TWI883261B - Method and non-transitory computer readable medium for automated classification of immunophenotypes represented in flow cytometry data - Google Patents

Abstract

Introduced here is an approach to improving the automatic identification of hematological diseases using computer-implemented models that are trained to rapidly distinguish between different collections of immunophenotypes that represent different disease types or disease states. Understanding the different patterns of immunophenotype collections contained in a given sample may permit a proposed diagnosis for a given hematological disease to be produced for the corresponding patient. For example, the proposed diagnoses may be output by a classification model based on the distribution of immunophenotypes across the given sample.

Description

Method and non-transient computer-readable medium for automatic classification of immunophenotypes of flow cytometric data

This invention claims priority to U.S. Patent Provisional Application No. 63/078,312 filed on September 14, 2020, entitled "Systems and Methods for Automatic Classification of Flow Cytometry Data", and U.S. Patent Provisional Application No. 63/078,662 filed on September 15, 2020, entitled "Methods for Automatic Preprocessing Flow Cytometry Data".

The specific embodiments provided in this disclosure relate to computer programs and related computer-implemented techniques for automatically classifying flow cytometer data. The present invention discloses a method for automatically classifying data, in particular, an automatic classification of immunophenotypes for flow cytometer data.

Leukemia (or leukaemia) is a blood disease that begins in normally developing blood cells that develop into different types. Generally, leukemia begins in the bone marrow and causes large numbers of abnormal blood cells to develop. These abnormal blood cells may be called "leukemic cells" or "blast cells." The exact cause of leukemia is unknown, so it is usually diagnosed based on the results of a blood test or a bone marrow test (also called a "bone marrow biopsy"). Generally, a blood test or bone marrow biopsy is done when a person (the "patient" or "subject") reports symptoms such as bleeding, bruising, fatigue, and fever.

There are four main types of leukemia: acute lymphocytic leukemia (ALL), acute myeloid leukemia (AML), chronic lymphocytic leukemia (CLL), and chronic myeloid leukemia (CML), as well as some rarer types. Leukemia is a broad group of diseases that affect the blood, bone marrow, and lymphatic system. This large group of diseases is often collectively referred to as "hematopoietic and lymphoid tissue tumors."

These categories have been primarily divided based on (i) whether the leukemia is acute (i.e., growing rapidly) or chronic (i.e., growing slowly), and (ii) whether the leukemia begins in myeloid or lymphocyte cells. Leukemias and acute myeloid leukemias generally begin in the bone marrow but often then spread to the blood and other parts of the body, including the lymph nodes, liver, and spleen. The rate at which blasts spread through the body corresponds to whether the underlying leukemia is acute or chronic. The presence and prevalence of blasts can also indicate other blood diseases, such as lymphoma and multiple myeloma.

Understanding the body's blood and lymphatic systems helps in understanding leukemia and lymphoma.

Bone marrow is a soft part inside some bones. At a higher level, bone marrow is made up of hematopoietic cells, fat cells, and support tissue. A small percentage of hematopoietic cells in the bone marrow are usually hematopoietic stem cells. Within the bone marrow, hematopoietic stem cells develop in order to develop into red blood cells, platelets, or white blood cells. Red blood cells (RBCs) carry oxygen from the lungs to other tissues in the body and carry carbon dioxide back to the lungs for excretion (such as by exhaling). Platelets are cell fragments made by a type of blood stem cell called a megakaryocyte. Platelets are important for plugging holes in blood vessels caused by cuts, scrapes, etc. White blood cells (WBCs) are responsible for helping the body fight infection.

There are three main types of white blood cells: lymphocytes, granulocytes, and monocytes. Lymphocytes are the main cells that make up the lymphatic tissues in lymph nodes and other parts of the body. Lymphocytes develop from "lymphoblasts" into mature, infection-fighting cells. There are two main types of lymphocytes: B lymphocytes (also called "B cells") and T lymphocytes (also called "T cells"). B cells protect the body by making proteins called antibodies that attach to germs, while T cells generally help to destroy these germs. ALL can develop from early lymphocytes. ALL can begin in the early maturation stages of B cells or T cells. Lymphoma also begins in lymphocytes, although it usually affects B cells or T cells in the lymph nodes rather than the blood and bone marrow. Granulocytes are white blood cells that contain granules. These granules often contain enzymes and other substances that help to destroy germs. There are three types of granulocytes: neutrophils, basophils, and eosinophils, which can be distinguished by the size and color of the granules. Monocytes also help protect the body from bacteria. Normally, monocytes circulate in the blood for a relatively short time (e.g., about a day) before entering tissues and becoming macrophages, which can destroy pathogens by surrounding and digesting them.

The term "bone marrow cells" is usually used to refer to those blood stem cells that can develop into red blood cells, platelets, or white blood cells other than lymphocytes. In AML, compared to ALL, it is these bone marrow cells that are abnormal.

The lymphatic system is an organ that is part of the circulatory system and the immune system. The lymphatic system is a large network of lymph, lymphatic vessels, lymph nodes, lymphatic organs, and lymphatic tissue. These vessels carry a clear fluid called lymph toward the heart. Unlike the cardiovascular system, the lymphatic system is not a closed system. This means that problems affecting the lymphatic system can quickly spread throughout the body if not treated promptly.

As mentioned above, the diagnosis of leukemia is usually made by a healthcare provider based on the results of a blood test or a bone marrow biopsy. By looking at a sample of a person's blood, a healthcare provider can determine if there are abnormalities in the red blood cells, platelets, or white blood cells that could be a sign of leukemia. A blood test can also detect blasts, although not all types of leukemia produce blasts that circulate in the blood. Sometimes blasts stay in the bone marrow. For this reason, a healthcare professional may recommend a bone marrow test, which involves taking a sample of the bone marrow to test for blasts.

Although recent medical advances have improved the survival rate of individuals diagnosed with leukemia, in some cases, unexpected results can still suddenly affect the prognosis. Current clinical practice uses the identification of minimal residual disease (MRD) as a prognostic indicator, which is detected using flow cytometry (FC). Generally, FC is a technology used to detect and measure the characteristics of cell populations.

In an FC experiment (or just "the experiment"), a sample containing cells is initially suspended in a liquid. Typically, these cells are labeled with a fluorescent marker that only binds to certain types of cells, allowing for the definition of different types of cells. The sample is then injected into a flow cytometer, where it is focused (ideally one cell at a time) and passed through a laser beam. The light scattered by the cells is characteristic of the cells, resulting in an illumination pattern that reflects the type of cells contained in the sample. Because the cells are labeled with a fluorescent marker, the light is absorbed and then emitted in a specific wavelength band.

Accordingly, the experiment may involve measuring the level of fluorescence excitation on the antibody marker to generate FC data. Historically, medical professionals have manually reviewed FC data by visually analyzing two-dimensional images to determine whether a diagnosis is appropriate. This approach is not only laborious and time-consuming, as the number of cells often ranges from tens of thousands to millions, but also prone to error, as these medical professionals must make subjective decisions. Some institutions have proposed the use of machine learning (ML) algorithms or artificial intelligence (AI) algorithms to manage FC data; however, processing large amounts of FC data remains a challenge.

The present disclosure relates to a method for improving the automatic identification of blood diseases using computer executable models (or simply "models") that are trained to quickly distinguish different sets of immunophenotypes representing different disease types or disease states. Understanding the pattern of different sets of immunophenotypes contained in an input sample can generate a recommended diagnosis of one of the input blood diseases for the corresponding patient. The recommended diagnosis may be one of multiple outputs generated by a disease analysis platform (or simply "analysis platform") based on the immunophenotype distribution of the entire sample. For example, the analysis platform can generate a recommended diagnosis for more than one acute leukemia (e.g., ALL and AML), all hematopoiesis (i.e., myeloma and one or more non-neoplastic diseases), or another blood disease.

This method can be used as part of a training framework for training a model to automatically classify samples (represented by FC data). The training framework can include three steps: a first step of processing the FC data; a second step of converting the processed FC data into a format more suitable for training the model; and a third step of using the formatted and processed FC data for training the model. Generally, the training framework is executed tens, hundreds, or thousands of times because a variety of samples (e.g., corresponding to different blood diseases) can be used for training. This model (i.e., processing, conversion, and then training) can reliably and quickly gain insights into the distribution of cell types in the entire sample, which can then be further submitted to, for example, medical professionals for detection and judgment.

This approach may also be used as part of a classification framework for applying a trained model to FC data to produce one or more outputs. Each output may represent a proposed diagnosis for a hematological disorder. More generally, a classification framework may be similar to a training framework in that processing and transformation steps may also be performed. Thus, upon receiving an input that explicitly requests a proposed diagnosis for a sample based on an analysis of FC data, the FC data may first be processed and then transformed into a format that can be more easily processed by the trained model. The formatted and processed FC data may then be provided as input to the trained model to produce an output. Compared to manually analyzing FC data to classify cell types, this automated approach can improve the quality, consistency, and timeliness of healthcare services by quickly generating insights that can be used to diagnose and monitor patients.

Although specific embodiments may be described with reference to specific hematological diseases, these hematological diseases are selected for illustrative purposes only. For example, the methods disclosed herein can be described in the context of the following model, which, when applied to FC data corresponding to a sample, can generate and output a suggested diagnosis indicative of ALL, AML, and total hematopoiesis. However, the methods are also equally applicable to other hematological diseases, such as: CLL, CML, Hodgkin's lymphoma and non-Hodgkin's lymphoma (diffuse large B-cell lymphoma, follicular lymphoma, mantle cell lymphoma, T-cell lymphoma), multiple myeloma, acute erythroleukemia (AEL), acute promyelocytic leukemia (APL) and other solid tumors. Furthermore, the approach is equally applicable to both malignant and non-malignant blood diseases (e.g., pancytopenias). Thus, the model can stratify patients into various blood diseases (malignant and/or non-malignant) based on sample-level characterization of the cells found in the sample.

