TWI767415B - Automatic blood analysis system and automatic blood analysis method - Google Patents

Abstract

The present disclosure provides an automatic blood analysis system including a sample loading platform, a testing platform and processor. The sample loading platform is for preprocessing a blood sample of a subject and a blood sample of a donor so as to obtain a testing blood sample and a target sample. A mixed blood sample obtained from mixing the testing blood sample and the target sample is underwent a cross-matching test by a drug reacting device of the testing platform. An image capture device of the testing platform is for capturing a three-dimensional hemagglutination image data of a processed mixed blood sample. A blood cross-matching assessment program is for assessing a result of the hemagglutination of the testing blood sample and the target sample when the blood cross-matching assessment program is executed by the processor. Therefore, the automatic blood analysis system of the present disclosure can achieve a consistency in blood sample processing, and the accuracy of the assessment of hemagglutination thereof can be further enhanced.

Description

Automated blood analysis system and automated blood analysis method

The present invention relates to a medical information analysis system and method, in particular to an automatic blood analysis system and an automatic blood analysis method.

Mass blood loss from surgery, accident, burns, childbirth, cancer chemotherapy, or other conditions such as anemia may require transfusions of blood from a patient's source to relieve symptoms or save lives. When blood transfusion treatment is required, the physician will draw the patient's blood and the blood in the blood bank for a cross-matching test to determine the blood compatibility between the patient and the blood donor to prevent the occurrence of hemolytic transfusion reactions .

The current cross-experiments are mostly performed manually by the blood bank medical examiners. Therefore, the pre-processing of blood samples and the subsequent drug experiments often show different appearances due to differences in the operating habits of different medical examiners. Affect the results of subsequent analysis and interpretation. Therefore, in order to ensure the consistency of sample processing between the patient's blood sample and the donor's blood sample, the patient's blood sample and the donor's blood sample need to be processed and analyzed by the same blood bank medical examiner. The blood samples and blood samples of the blood donors must be pre-processed in sequence, which will lead to a slow detection speed and affect the subsequent blood product distribution operations. Furthermore, the results of the cross-test must be interpreted by the blood bank medical examiners, and the combined blood assessment report can only be issued after two or more blood bank medical examiners cross-check the interpretation results. However, the results of the cross-test between the blood samples of the same patient and the blood samples of the same blood donor are often different due to the subjective interpretation habits of different blood bank medical examiners, which may lead to the accuracy of the cross-test results based on manual interpretation. There is a gap between the consistency and the actual situation.

In view of this, an automated blood analysis system is introduced on the market, which automatically draws blood samples from patients and blood donors into a blood group column card, and uses the aforementioned blood group column card as a carrier for cross-examination. After the reaction is completed, the conventional automated blood analysis system will capture the two-dimensional hemagglutination image data of the mixed blood samples in the blood group column card, and use the aforementioned two-dimensional hemagglutination image data as the interpretation basis of the cross-experiment result. However, the conventional automated blood analysis system can only capture and determine the 2D image of the upper column in the blood group column card when capturing 2D hemagglutination image data, and cannot capture or even accurately determine the sedimentation to the column Bottom and edge images. Furthermore, when the patient's blood sample and the blood donor's blood sample have a low agglutination valence or a high hemolytic valence, or when the patient is taking special drugs such as cancer-targeted drugs or immunomodulators, it is known that The automated blood analysis system also cannot achieve accurate judgment based on two-dimensional hemagglutination image data.

Therefore, how to develop a fast, low-cost and highly accurate automated blood analysis system and method is a technical subject with clinical application value.