To further illustrate, the following specific embodiments can also be executed in the context of executable instructions. However, those with ordinary knowledge in the field of the present invention can understand that the present disclosure is implemented by hardware, firmware or software. For example, a disease analysis platform (or simply "analysis platform") can be an example of a computer program that provides support for testing information related to the process and/or state of hematological malignancies, arranging treatment methods, and reviewing model-recommended diagnoses.

200:Framework

202: Data acquisition phase

204: Data acquisition phase

206: Data conversion phase

208: Training

210: Classification

300:Database

302: FCS file

304, 800: FC data matrix

350, 351, 352, 601, 602, 603, 701, 702, 703, 850, 851, 852, 901, 902, 903, 904, 905, 1001, 1002, 1003, 1004, 1005: Steps

600, 700, 900, 1000: Program

802: Hybrid Model

804: Vector

1100: Network environment

1102:Analysis Platform

1104: Interface

1106a, 1106b, 1210, 1314: Network

1108, 1204: Server system

1200: System

1202: Flow cytometer

1206: Data storage device

1208:Computer device

1300: Processing system

1302: Processor

1304, 1308, 1328: Instructions

1306: Main memory

1310: Non-volatile memory

1312: Network interface card

1316: Bus

1318: Video display unit

1320: Input/output device

1322: Control device

1324:Drive device

1326: Storage media

1330:Signal generating device

Figure 1 is a diagram showing how hematological diseases have been classified historically.

FIG2A is a schematic diagram illustrating a framework that can be executed by an analysis platform to acquire, process, and transform flow cytometer (FC) data to facilitate automated detection of hematological diseases and indication of hematological abnormalities.

Figure 2B illustrates how the framework shown in Figure 2A can be used to (i) obtain "raw" FC data related to patients, (ii) select intersecting or interrelated parameters, (iii) transform the "raw" FC data by encoding the patient level, and then (iv) apply a classification model to the transformed FC data to classify patients, or (v) train a classification model to do the same.

Figure 3 is a high-level diagram to reveal the process of obtaining FC data from the source.

Figure 4 is used to illustrate how spillover signals from other fluorescent intensities can bias the pure signal of the primary fluorescent intensity of current interest.

Figure 5 is used to illustrate how to generate a scatter plot using forward scattered light height (FSC-H) along the Y axis and forward scattered light area (FSC-A) along the X axis to facilitate manual single circle selection.

Figure 6 is a schematic flow chart for explaining the automatic execution of a single lap selection procedure.

FIG7 is a schematic flow chart for explaining the standardization of FC data sets extracted from flow cytometer standard files (FCS files).

Figure 8 is a high-level diagram to illustrate the process of converting the processed FC data from its matrix form to a vector.

Figure 9 is a schematic flow chart illustrating the process for training a model to classify blood diseases.

Figure 10 is a schematic flow chart for explaining the process of classifying samples by applying a classification model.

Figure 11 is a schematic diagram showing the network environment of the analysis platform.

FIG. 12 is a schematic diagram illustrating an embodiment of a system capable of automatically classifying different patterns of immunophenotype sets for identifying blood diseases.

FIG. 13 is a block diagram illustrating an example of a processing system that can perform the operations described in this disclosure.

By combining the drawings and detailed descriptions, people with ordinary knowledge in the field of the present invention will more clearly understand the various features of the technology disclosed in the present invention. Some of the embodiments disclosed in the drawings are mainly used for illustration. However, people with ordinary knowledge in the field of the present invention can understand that alternative implementation schemes can be adopted without deviating from the principles of the present technology. Therefore, the specific examples of the present invention disclosed in the drawings do not limit the present invention, but can be modified in various ways.

Glossary

The reference to "one specific embodiment" or "certain specific embodiments" in this disclosure means that the features, functions, structures or characteristics described are included in at least one specific embodiment. Such terms do not necessarily refer to the same specific embodiment, nor do they necessarily refer to mutually exclusive alternative embodiments.

Unless otherwise expressly noted in the context, the terms "include", "including" and "by" shall be interpreted in an inclusive sense, rather than in an exclusive or exhaustive sense (in other words, "including but not limited to"). The term "based on" shall also be interpreted in an inclusive sense, rather than in an exclusive or exhaustive sense. Therefore, unless otherwise expressly noted, the term "based on" means "based at least in part on".

The terms "connected," "coupled," and their synonyms refer to any connection or coupling between two or more elements, whether directly or indirectly. The connection or coupling can be physical, logical, or a combination of both. For example, elements can be electrically or communicatively coupled to each other despite not having a common physical connection.

The term "module" may refer to software, firmware, hardware, or any combination of these. A module is generally a functional component that produces one or more outputs based on one or more inputs. A computer program may include or utilize one or more modules. For example, a computer program may utilize multiple modules that are responsible for completing different tasks, or a computer program may utilize a single module that is responsible for completing all tasks.

When used in reference to a list of multiple items, the term "or" is intended to cover all of the following interpretations: any item in the list, all items in the list, and any combination of items in the list.

Introduction to immunophenotype

Immunophenotyping by FC is a laboratory technique that is typically used to detect the presence or absence of WBC markers (abbreviated as antigens). These antigens are protein structures found on or in white blood cells, and specific groups of these antigens are unique to specific cell types. Because FC immunophenotyping can serve as a sensitive filter or test for blood disorders, it is a useful tool for staging previously diagnosed blood disorders, proving the presence or absence of blood disorders, monitoring response to treatment (e.g., by analyzing MRD), documenting the recurrence or progression of blood disorders, and detecting inter-complications of blood disorders. In simple terms, FC immunophenotyping can be used to detect normal and abnormal cells whose marker patterns are generally observed in specific blood disorders.

In the conventional art, FC data generated by a flow cytometer is either plotted in a single dimension to produce a bar graph, or in multiple dimensions to produce a "dot plot" or "scatter plot." Regions on these plots are sequentially separated by fluorescence intensity by creating a series of subset captures (also called "gates"). Different gate protocols are utilized for different diagnostic purposes, particularly those related to hematology. In the conventional art, single cells can be distinguished from doublets and larger aggregates of cells by visual analysis of these plots. The term "doublet" as used herein may refer to an event in which a flow cytometer measures more than one cell. Doublets are usually identified based on the "time-of-flight" or "pulse-width" of the laser beam. Correct identification of doublets is crucial in cell sorting, as the corresponding values in the FC data should not affect the analysis. However, since the exclusion of doublets relies heavily on visual analysis, this process is prone to errors, which will be discussed further below.

Using FC data, the user can determine the relative size of cells using known controls. For example, forward scatter (FSC) and side scatter (SSC) values are often used in gated acquisitions. More specifically, FSC and SSC values can be used to identify cells of interest based on cell size and cell granularity. In general, FSC and SSC values are used to normalize data relative to other light scatter parameters, particularly fluorescent markers used to identify different cell types through traditional visual analysis of FC data.

However, FC immunophenotyping has several disadvantages. First, the data generated by flow cytometry during the experiment can be difficult to interpret. This can further lead to errors because the medical professional responsible for analyzing the data may need to make subjective decisions. Second, determining an appropriate classification of blood disorders can be difficult. Figure 1 reveals a schematic diagram that illustrates how blood disorders have been classified historically. However, correctly viewing this diagram relies on an accurate understanding of the distribution of cell types in the sample. Due to the limitations of visual analysis of the FC data populated on the graph, cells can easily be misclassified.

The methods proposed in the present disclosure not only involve classifying single cells in an automated manner to reduce errors, but may also involve generating representations of cell types across samples to determine how to stratify corresponding patients in different blood diseases. In other words, sample-level representations of cell types (also referred to as "patient-level representations") can be used to determine which blood disease (if the patient has a disease) is predicted for an input sample (i.e., the input patient). Sample-level representations help classify patients in different blood diseases, as well as assign pathological states (e.g., relapse, progression, etc.), and correlate cell type distribution with clinical intervention to determine treatment efficacy. Examples of clinical interventions include: chemotherapy, targeted therapy, immune checkpoint inhibitors, and chimeric antigen receptor (CAR) T cell therapy.

Overview of a computational pipeline for automated analysis of FC data

The present disclosure relates to a method for improving the automatic identification of blood diseases using a method trained to quickly (i) distinguish different cell types in a sample and then (ii) determine an appropriate prediction based on the distribution of a set of immunophenotypes across the sample. As further discussed below, the method can be implemented through a framework that supports multiple computational pipelines, namely a first computational pipeline: which is used to train a model to classify cells by the distribution of a set of immunophenotypes and then classify samples based on the cell type distribution; and a second computational pipeline: which is used to apply the trained model to FC data to generate an output indicating a suggested diagnosis of a blood disease.

Figure 2A is a top-level diagram schematically illustrating a framework 200 that can be executed by an analysis platform to acquire, process and transform FC data to facilitate the automatic detection of hematological abnormalities indicative of hematological diseases. As further discussed below, FC data can be provided as input to a classification model for training, or FC data can be provided as input to a classification model for classification. The classification model (also referred to as a "classifier model" or simply "classifier") can perform multi-class classification. Therefore, when applied to FC data, the classification model can generate multiple outputs representing recommended diagnoses for different blood diseases. For example, the analysis platform can utilize a classification model of multidimensional multicolor flow cytometer (MFC) phenotypes, which is trained using, for example, a deep neural network (DNN) or support vector machines (SVMs) in combination with a Gaussian mixture model (GMM). In some embodiments, the classification model is developed through supervised learning by analyzing MFC datasets to develop an explanation or understanding of MFC in order to objectively detect MRD. Supervised learning refers to a branch of artificial intelligence in which datasets and accompanying markers are used to train models to provide reliable predictions.

As shown in FIG2A, the framework 200 may include various phases. The phases may include: a data acquisition phase 202, a data capture phase 204, and a data transformation phase 206. Please refer to FIG3 for further discussion of the data acquisition phase 202; please refer to FIG4 to FIG7 for further discussion of the data capture phase 204; please refer to FIG8 for further discussion of the data transformation phase 206. After completing the data transformation phase 206, the analysis platform may provide the output to the classification model for the purpose of training 208 or classification 210. Please refer to FIG9 and FIG10 for further discussion of training and classification.

In the data acquisition phase 202, the analysis platform can obtain FC data, which describes the characteristics of a sample from a source, wherein the sample includes cells labeled with fluorescent markers. The FC data can be contained in a file whose format is a flow cytometer standard file (FCS file). The FCS file is a file format standard used to read and write data from FC experiments. The file format describes a file composed of text data followed by binary data, and the order of the file format is generally: (1) header segment, (2) text segment, (3) data segment, (4) optional analysis segment, (5) loop redundancy check (CRC) value, and (6) optional other segments. FC data can be represented as a measurement matrix consisting of N parameters at M wavelengths, where M and N are integers, which can be extracted from the data segment of the FCS file. These parameters can include light scattering parameters and/or fluorescence signature parameters.

In some specific embodiments, the source of FC data obtained by the analysis platform is the flow cytometer that generates the FC data. In other specific embodiments, the source is a storage medium that can be accessed by the analysis platform (e.g., via the Internet). The storage medium can be associated with an entity that manages the flow cytometer or other entities. In some specific embodiments, the storage medium is publicly accessible (e.g., via the Internet). In these specific embodiments, in order to obtain FC data from the storage medium, the analysis platform can initiate a connection with the storage medium through a data interface (e.g., an interface designed by an application). In other specific embodiments, the storage medium is privately maintained and managed. For example, the storage medium may contain proprietary clinical data generated over a long period of time by a medical system, and the analytics platform may be granted access to the storage medium based on an agreement between the medical system and the entity managing the analytics platform.

In the data capture phase 204, the analysis platform may process the FC data in preparation for further processing. The nature of the data capture phase 204 may depend on the form of the FC data obtained by the analysis platform. For example, assume that the analysis platform captures the FC data matrix from the FCS file as described above. In this specific embodiment, the analysis platform may process the values contained in the FC data matrix by performing compensation operations, selection operations and/or normalization operations, which will be discussed further below. More generally, the data capture phase 204 can ensure that the analysis platform is able to analyze large batches of FC data in a consistent and accurate manner within a relatively short time interval. Since the processing occurs before the analysis platform examines the FC data in order to gain insights from it, the data acquisition phase 204 may also be preferably referred to as the "data pre-processing phase" or simply the "data processing phase".

In the data transformation stage 206, the analysis platform may transform the FC data into a form more suitable for further processing. For example, the analysis platform may implement an ML algorithm to transform the FC data matrix into a function of a multi-dimensional vector. Thus, the analysis platform may transform the FC data matrix into a FC data vector.

This FC data vector can be used in different ways, depending on the computational pipeline currently being implemented or executed by the analysis platform.

For example, suppose the analysis platform is interested in training a classification model to identify hematological abnormalities in blood diseases. In such a case, the analysis platform may provide (i) FC data vectors and (ii) a set of markers (used to represent an immunophenotypic set pattern for each cell described in the flow cytometer data vector) to a classification model for training purposes 208. For example, the markers may be representations of the disease type, disease state, or physiological state (also known as "pathological state") describing each cell in the vector. Generally, the FC data vector is one of a plurality of FC data vectors provided to the classification model 208 for training purposes, and the plurality of FC data vectors may correspond to different blood diseases (i.e., the classification model is trained to recognize blood diseases). Therefore, the classification model may learn how to classify samples in a plurality of hematological diseases based on the FC data vectors as input, and these immunophenotypes may represent each of the plurality of hematological diseases.

On the other hand, when the analysis platform is interested in applying a classification model to the FC data vector for classification purposes 210. In such a case, the analysis platform can provide the FC data vector as an input to the classification model in order to obtain an output indicating a recommended diagnosis of a blood disease. As will be discussed further below, in some specific embodiments, the classification model can classify the FC data vector generated for a specified sample (and therefore also for a specified patient) in more than one blood disease. In this specific embodiment, the classification model can generate multiple outputs, each of which can represent a recommended diagnosis for a different blood disease.

Figure 2B illustrates how the framework disclosed in Figure 2A can be used to (i) obtain “raw” FC data related to patients, (ii) select parameters that are mutually or mutually correlated (e.g., fluorescence signature parameters), (iii) transform the “raw” FC data by patient-level encoding (e.g., using GMMs and Fisher Vectorization), and then (iv) classify patients by applying a classification model (e.g., multivariate SVM) to the transformed FC data, or (v) train a classification model (e.g., multivariate SVM) to do the same. At a high level, Figure 2B represents an overview of a framework that provides the general steps of the above computational pipeline. However, the nature of the training may depend on the target task. For example, step (ii) can be implemented by any of resampling, padding or selecting fluorescence signature parameters (e.g., based on human knowledge or model-generated output) to obtain feature dimensions. If the target task involves patients with different fluorescence signature representations, then step (ii) can be implemented to match the feature dimensions of the respective FC data from the different representations. In addition, step (ii) may involve a method of forming the original FC data into a matrix that includes both training data and test data. Therefore, all fluorescence signature parameters can be considered simultaneously when encoding and classifying.

A. Data Acquisition

FIG3 is a high-level diagram illustrating the process of obtaining FC data from a source. Here, the source is a database 300 in which one or more flow cytometers are capable of storing FC data generated by experiments. This process can be performed by an analysis platform as part of a data acquisition step (e.g., data acquisition stage 202 of FIG2A ). More generally, this is the process by which an analysis platform can obtain FC data that can be used to train a classification model to classify a sample based on an analysis of cells that have been characterized by a flow cytometer. Likewise, this is also the process by which the analysis platform can obtain FC data, and the classification model can be applied to generate one or more outputs (e.g., recommended diagnosis).

Generally speaking, the database 300 has entries containing FC data of different samples (i.e., different patients) tested by the experiment. In FIG. 3 , the entries include an FCS file 302 in which the FC data associated with the corresponding sample (i.e., the corresponding patient) is stored. However, it is worth noting that the FC data can be stored in other formats in the database 300.

Database 300 may be one of a plurality of databases from which the analysis platform can obtain FC data. For example, assume that the analysis platform is interested in obtaining FC data generated by flow cytometers located in different medical institutions associated with different medical systems. In this case, the analysis platform may be allowed to access (i) a first database in which FC data generated by the first flow cytometer is stored; and (ii) a second database in which FC data generated by the second flow cytometer is stored. Thus, the analysis platform may obtain FC data from more than one source sequentially or simultaneously. In general, FC data stored in different databases will be associated with different sets of patients, although there may be some overlap (e.g., a patient may have a sample examined by a first flow cytometer associated with a first medical system and another sample examined by a second flow cytometer associated with a second medical system).

Typically, a flow cytometer will process a large number of fluorescent markers simultaneously, in addition to several forward and side scatter properties. For example, the FC data generated during an experiment may include 17 to 23 channels, with 6 channels corresponding to forward and side scatter properties and the remaining channels corresponding to different fluorescent marker properties. Forward and side scatter properties (also called "optical properties" or "optical parameters") may include forward scatter area (FSC-A), forward scatter width (FSC-W), forward scatter height (FSC-H), side scatter area (SSC-A), side scatter width (SSC-W), and side scatter width (SSC-H). Meanwhile, the fluorescence marker attributes (also called "fluorescence marker parameters") can include: CD117_PerCP-Cy5-A, KAPPA_FITC-A, HLA-DR_V450-A, CD38_APC-H7-A, and CD123_PE-A, etc. Therefore, one experiment can generate a large data set.

When a flow cytometer analyzes a sample, FC data is generated as output. The FC data may be in the form of a matrix with more than one dimension. For example, FC data may include: FSC signals (e.g., FSC-A, FSC-W, or FSC-H signals), SSC signals (e.g., SSC-A, SSC-W, or SSC-H signals), or fluorescence signals, and each of these signals may be considered as a separate dimension. The characteristics of these signals may also be treated as dimensions. Examples of such characteristics include: amplitude, frequency, amplitude variation, frequency variation, time dependency, spatial dependency, and the like. In addition, the fluorescence signal may include: a red fluorescence signal, a green fluorescence signal, or one or more other color fluorescence signals. In general, the matrix will have at least three dimensions (it can have seven or more). For normalization purposes, FC data can be presented as a two-dimensional matrix with individual signal values for training, validation, or testing as columns and features as rows. This FC data matrix can be exported from a flow cytometer in an FCS file.

As described above, the analysis platform can apply a classification model to a FC data extracted from an FCS file in order to classify individual cells and then classify the sample as a whole (e.g., as a representative of a blood disease). However, in its original form, FC data can be difficult for the classification model to handle. For example, it may require a large amount of computational resources for the classification model to quickly process FC data in matrix form. Therefore, the classification model can instead be trained to operate on FC data that has been transformed or converted into another form that can be more easily processed by the classification model.