One aspect of the present invention is to provide an automated blood analysis system, which includes a sample loading platform, a testing platform and a processor. The aforementioned sample loading platform includes a sample loading device and a blood sample loading device. The sample loading device is used for pre-processing a blood sample of a subject to obtain a test blood sample. The blood sample loading device is used for pre-loading a blood sample of a blood donor to obtain a target blood sample. The test platform is adjacent to the above-mentioned sample platform, wherein the test platform includes a reagent reaction device and an image capture device. The reagent reaction device is used for mixing the test blood sample and the target blood sample to obtain a mixed blood sample, and performing a blood cross test on the mixed blood sample to obtain a mixed blood sample after reaction. The image capturing device is used for capturing a three-dimensional hemagglutination image data of the mixed blood sample after the reaction. The processor is telecommunicationly connected to the aforementioned image capturing device, wherein the aforementioned processor includes a blood cross test evaluation program, and the blood cross test evaluation program includes a first image preprocessing module, a first training module and a first Judgment module. The first image preprocessing module is used for adjusting an image size and a background value of the aforementioned three-dimensional hemagglutination image data, so as to obtain a processed three-dimensional hemagglutination image data. The first training module uses a first neural network classifier to train the aforementioned processed 3D hemagglutination image data to converge to obtain an image feature value. The first judgment module utilizes the aforementioned first neural network classifier to output an agglutination judgment result according to the aforementioned image feature value. Wherein, the aforementioned agglutination interpretation result includes a blood agglutination valence interpretation result and a blood agglutination type interpretation result.

According to the aforementioned automated blood analysis system, the aforementioned mixed blood sample can be placed in a centrifuge tube to perform a blood cross test, and the aforementioned image capture device can capture the mixed blood sample after the reaction in the centrifuge tube Three-dimensional hemagglutination imaging data in .

According to the aforementioned automated blood analysis system, the aforementioned image capturing device may include a light source and a camera module. The light source is used for projecting a scattered light or direct light to the mixed blood sample after the reaction. The camera module includes a low-magnification imaging module and a high-magnification imaging module.

According to the aforementioned automated blood analysis system, the aforementioned three-dimensional hemagglutination image data may include at least one dynamic three-dimensional hemagglutination image data and at least one static three-dimensional hemagglutination image data.

According to the aforementioned automated blood analysis system, the aforementioned first neural network classifier may be a support vector machine (SVM) neural network classifier.

According to the aforementioned automated blood analysis system, the aforementioned image capture device can further capture a high-magnification three-dimensional hemagglutination image data of the mixed blood sample after the reaction, and the aforementioned blood cross test evaluation program can further include a first Two image preprocessing modules, a second training module and a second judgment module. The second image preprocessing module can be used for preprocessing the aforementioned high-magnification three-dimensional hemagglutination image data to obtain at least one image pixel distribution data. The second training module can use a regression analysis algorithm module to analyze the image pixel distribution data, and then use a second neural network classifier to train the image pixel distribution data to converge to obtain a double image Eigenvalue weight data. The second judgment module can utilize the aforementioned second neural network classifier to output a double agglutination interpretation result according to the weight data of the double image feature value. Wherein, the aforementioned double agglutination interpretation result may include a double blood agglutination valence interpretation result and a double blood agglutination type interpretation result.

According to the aforementioned automated blood analysis system, the aforementioned high-magnification 3D hemagglutination image data may include at least one dynamic high-magnification 3D hemagglutination image data and at least one static high-magnification 3D hemagglutination image data, and the aforementioned image pixel distribution data may include Information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between waves in pixel distribution in a single field of view, information on the distribution area of pixels in a single field of view, and information on an average peak distance of pixel distribution in a single field of view.

According to the aforementioned automated blood analysis system, the aforementioned regression analysis calculation module may be a Logistic regression calculation module.

According to the aforementioned automated blood analysis system, the aforementioned second neural network classifier may be a support vector machine neural network classifier.

According to the aforementioned automated blood analysis system, the aforementioned blood cross-test evaluation program may further include a combined blood evaluation module. The blood compatibility assessment module can train a third neural network classifier to converge a copy of the aforementioned agglutination interpretation result to output a blood compatibility assessment result.

According to the aforementioned automated blood analysis system, the aforementioned third neural network classifier may be a support vector machine neural network classifier.

In this way, the subject's blood sample and the donor's blood sample are automatically pre-loaded through the sample loading platform, and the subject's blood sample and the donor's blood sample are mixed to conduct a blood cross test, and then The three-dimensional hemagglutination image data of the mixed blood sample after the reaction is automatically analyzed and trained by the first neural network classifier, and then the blood agglutination valence interpretation result and the hemagglutination type interpretation result are further output according to the obtained image feature values. The automated blood analysis system of the present invention can not only achieve the consistency of blood product processing, greatly reduce the error of cross-test results caused by the subjective interpretation habits of different blood bank medical examiners, and improve the detection efficiency, but also can quickly carry out multi-batch analysis. The blood cross test and the interpretation accuracy of the cross test are greatly improved, so that the automated blood analysis system of the present invention has application potential in related fields.