Referring to FIG8 , although the process of converting the FC data matrix is described below, the analysis platform can extract the FC data matrix when obtaining the FCS file disclosed in FIG3 . Therefore, the analysis platform can initiate a connection with a database to which one or more flow cytometers can upload FCS files (step 350), obtain a series of FCS files 302 from the database 300 (step 351), and then extract the FC data matrix from each FCS file, thereby obtaining a series of FC data matrices 304 (step 352). In a specific embodiment in which the analysis platform is interested in classification samples rather than training classification models, the analysis platform can only obtain one FCS file from the database.

B. Data Acquisition

As mentioned above, flow cytometry measures cell types based on the fluorescence response expressed by antibodies. Depending on the application, the number of cells whose fluorescence is measured in an experiment can range from thousands to millions. For this reason, the FC data sets (e.g., in matrix form) generated by flow cytometry can be very large.

Traditionally, these large FC datasets are processed using dimensionality reduction techniques that result in scatter plots of single values (e.g., FSC, SSC, and fluorescence). Although some information about clustering may be revealed on these scatter plots, medical professionals are usually responsible for defining "circles" to identify distinct regions on these scatter plots.

Instead of proposing clusters of cells at the cellular level, the analysis platform can encode large amounts of cell-level data into patient-level representations for automatic classification. To achieve this, the analysis platform can adopt a method for encoding FC data that relies on ML-based techniques (e.g., GMMs and Fisher vectors) to cluster FC data for different levels of recognition tasks. Training of the GMM model involves concatenating all cell-level data for all patients represented in the FC data used for training. Therefore, this method consumes a lot of computational resources. Therefore, in order to save computational resources, downsampling and/or aggregation can be used. Downsampling can be achieved by selecting subsets of the data (e.g., by uniformly sampling the data), while aggregation can be achieved by statistically representing clusters of cells that are grouped together under the assumption that the processed data are still likely to form a distribution similar to the original data. For example, the analysis platform can represent clusters of cells (e.g., 3, 5, or 10 cells) with a mean vector to reduce memory consumption. Importantly, the FC data provided as input to the classification model is of high quality. Therefore, the analysis platform can extract or process the FC data before further processing it (e.g., converting from matrix form to vector form).

As part of the data extraction phase, the analysis platform can perform (i) compensation operations, (ii) selection operations, and (iii) normalization operations. Each of these operations is discussed further below.

Compensation is the process by which the analysis platform attempts to obtain a pure signal for each fluorescence intensity by eliminating the spillover signal from other fluorescence intensities contained in the FC dataset. Compensation is therefore performed to ensure that the laser performance of the flow cytometer is within an appropriate range. Figure 4 illustrates how the spillover signal from other fluorescence intensities deviates from the pure signal of the primary fluorescence intensity of interest. This can (and often does) lead to inappropriate results when the user manually circles the fluorescence intensities on the scatter plot. To address this issue, the user uses compensation beads during the flow cytometer run to establish a compensation setting before any experiment is performed. In other words, the compensation beads can be passed through the flow cytometer without any sample in an attempt to create a spillover signal. In doing so, the flow cytometer will generate a spillover matrix that can later be used to calculate, infer, or otherwise determine the pure signal of each fluorescence intensity. The spillover matrix is typically saved by the flow cytometer in a text segment of each FCS file generated over time.

When the analysis platform obtains an FCS file, the analysis platform can not only extract the FC dataset from the data segment, but also extract the overflow matrix from the text segment. Then, the analysis platform can use the overflow matrix to perform compensation operations. In other words, the analysis platform can use the overflow matrix to generate a compensated FC dataset from the original FC dataset extracted from the FCS file. Generally speaking, the overflow matrix is an n x n matrix, where "n" is the number of fluorescent markers associated with the corresponding sample. Considering that each row is the original measurement value of the corresponding fluorescent marker, each digit in the same row can represent the contribution of a fluorescent marker to the measurement. This contribution is called the "overflow coefficient" and has a maximum value of 1. Therefore, the diagonal elements of the overflow matrix are all 1, and the remaining numbers are between 0 and 1. The overflow matrix can be used to calculate the compensated measurement for each fluorescence flag by multiplying the inverse of the overflow matrix with the uncompensated data matrix for each fluorescence flag.

It is worth noting that compensation operations may not be performed in every embodiment. For example, compensation may only be required when the analysis platform determines that the quality of the FC dataset is not sufficient for training or classification. The analysis platform can determine the quality based on analysis of the original FC dataset. For example, the analysis platform can attempt to determine whether the density, distribution, or absolute values of the measurements contained in the FC dataset meet the criteria that jointly define quality. For example, the analysis platform can determine through computational analysis that the FC data set labeled "uncompensated" in the scatter plot similar to Figure 4 is of sufficient quality, and the analysis platform can determine through computational analysis that the FC data set labeled "compensated" in the scatter plot similar to Figure 5 is of sufficient quality. It may be necessary to require better quality so that the analysis platform can perform automatic analysis with better accuracy.

Single binning (or simply "binning") refers to the process of removing inaccurate signals from non-specific binding events or doublets from an FC dataset before its contents are actually binned. For example, suppose two cells are measured simultaneously by a flow cytometer because the cells are aligned as they pass through the laser beam. In order to ensure that the corresponding measurements produced by the flow cytometer do not affect the performance of the classification model, it may be desirable to remove the corresponding measurements from the FC dataset.

This process (also known as "doublet exclusion") has traditionally involved plotting the area of height or width versus FSC or SSC. For example, Figure 5 illustrates how a scatter plot is generated with FSC-H along the Y-axis and FSC-A along the X-axis to facilitate manual haploid binning. To perform binning, users have traditionally determined the area of single cells by defining a region on the scatter plot. This method relies on the linear relationship between FSC-H and FSC-A, so the region is usually plotted along a straight line, roughly corresponding to the diagonal line shown in Figure 5.

To eliminate the ambiguity inherent in manual ring selection, the analysis platform can perform a function that performs ring selection or doublet identification in an automated manner. This function can be explained to ensure that each value in the FC data set corresponds to a single cell. Figure 6 is a schematic flow chart to reveal the automatic execution of a single ring selection process 600. At the beginning, the analysis platform can delete cells whose FSC-A values reach a threshold (step 601). For example, the analysis platform can delete all cells whose FSC-A values are the maximum value. The maximum value can be 218, which is the highest possible value of FC data in a linear scale. In another specific embodiment, the analysis platform can delete all cells whose FSC-A values are in the top 2%, 3% or 5% of the entire FC data set. Therefore, the threshold can be programmed into the command executable by the analysis platform, or the threshold can be determined dynamically by the analysis platform based on the FC dataset. When performing singleton gating, FSC-H and FSC-A are displayed in a linear scale, so this step can be performed by the analysis platform to simulate the operation of medical professionals. More specifically, this step can be performed automatically to remove cells that are unnaturally "stuck" to the right side of the scatter plot, as revealed in Figure 5, because these cells would not be included in this area if it was manually defined by a medical professional.

Then, the analysis platform can circle the most densely distributed cells on the scatter plot containing the remaining cells (step 602). More specifically, the analysis platform can generate a scatter plot based on the FSC-H and FSC-A values contained in the compensated FC data set of the remaining cells, and then the analysis platform can circle the most densely distributed cells on the scatter plot. For example, the analysis platform can circle the 90%, 95%, or 98% most densely distributed cells on the scatter plot. This percentage can be called the "circle score". Due to the highly linear relationship between FSC-H and FSC-A, these circles should mainly capture single cells rather than doublets.

Therefore, the analysis platform can calculate the coefficient of determination (R ² ) between the gated cells that remain after step 602 (step 603). If the R ² value exceeds an upper limit (e.g., 0.80, 0.85, or 0.90), the function executed by the analysis platform can return the data associated with these cells in the FC data set and then terminate. Otherwise, the function can instruct the analysis platform to repeatedly execute steps 602 to 603, each time reducing the gated score by a predetermined amount (e.g., 2%, 3%, 5%, or 10%) until the R ² value exceeds the upper limit. If the ^R2 value still does not exceed the upper limit when the gated score reaches a lower threshold (e.g., 70%, 75%, or 80%), the analysis platform can generate an alert indicating that the sample lacks linearity between FSC-H and FSC-A. Because it may cause further problems in using the FC dataset (e.g., in training or classification), the analysis platform can simply return to the original FC dataset or the compensated FC dataset.

AI-assisted analysis represents an attractive option to classify samples represented by FC datasets in a more systematic and consistent manner than individuals manually reviewing FC datasets. However, in order for AI-assisted analysis to have a widespread impact on clinical and diagnostic practice, it may be necessary for institutions to adhere to protocols.

Normalization is a process by which the analysis platform can overcome the problem of non-standardized processing of FC datasets. Normalization can be used as a means to improve the performance and training stability of the classification model to which the FC dataset is provided as input for training or classification purposes. FIG. 7 is a schematic flow chart of a process 700 for normalizing an FC dataset extracted from a flow cytometer standard file (FCS file). As described above, the analysis platform will typically perform a normalization operation after performing compensation and ringing operations to ensure that inappropriate and inaccurate values are removed before these values are normalized.

As previously mentioned, FC data sets will typically contain values for multiple parameters. For example, in addition to values for one or more fluorescence marker parameters, FC data sets may also contain values for one or more light scattering parameters. As part of the normalization operation, the analysis platform may initially aggregate the values belonging to each parameter into a unique feature dimension (step 701). The analysis platform may then resample the unique feature dimension to the same sample size to ensure that each parameter has the same number of cells (step 702). It is worth noting that in some specific embodiments, the analysis platform may resample the unique feature dimension so that the parameters have approximately the same number of cells (e.g., within 2%, 5%, or 10%) rather than exactly the same number of cells.

In one embodiment, the values of a fluorescent marker parameter (e.g., CD56-APC) can be aggregated across multiple samples as a single parameter to ensure that the number of values (i.e., the number of cells) meets the counting criteria determined by resampling. In another embodiment, the values of a light scatter parameter (e.g., FSC-A or SSC-A) can be aggregated across multiple samples and then downsampled to ensure that the number of values (and thus the number of cells) meets the counting criteria determined by resampling. More generally, the counting criteria can represent the appropriate number of samples determined by the analysis platform.

其中μ是平均值，σ是標準差。z-score標準化技術可以確保分佈的平均值為0並且標準差為1，因此其為有用的，當有一些離群值但沒有多到需要採取更嚴厲的措施(例如：削減)時。分析平臺還可使用其他標準化技術，舉例來說，分析平臺可以執行對範圍的縮放、削減或對數縮放，以取代或補充Z-score。 As will be further described below, by standardization, the analysis platform can generate a processed FC dataset that can be used as an input by other elements of the framework. For example, the analysis platform can perform standardization based on the z-score standardization technique to ensure that the values in the FC dataset are on a similar scale (step 703), thereby generating a processed FC dataset. The z-score standardization technique is a variation of scaling, which represents the number of standard deviations away from the mean. The formula for calculating the value (x) in the z-score is as follows:

Where μ is the mean and σ is the standard deviation. The z-score normalization technique ensures that the mean of the distribution is 0 and the standard deviation is 1, so it is useful when there are some outliers but not so many that more drastic measures (such as clipping) are required. The analytics platform can also use other normalization techniques. For example, the analytics platform can perform range scaling, clipping, or logarithmic scaling instead of or in addition to the z-score.

Therefore, to process the FC dataset extracted from the FCS file, the analysis platform can perform (i) compensation operations, (ii) selection operations, and (iii) standardization operations. Thereafter, the FC dataset can be stored in a storage medium accessible to the analysis platform, or the FC dataset can be further processed by the analysis platform according to an appropriate computational pipeline. Other steps can also be performed. For example, the analysis platform can generate visual indicators of the values remaining in the FC dataset after processing (e.g., FSC-H and FSC-A values) as a method to allow the user to review how the analysis platform automatically compensates, selects, and standardizes the FC dataset. In one embodiment, the analysis platform can generate a report that includes an analysis of the values remaining in the FC dataset after processing. Alternatively, the analysis platform can generate a scatter plot containing the values remaining in the FC dataset after processing. Regardless of its form and content, the visual indicators can be published to an interface generated by the analysis platform for user review.

Processing FC data sets in a predetermined manner ensures that the analysis platform can analyze large amounts of data in a relatively short time with improved quality, and the effects of signal offset are largely (if not completely) mitigated.

C. Data conversion

Although useful analysis can be obtained by analyzing raw or "processed" FC data, classification models may have difficulty processing such data, especially if the classification model is tasked with processing data from dozens or hundreds of samples in a short period of time. Therefore, analysis platforms can transform processed FC data into a form that is more suitable for further use. In particular, analysis platforms can transform processed FC data into a form that is well suited for input into classification models.

FIG8 is a high-level diagram illustrating the process of converting processed FC data from its matrix form to a vector. The process can be performed by the analysis platform as part of a data conversion step (e.g., data conversion step 206 of FIG2A ). More generally, the analysis platform can perform the process to convert the processed FC data into a form that can be more easily processed by a classification model. In one embodiment, the processed FC data can be converted into a vector 804 using Fisher vector encoding and GMM distribution. After the conversion occurs, the representation of each sample can be a high-dimensional vector that corresponds to the sample phenotype of the patient. The representation can be easily used by different types of classification models (including SVMs, DNNs, and random forests).

At the beginning, the analysis platform can obtain the processed FC data matrix 800 (step 850). As described in Figures 4 to 7 above, the processed FC data matrix 800 is usually generated by the analysis platform by processing the "raw" FC data matrix. Therefore, the processed FC data matrix 800 can be provided to the analysis platform at any time, and the process can simply become the next stage in the framework (e.g., framework 200 in Figure 2A) executed by the analysis platform.

Alternatively, the analysis platform may obtain the processed FC data matrix 800 from elsewhere. For example, the analysis platform may continuously or periodically obtain FCS files generated by a flow cytometer. As described above, the analysis platform may process the "raw" FC data matrix contained in each FCS file. However, the analysis platform may store the processed FC data matrix in a storage medium for future use, rather than immediately converting the processed FC data matrix. Therefore, the analysis platform may not immediately perform the process disclosed in FIG. 8 after processing the "raw" FC data matrix. Instead, the analytics platform can store the processed FC data matrix so that it can perform a "batch training" scheme where training occurs periodically (and the processed FC data matrix only needs to be transformed periodically).

The analysis platform can then create a mixture model 802 based on the processed FC data matrix 800 (step 851). More generally, a mixture model is a probabilistic model that aims to represent the presence of cell types within the processed FC data matrix by clustering comparable values. Thus, the mixture model 802 can correspond to a mixture distribution representing the probability distribution of cell type observations in the entire sample represented by the processed FC data matrix 800. An example of such a mixture model is a GMM.

Therefore, the gradient of the mixture model 802 can be calculated using an ML algorithm to derive a vector representation of the processed FC data matrix 800 (step 852). This gradient-based feature space transformation can rely on a distance function to estimate the relationship between cells in the processed FC data matrix 800 and the clusters defined by the GMM. The Fisher kernel distance used in the Fisher vector is an example of a distance function that measures high-order relationships based on a probabilistic cluster distribution. Therefore, the derived vector representation can represent the complex cell distribution of the processed FC data matrix using the relationship with each cluster. For example, the analysis platform can use the mixture model to calculate the Fisher vector to construct the vector 804. While the mixture model 802 may attempt to cluster comparable values, Fisher vectorization (when performed by the analysis platform) may further encode the processed FC data matrix based on the trained parameters of the mixture model 802.

The dimension of vector 804 may be based on the dimension of processed FC data matrix 800 and the cluster number (also called "mixture number"). Accordingly, if processed FC data matrix 800 includes various dimensions as described above, then vector 804 may be a high-dimensional vector. Each cell represented in processed FC data matrix 800 may be associated with multiple entries in the high-dimensional vector, and each of these entries may correspond to different parameters (e.g., FSC, SSC, fluorescence intensity, and features such as amplitude, frequency, etc.) to describe the relationship with the cluster distribution in the GMM. The benefits of representing the underlying FC data as high-dimensional vectors come from the high dimensionality (e.g., n=17, 23, or more parameters times the number of mixtures, which may result in hundreds or thousands of dimensions), and in part from the ability to gain greater insight into the connections between different dimensions through learning.

By using a GMM trained on the entire training dataset (e.g., consisting of multiple FC datasets), the analysis platform can calculate the posterior probability for each cell-level FC dataset to determine the likelihood that the cell belongs to each "cluster" or "mixture" defined by the GMM. Fisher vectors can be used to transform the cell vector by considering the posterior probability of each cluster and the distance between the cell vector and the center vector created for each cluster. This distance used in the Fisher vector takes into account the mean vector, the covariance matrix, and the weights of the GMM, so it can represent the complex, high-order relationship between the cell vector and each cluster. Fisher vectors are an example of a method that weighs distances by posterior probabilities. With the GMM parameters, other distance functions can also be applied to estimate the relationship between cells and clusters. Finally, each FC dataset can be represented by an averaged cell, embedding the information of its posterior probability and its relationship with the cluster.

D. Training

FIG9 is a schematic flow chart illustrating a process 900 for training a model to classify blood diseases. Initially, the analysis platform may receive an input indicating the selection of one or more sources to obtain FC data (step 901). In one embodiment, the input may specify multiple databases storing separate sets of FCS files (e.g., associated with different patients and generated by different flow cytometers). In another embodiment, the input may specify multiple flow cytometers from which FCS files are obtained. In a specific embodiment in which more than one source is selected, the FC data obtained from each source is typically associated with a different group of patients. However, a patient may be included in two groups. Alternatively, the input can specify a single database or flow cytometer from which to obtain FCS files.

Then, the analysis platform can obtain multiple matrices of FC data from one or more sources, the matrices describing samples containing cells labeled with fluorescent markers (step 902). For example, the analysis platform can obtain multiple FCS files generated by the above-mentioned flow cytometer, and then the analysis platform can extract the matrix of FC data from each FCS file.

The nature of the multiple FC data matrices may depend on the analysis platform's goal of training a classification model. For example, suppose the analysis platform is interested in training a classification model to distinguish four different hematological diseases. In this case, the samples corresponding to the multiple matrices of FC data may be confirmed instances known to correspond to the four different hematological diseases. Therefore, the analysis platform may obtain at least one FC data matrix for each hematological disease of interest.

Although the contents of each FC data matrix may be different, their structure is often quite consistent. For example, each matrix may contain: FSC values, SSC values, or fluorescence values at M wavelengths from N parameters, where M and N are integers. Therefore, each matrix may contain: a first set of FSC values, a second set of SSC values, or a third set of fluorescence values.

Then, the analysis platform may execute a function that converts multiple matrices of FC data into multiple vectors of FC data (step 903). When executed, the function may independently convert each matrix of FC data into a corresponding vector of FC data. Generally, this is accomplished by using an ML algorithm. For example, as described above, each matrix of FC data may be converted into a corresponding vector of FC data using Fisher vector encoding and GMM distribution. In a specific embodiment where the function converts matrices of FC data by Fisher vector encoding, each vector may be a Fisher vector representation of the FC data contained in the corresponding matrix.

Therefore, the analysis platform can provide (i) multiple vectors of FC data and (ii) corresponding marker sets to the classification model as training data to generate a trained classification model (step 904). Each set of markers can represent the type of immunophenotype set encoded or represented in the corresponding vector, as well as the type of hematological disease represented by the corresponding sample. For example, each marker set can be used to represent the type of disease, disease state, or physiological state represented by each described cell in its corresponding vector. Therefore, the marker may not explain how the classification model learns to classify a single cell, but rather learns how to classify the entire sample (for example: in multiple blood diseases) based on the distribution of its immunophenotype set. As mentioned above, if the vector contains FC data related to more than one blood disease, then the classification model can be trained to distinguish multiple blood diseases (e.g., ALL, AML, APM, and pancytopenia).

In some embodiments, multiple vectors of FC data and corresponding sets of signatures are included in a larger training dataset used to train a classification model. The larger training dataset may further include information about one or more optical parameters and/or one or more fluorescence signature parameters. Examples of optical parameters include: forward scattered light area (FSC-A), forward scattered light width (FSC-W), forward scattered light height (FSC-H), side scattered light area (SSC-A), side scattered light width (SSC-W), and side scattered light width (SSC-H). Meanwhile, examples of fluorescent marker parameters include: CD117_PerCP-Cy5-A, KAPPA_FITC-A, HLA-DR_V450-A, CD38_APC-H7-A, and CD123_PE-A.

The analysis platform may then store the trained classification model in a data structure (step 905). As further described below, the analysis platform may then use the trained classification model to generate classifications representing recommended diagnoses for different blood disorders. Thus, the analysis platform may programmatically associate a trained classification model with each blood disorder for which it can generate a recommended diagnosis. For example, the analysis platform may populate the data structure with an identifier (e.g., an alphanumeric identifier) that identifies the blood disorder for which the classification model can generate a recommended diagnosis.

In summary, multiple vectors of FC data and corresponding sets of markers representing the types of immunophenotypes represented in the FC data can be sent to the classification model for training. Therefore, multiple vectors can represent training data that can be used to train the classification model to classify a given sample in different blood diseases. The training data used to train the classification model can include a set of high-dimensional vectors associated with different samples (i.e., different patients). Once the training is completed, the classification model can classify the samples based on the analysis of their corresponding FC data to determine different patterns of immunophenotype sets, and then determine whether the sample represents a blood disease based on the full sample distribution of the immunophenotype.

E. Classification

FIG. 10 is a schematic flow chart for illustrating a process 1000 for classifying a sample by applying a classification model. For example, assume that an analysis platform receives an input indicating a request to make a diagnosis of one or more blood diseases based on the contents of a file (step 1001). The input may represent the selection of a file (i.e., the corresponding patient) through an interface generated by the analysis platform, or the input may represent the receipt of a file (e.g., from a flow cytometer). In one embodiment, the file may be formatted in accordance with FCS. In this case, the analysis platform may extract FC data in a first form from the file and then convert the FC data into a second form to make it easier to be processed by the classification model. For example, the analysis platform can extract the matrix of FC data from the file (step 1002). Then, the analysis platform can execute a function to convert the matrix of FC data into a vector of FC data (step 1003). The function can be the same function discussed in step 903 of Figure 9 above.

The analysis platform can then provide the vector of FC data to the classification model as input to obtain one or more outputs (step 1004). Each output can represent a recommended diagnosis for a different blood disease. Therefore, the analysis platform can derive a classification for the sample based on the output results, which classification is represented by the FC data (step 1005). As described above, the number of outputs generated by the classification model can depend on the number of hematological diseases for which training data is provided during the training phase. Typically, the classification model is trained to generate outputs for multiple hematological diseases when applied to the FC data vector; however, the classification model can be trained to generate a single output for one hematological disease when applied to the FC data vector. In specific embodiments where the classification model is associated with a single blood disease, the analysis platform may apply multiple classification models that have been trained to classify different blood diseases according to the methods described herein. Additionally or alternatively, the number of outputs generated by the classification model may be based on the number of disease states defined for the specified blood disease and/or the number of digital ranges defined for MRD.

It is worth noting that the order of steps in the disclosed processes is exemplary, but the steps can be performed in various orders and combinations. For example, steps can be added to these processes, or removed from these processes. Likewise, steps can be replaced or reordered. Therefore, the description of these processes is open-ended.

Additional steps may also be included in some specific embodiments. For example, the analysis platform can generate a classification (e.g., a recommended diagnosis for a blood disease) based on the output generated by the classification model as described above. In such a case, the analysis platform can display the classification on an interface that is accessible to patients associated with basic FC data. Similarly, the analysis platform can display the classification on an interface accessible to medical professionals. In some specific embodiments, the analysis platform can be connected to a central computer system of a healthcare provider. For example, the analysis platform can access the central computer system through a data interface to access FC data. In this case, the analytics platform may be able to automatically populate the classification into the corresponding patient's electronic health record (EHR). For example, the analytics platform can transmit the classification to a central computer system and issue a command to populate the classification into the EHR for recording.

E. Usage examples

In a typical setting, the methods described in the present disclosure may be used to further examine the FC data of interest. In one embodiment, the FC data of interest may correspond to an uncertain laboratory result, and therefore the medical professional wishes to obtain further information about the result before determining an appropriate treatment plan. To achieve this goal, the analysis platform can apply a classification model to the FC data of interest. For example, assume that the classification model is trained to classify a set of immunophenotypes of different patterns in order to distinguish between multiple blood diseases (e.g., ALL, AML, APM, and total cytopenia). With the rapid and accurate classification of the classification model, the medical professional is able to select an appropriate treatment plan.

As described above, the analysis platform can execute a classification model to automatically classify diseases or physiological conditions by type. Generally speaking, the analysis platform is part of an automatic classification system (or simply "system"), which is further discussed below with reference to Figures 11 to 12. The system may include: a flow cytometer, a server system accessible via a network, a data memory, and a computer device (also known as an "electronic device" or "user device"). In some specific embodiments, the entire system is disposed in a single housing.

Generally speaking, the process of automatically classifying samples by the analysis platform begins with the user preparing the sample and inserting the sample into the flow cytometer. In one embodiment, the user can prepare a series of tubes, and each tube contains a different sample. Each tube may be subjected to different and appropriate fluorescent markers. When the series of test tubes are examined by the flow cytometer, FC data is generated and the FC data is encoded into independent files. As described above, these files can be used by the analysis platform to train classification models to produce outputs useful for diagnosis.

To ensure good coverage, the training dataset used by the analysis platform to train the classification model can be based on or obtained from a large number of files. In one embodiment, the training dataset can contain FC data for several thousand (e.g., 1,000, 2,000, or 4,000) patients who are known to have been diagnosed with ALL, AML, or APL. Each sample can be associated with a single patient, although a sample can be associated with multiple test tubes (thus generating multiple files by the flow cytometer). For example, due to dimensionality limitations, a sample set of approximately 1,000 to 2,000 samples may be associated with approximately 4,000 to 12,000 test tubes.

To further illustrate its efficacy, the framework described in this disclosure was used to develop a four-class classification model using FCS files generated by a flow cytometer (i.e., FASCantoII from Becton Dickinson Bioscience). The FCS files corresponded to approximately 550 bone marrow samples, of which approximately 100 were ALL, approximately 200 were AML, and approximately 200 were total cytopenias without blood disease. These diagnoses were determined based on conventional morphological, cytogenetic, molecular, and clinical findings. GMMs were established using raw fluorescence intensity and light scattering parameters of antibody and fluorochrome conjugates for 90% of the samples in each of the four classes. For each GMM, the gradient of each light scattering parameter is calculated using Fisher vectorization, resulting in a high-dimensional representation used to train four-class classification models.

To evaluate its performance, accuracy (ACC) was further used and was used to define the agreement between the diagnoses made by manual and automatic methods. In addition, sensitivity and specificity were further evaluated based on the area under the receiver operating characteristic (ROC) curve, also known as "AUC".

A single parameter analysis was first performed and it was found that FSC-A provided the highest accuracy compared with 36 other parameters, including 31 markers commonly used to measure the performance of FC analysis. The complete list of parameters used in the study includes: FSC-A, SSC-H, CD117_PerCP-Cy5-A, FSC-H, KAPPA_FITC-A, HLA-DR_V450-A, CD38_APC-H7-A, CD123_PE-A, FSC-W, CD34_APC-A, CD19_PE-Cy7-A, CD2_V450-A, CD14_APC-H7-A, SSC-W, CD4_PerCP-Cy5-A, CD45_V500-A, CD64_PerCP-Cy5-A, CD7_FITC-A, CD8_APC-H7-A, CD10_APC-A, SSC-A, CD7_PE-A, CD19_PerCP-Cy5-A, CD33_PE-Cy7-A. Accordingly, these parameters include optical parameters and fluorescent marker parameters.

Because one optical parameter (i.e., FSC-A) showed the best performance, all six optical parameters were further investigated and the additional effects in terms of accuracy and AUC were compared. The results are revealed in Table 1 below. As can be seen from Table 1, the combination of three optical parameters (i.e., FSC-A, SSC-H, and SSC-W) showed a reasonable accuracy of 0.921, while the combination of all six optical parameters showed an accuracy of 0.938. The analysis showed that when an additional optical parameter (i.e., FSC-W) was included, the accuracy rose to 0.928 and the AUC was 0.990. Meanwhile, adding two optical parameters (i.e., FSC-W and FSC-H) only increased the accuracy to 0.940 and the AUC to 0.991. Therefore, a classification model trained with only three parameters performs almost as well as a classification model trained with all six parameters.

Further investigation revealed that accuracy and AUC can be improved with selected fluorescent marker parameters without using all 37 fluorescent marker parameters tested. The results are shown in Table 2 below. As revealed in Table 2, the addition of a CD117 marker (i.e., CD117_PerCP-Cy5-5-A) resulted in an accuracy of 0.932 and an AUC of 0.983 - a significant improvement over the two-parameter combination of FSC-A and SSC-H. The addition of another fluorescent marker parameter (i.e., KAPPA_FITC-A) increased the accuracy to 0.948 and the AUC to 0.990. As shown in Table 2, the addition of HLA-DR_V450-A, CD38_APC-H7-A, and CD123_PE-A also provided better accuracy and AUC.

Analysis platform overview

FIG. 11 is a schematic diagram illustrating a network environment 1100 of an analysis platform 1102. A person (also referred to as a "user") can interface with the analysis platform 1102 via an interface 1104. In one embodiment, the user can access an interface through which information about a patient and a recommended diagnosis for the patient can be viewed. These interfaces 1104 can allow the user to interact with the analysis platform 1102 because it implements the framework described in this disclosure. The term "user" used in this disclosure can refer to a person who is interested in reviewing a recommended diagnosis (e.g., a patient or a medical professional, or a person who is interested in developing, training, or implementing a model).

As shown in FIG. 11 , the analysis platform 1102 can reside in a network environment 1100. Therefore, the computer device implementing the analysis platform 1102 can be connected to one or more networks 1106a to 1106b. These networks 1106a to 1106b can be a personal area network (PAN), a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a cellular network, or the Internet. For example, the analysis platform 1102 can be indirectly connected to one or more flow cytometers through the Internet (e.g., through a corresponding application programming interface), or the analysis platform 1102 can be directly connected to one or more flow cytometers (e.g., through a corresponding pipeline). In another embodiment, the analysis platform 1102 can be directly or indirectly connected to storage media managed by respective medical systems. These storage media can be part of laboratory information systems, electronic health record systems, etc. In addition or alternatively, the analysis platform 1102 can be communicatively coupled with one or more computer devices via short-range wireless connection technology (e.g., Bluetooth, Near Field Communication (NFC), Wi-Fi Direct (also known as "Wi-Fi P2P"), etc.).

The interface 1104 can be accessed via a web browser, desktop application, mobile application, or OTT application. In one embodiment, a healthcare professional can access an interface through which they can enter information about a patient. This information can include: name, date of birth, symptoms, medications, and laboratory results (e.g., in the form of an FCS file). With this information, the healthcare professional can execute the framework to generate a classification representing a recommended diagnosis. In another embodiment, an individual can access an interface through which they can identify a dataset and then monitor as the analysis platform 1102 executes the framework to train a classification model using the dataset. Accordingly, interface 1104 can be viewed on a computer device, such as a mobile workstation (also known as a "medical cart"), a personal computer, a tablet computer, a mobile phone, a wearable electronic device, etc.

In some embodiments, at least some elements of the analysis platform 1102 are hosted locally. In other words, a portion of the analysis platform 1102 may reside on a computer device used to access the interface 1104. For example, the analysis platform 1102 may be a desktop application that is executable by a mobile workstation accessed by one or more medical professionals. However, it is noted that the desktop application may be communicatively connected to a server system 1108 on which other elements of the analysis platform 1102 are hosted.

In other embodiments, the analysis platform 1102 is entirely executed by a cloud computing service (e.g., Amazon Web Services, Google Cloud Platform, Microsoft Azure). In this embodiment, the analysis platform 1102 can reside on a server system 1108 consisting of one or more computer servers. These computer servers can include models, algorithms (e.g., for processing FC data, generating reports, etc.), patient information (e.g., files, certificates, and health-related information, such as age, date of birth, disease classification, healthcare providers, etc.), and other assets. It should be understood by those skilled in the art that the information can also be distributed between the server system 1108 and one or more computer devices. For example, for security or privacy purposes, some data generated by the computer device on which the analysis platform 1102 resides may be stored and processed by the computer device.

Figure 12 is a schematic diagram illustrating an embodiment of a system 1200 that can automatically classify different patterns of immunophenotype sets to identify hematological diseases. The system 1200 may include a flow cytometer 1202 that is communicatively connected to an analysis platform. In this embodiment, the analysis platform is executed on a server system 1204 that can be accessed via a network, although the above-mentioned analysis platform can be executed elsewhere. The system 1200 also includes a data storage device 1206 and a computer device 1208. The computer device 1208 can be one of a plurality of computer devices that can be used to connect to the analysis platform. For example, in one embodiment, multiple users (e.g., medical professionals employed by a medical system) can connect to the analysis platform, and more than one computer device can be part of the system 1200. As shown in FIG. 12, the components of the system 1200 can be connected to each other in a direct or indirect manner through a network 1210. In addition or alternatively, the components of the system 1200 can be connected to each other in a communicative manner through a physical communication interface.

As described above, the functions of the network accessible server system 1204, the data storage 1206, and the computer device 1208 can be performed in a single device. Similarly, the functions of the flow cytometer 1202, the network accessible server system 1204, the database 1206, and the computer device 1208 can be performed in a single flow cytometer, in which case the flow cytometer can be referred to as a "combined flow cytometer" or "integrated flow cytometer."

Processing system

FIG. 13 is a block diagram illustrating an example of a processing system 1300 that can perform the operations described in the present disclosure. For example, the components of the processing system 1300 can be hosted on a computer device including an analysis platform (e.g., analysis platform 1102 in FIG. 11 ). In another specific embodiment, the components of the processing system 1300 can be hosted on a flow cytometer (e.g., flow cytometer 1202 in FIG. 12 ).

The processing system 1300 may include: a processor 1302, a main memory 1306, a non-volatile memory 1310, a network interface card 1312, a video display unit 1318, an input/output device 1320, a control device 1322 (e.g., a keyboard, a pointing device, or a mechanical input such as a button), a drive device 1324 including a storage medium 1326, or a signal generating device 1330, which are connected to the bus 1316 in a communicative manner. The bus 1316 is an abstract concept that represents one or more physical buses and/or point-to-point connections that are connected by appropriate bridges, adapters, or controllers. Thus, bus 1316 may include a system bus, a peripheral component interconnect (PCI) bus, a PCI-Express bus, a HyperTransport bus, an Industry Standard Architecture (ISA) bus, a Small Computer System Interface (SCSI) bus, a Universal Serial Bus (USB), an Inter-Integrated Circuit (I ² C) bus, or a bus compliant with the Institute of Electrical and Electronics Engineers (IEEE) 1394 standard.

Processing system 1300 may be similar in architecture to a computer processor such as a computer server, router, desktop computer, tablet computer, mobile phone, video game console, wearable electronic device (e.g., watch or fitness tracker), network-connected ("smart") device (e.g., television or home assistant device), augmented or virtual reality system (e.g., head mounted display), or other electronic device capable of executing a set of instructions (sequentially or otherwise) that are specified to be executed by processing system 1300.

Although main memory 1306, non-volatile memory 1310, and storage medium 1326 are shown as a single medium, the terms "storage medium" and "machine-readable medium" should be understood to include a single medium or multiple media that store instructions. The terms "storage medium" and "machine-readable medium" should also be understood to include any medium capable of storing, encoding, or carrying instructions for execution by processing system 1300.

Generally speaking, instructions executed to implement specific embodiments of the present disclosure may be implemented as part of an operating system or a specific application, component, program, object, module, or sequence of instructions (collectively referred to as a "computer program"). Computer programs typically include instructions (e.g., instructions 1304, 1308, 1328) that are located at various times in various memories and storage devices of a computing device. When read and executed by processor 1302, the instructions may cause processing system 1300 to perform operations to execute the present disclosure.

Although the specific embodiments have disclosed all functions of the computing device in this disclosure, it should be understood by those with ordinary knowledge in the field of the present invention that each specific embodiment can be released as a program product in various forms. Regardless of the specific type of machine or computer-readable medium used to actually cause the application, it falls within the scope of this disclosure. Specific examples of machine and computer-readable media include recordable media (e.g., volatile and non-volatile memory 1310), removable disks, hard disk drives, optical disks (e.g., compact disks read-only memory (CD-ROM) and digital versatile disks (DVD)), cloud storage space, and transmission media (e.g., digital and analog communication links).

The network interface card 1312 enables the processing system 1300 to mediate data with an external entity (which is external to the processing system 1300) in the network 1314 through any communication protocol supported by the processing system 1300. The network interface card 1312 may include: a network interface card, a wireless network interface card, a switch, a communication protocol converter, a gateway, a bridge, a hub, a receiver, a repeater, or a transceiver including an integrated circuit (for example: capable of communicating via Bluetooth or Wi-Fi).

Notes

The above description of various specific embodiments of the scope of the claims is for illustration and description purposes only. It is not possible to disclose in detail, nor does it limit the scope of this application to the disclosed content. For those with ordinary knowledge in the technical field of the present invention, it can be modified and changed accordingly. The selection and description of the specific embodiments disclosed in this disclosure are only to provide better principles and their practical applications, so that those with ordinary knowledge in the technical field of the present invention can understand the subject matter of this disclosure, various embodiments, and various modifications suitable for specific purposes.

Although preferred specific embodiments are disclosed, the technology disclosed herein can be implemented in many ways, no matter how detailed the description is. Therefore, the details in the specific embodiments may vary greatly, but the scope is still considered to fall within the scope of the present disclosure. The specific terms used to describe certain features or aspects of various specific embodiments should not be considered to imply that the terms are redefined in the present disclosure and are limited to any specific features, characteristics or aspects of the technology related to the terms. In general, the terms used in the scope of the following patent application should not be interpreted as limiting the technology to the specific embodiments disclosed in the present disclosure unless the terms are clearly defined in the present disclosure. Therefore, the scope of the present disclosure includes not only the disclosed specific embodiments, but also all equivalent ways of implementing or realizing the specific embodiments.

The terms used in this disclosure are primarily selected for the purpose of easy reading and guidance. The terms are not selected to define or limit specific subject matter. Therefore, the scope of the technology disclosed in this disclosure is not limited by the content of this disclosure, but should be limited by the content of the patent application scope. Therefore, the disclosure of various specific embodiments is for the purpose of illustrating rather than limiting the scope requested in the following patent application scope.

200: Framework 202: Data acquisition phase 204: Data extraction phase 206: Data conversion phase 208: Training 210: Classification

Claims (35)

A non-transitory computer-readable medium storing instructions for automatic classification of flow cytometer data, when executed by a processor of a computer device, causes the computer device to execute the following operation steps, including: Obtaining a flow cytometer data matrix, which is used to describe a sample and the sample includes cells marked with fluorescent markers; Executing a function, which converts the flow cytometer data matrix into a flow cytometer data vector; Providing (i) the flow cytometer data vector and (ii) a set of markers as input to a classification model to generate a trained classification model, wherein the set of markers is used to represent an immunophenotype set pattern for each described cell in the flow cytometer data vector; and Storing the trained classification model in a data structure. The non-transitory medium of claim 1, wherein the flow cytometer data matrix comprises fluorescence values of M wavelengths composed of N parameters, wherein the values of M and N are integers. The non-transitory medium of claim 1, wherein the flow cytometer data matrix is one of a plurality of flow cytometer data matrices obtained by the computer device; and wherein the plurality of flow cytometer data matrices correspond to different samples, and the different samples are known to represent at least two blood diseases. A non-transient medium as described in claim 3, wherein each of the plurality of flow cytometer data matrices is converted into a corresponding flow cytometer data vector to generate a plurality of flow cytometer data vectors; and wherein the plurality of flow cytometer data vectors are provided as input to the classification model, so that the classification model learns to distinguish the at least two blood diseases. The non-transitory medium of claim 1, wherein when applied to a new flow cytometer data vector corresponding to a new sample, the trained classification model produces a classification for the new sample based on sample-level analysis rather than cell-level analysis output. The non-transitory medium of claim 5, wherein the classification is used to represent a suggested diagnosis of a transfused blood disorder and is determined based on the distribution of an immunophenotype set across the new sample. The non-transitory medium of claim 1, wherein the flow cytometer data matrix comprises: a first value set of fluorescence intensity, a second value set of forward scattered light (FSC), and a third value set of side scattered light (SSC). The non-transitory medium of claim 1, wherein the flow cytometer data matrix is contained in a file and the file is received from a flow cytometer describing a sample. The non-transitory medium of claim 1, wherein the streaming cytometer data matrix is retrieved from a storage medium that is accessible by the computer device via a network. The non-transitory medium of claim 1, wherein the function converts the flow cytometer data matrix into the flow cytometer data vector by Fisher vector (FV) encoding, so that the flow cytometer data vector is a Fisher vector representation of the flow cytometer data contained in the flow cytometer data matrix. A method for automatic classification of flow cytometer data, comprising: Receiving a flow cytometer standard file (FCS file) generated by a flow cytometer, which describes a sample and the sample contains cells marked with fluorescent markers of different wavelengths; Extracting a flow cytometer data matrix from the FCS file; Converting the flow cytometer data matrix into a flow cytometer data vector; and For each cell defined in the flow cytometer data vector, Providing (i) the vector of flow cytometer data and (ii) a set of markers representing a disease type, a disease state, or a physiological state to a classification model as input to generate a trained classification model, wherein the set of markers is used to represent each described cell in the vector of flow cytometer data. A method as claimed in claim 11, wherein the FCS file is one of a plurality of FCS files received from a source, each of the plurality of FCS files corresponding to a different sample; wherein a separate vector is generated for each file in the plurality of FCS files according to a flow cytometer data corresponding matrix, thereby obtaining a plurality of flow cytometer data vectors; and wherein the classification model is trained using the plurality of flow cytometer data vectors so that the classification model learns how to distinguish between different blood diseases. The method of claim 11, wherein the transformation comprises: generating a mixture model based on the flow cytometer data matrix; and calculating a gradient of the mixture model to obtain the flow cytometer data vector. A method as described in claim 11, wherein the flow cytometer data vector and the marker set are included in a training data set and further include information about one or more optical parameters and one or more fluorescent marker parameters. A method as described in claim 14, wherein the one or more optical parameters include: forward scattered light area (FSC-A), forward scattered light width (FSC-W), side scattered light area (SSC-A), side scattered light width (SSC-W), side scattered light height (SSC-H) or any combination of the above. A method for automatic classification of flow cytometer data, comprising: receiving a request input to propose a diagnosis of a plurality of blood diseases based on an analysis of a data file; extracting a flow cytometer data matrix from the data file, which describes a sample and the sample includes cells labeled with fluorescent markers of different wavelengths; converting the flow cytometer data matrix into a flow cytometer data vector; and inputting the flow cytometer data vector into a classification model to obtain a plurality of outputs, wherein each of the plurality of outputs represents a suggested diagnosis of a corresponding blood disease in the plurality of blood diseases. A method as described in claim 16, wherein the flow cytometer data vector is a high-dimensional vector that includes, for each cell, values of (i) a forward scatter (FSC), (ii) a FSC feature, (iii) a side scatter (SSC), (iv) a SSC feature, (v) a fluorescence, and (vi) a fluorescence feature. The method of claim 17, wherein the FSC, SSC and fluorescence properties describe the same characteristic. A method as described in claim 17, wherein the FSC, SSC and fluorescence properties are selected from amplitude, frequency, amplitude variation, frequency variation, time dependence or spatial dependence. A non-transitory computer-readable medium containing instructions for automatic classification of flow cytometer data, which, when executed by a processor of a computer device, causes the computer device to perform the following operation steps, including: Receiving a flow cytometer standard file (FCS file) generated by a flow cytometer, which describes a sample and the sample includes cells marked with fluorescent markers of different wavelengths; Extracting (i) a flow cytometer data set and (ii) an overflow matrix from the FCS file; Performing a compensation operation involving the flow cytometer data set according to the overflow matrix to generate a compensated flow cytometer data set; Performing a function that performs doublet discrimination to ensure that each value contained in the compensated flow cytometer data set corresponds to a single cell; and Performing a normalization operation on the compensated flow cytometer data set to generate a normalized flow cytometer data set. The non-transitory medium of claim 20, wherein the flow cytometer data set is in the form of a matrix. The non-transitory medium of claim 20, wherein the flow cytometer data set is extracted from a data segment of the FCS file, and wherein the overflow matrix is extracted from a text segment in the FCS file. The non-transitory medium of claim 20, wherein the operation further comprises: performing a determination based on an analysis of a flow cytometer data set, wherein compensation is necessary to improve the quality of the flow cytometer data set; wherein the compensation operation is performed in response to the determination. The non-transitory medium as described in claim 20, wherein the operation further comprises: Creating a scatter plot based on the forward scattered light area value (FSC-A value) and the forward scattered light height value (FSC-H value) contained in the compensated flow cytometer data set. The non-transitory medium of claim 24, wherein when executed, the function causes the computer device to perform: (i) deleting cells whose FSC-A values in the scatter plot reach the maximum value; (ii) selecting a portion of cells still in the scatter plot; and (iii) calculating a determination coefficient between the selected cells. The non-transitory medium of claim 25, wherein the function, when executed, causes the computer device to perform: (iv) detecting whether the assay coefficient exceeds a threshold; and (v) returning data of the selected cells from the compensated flow cytometer data set in response to an assay, wherein the assay coefficient exceeds the threshold. The non-transitory medium of claim 25, wherein the function, when executed, causes the computer device to perform: (iv) detecting whether the assay coefficient exceeds a second threshold; and (v) repeatedly performing steps (ii) and (iii) and each time reducing the number of cells selected by a predetermined amount in response to an assay, and the assay coefficient does not exceed the predetermined threshold. The non-transitory medium of claim 27, wherein steps (ii) and (iii) are repeated and each time the gated cells are reduced by a predetermined amount in response to an assay until the assay coefficient exceeds the predetermined threshold. The non-transitory medium of claim 20, wherein the flow cytometer data set comprises values of a plurality of parameters. The non-transitory medium of claim 29, wherein the plurality of parameters comprises one or more optical parameters and one or more fluorescent marker parameters. The non-transient media of claim 29, wherein the normalization operation comprises: aggregating multiple values by treating each parameter in the multiple parameters as a unique feature dimension; resampling the unique feature dimensions to achieve the same sample size to ensure that each parameter in the multiple parameters has the same number of values; and normalizing the unique feature dimensions so that the multiple values have a similar scale. The non-transitory media of claim 31, wherein the normalization process involves implementing a z-score normalization technique. A method for automatic classification of flow cytometer data, comprising: Receiving a flow cytometer standard file (FCS file) generated by a flow cytometer, which describes a sample and the sample contains fluorescent markers of different wavelengths. (i) extracting a flow cytometer data matrix from a data segment of the FCS file and (ii) extracting an overflow matrix from a text segment of the FCS file; Performing a compensation operation involving the flow cytometer data matrix based on the overflow matrix to generate a compensated flow cytometer data matrix; Executing a function to perform doublet recognition (doublet discrimination) to ensure that each value contained in the compensated flow cytometer data set corresponds to a single cell; normalizing the compensated flow cytometer data set to generate a normalized flow cytometer data set; and storing the normalized flow cytometer data matrix in a memory. The method of claim 33, further comprising: generating a visual indicator of complex values in the normalized flow cytometer data matrix; and displaying the visual indicator on an interface for viewing by a person. The method of claim 34, wherein the visual indicator is a report comprising an analysis of the complex values in the normalized flow cytometer data.

Images