Another aspect of the present invention provides an automated blood analysis method, comprising the following steps. A blood sample from a subject is provided. Provide a blood sample from one of the donors. A sample pre-processing step is performed, which is to perform pre-loading pre-processing on the aforementioned subject's blood sample to obtain a test blood sample, and perform the aforementioned pre-loading pre-processing on the blood sample of the blood donor to obtain a sample pre-processing step. target blood sample. A test drug reaction step is performed, which is to mix the aforementioned test blood sample and the aforementioned target blood sample to obtain a mixed blood sample, and perform a blood cross test on the mixed blood sample to obtain a post-reaction mixing blood sample. An image capture step is performed, which captures a three-dimensional hemagglutination image data of the mixed blood sample after the reaction. An image preprocessing step is performed, which adjusts an image size and a background value of the aforementioned three-dimensional hemagglutination image data, so as to obtain a processed three-dimensional hemagglutination image data. An analysis and judgment step is performed, which uses a first neural network classifier to train the aforesaid processed 3D hemagglutination image data to converge to obtain an image feature value, and uses the first neural network classifier according to The aforementioned image feature value outputs an agglutination judgment result. Wherein, the aforementioned agglutination interpretation result includes a blood agglutination valence interpretation result and a blood agglutination type interpretation result.

According to the aforementioned automated blood analysis method, wherein the aforementioned mixed blood sample can be placed in a centrifuge tube to perform a blood cross test, and the aforementioned three-dimensional hemagglutination image data can be a three-dimensional hemagglutination image of the mixed blood sample after reaction in the centrifuge tube material.

According to the aforementioned automated blood analysis method, the aforementioned three-dimensional hemagglutination image data may include at least one dynamic three-dimensional hemagglutination image data and at least one static three-dimensional hemagglutination image data.

According to the aforementioned automated blood analysis method, the aforementioned first neural network classifier may be a support vector machine neural network classifier.

According to the above-mentioned automated blood analysis method, it may further include a double analysis and judgment step, and the double analysis and judgment step may include the following steps. A double image capture step is performed, which can capture a high-magnification three-dimensional hemagglutination image data of the mixed blood sample after the reaction. A double image preprocessing step is performed, which can perform preprocessing on the aforementioned high-magnification three-dimensional hemagglutination image data, so as to obtain at least one image pixel distribution data. A double analysis step is performed, which can use a regression analysis algorithm module to analyze the image pixel distribution data, and then use a second neural network classifier to train the image pixel distribution data to convergence, so as to obtain a Double image eigenvalue weight data. A double judgment step is performed, which can output a double agglutination judgment result according to the weight data of the double image feature value by using the second neural network classifier. Wherein, the aforementioned double agglutination interpretation result may include a double blood agglutination valence interpretation result and a double blood agglutination type interpretation result.

According to the aforementioned automated blood analysis method, the aforementioned high-magnification 3D hemagglutination image data may include at least one dynamic high-magnification 3D hemagglutination image data and at least one static high-magnification 3D hemagglutination image data, and the aforementioned image pixel distribution data may include Information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between waves in pixel distribution in a single field of view, information on the distribution area of pixels in a single field of view, and information on an average peak distance of pixel distribution in a single field of view.

According to the aforementioned automated blood analysis method, the aforementioned regression analysis algorithm module can be a Logis regression algorithm module, and the aforementioned second neural network classifier can be a support vector machine neural network classifier.

According to the aforementioned automated blood analysis method, it may further include a step of performing a blood compatibility assessment, which can train a copy of the aforementioned agglutination interpretation result with a third neural network classifier to convergence, so as to output a blood compatibility assessment result.

According to the aforementioned automated blood analysis method, wherein the aforementioned third neural network classifier is a support vector machine neural network classifier.

Thereby, by mixing the blood samples of the subjects and the blood samples of the donors and performing the blood cross test, the obtained three-dimensional hemagglutination image data is automatically analyzed and trained by the first neural network classifier to output the blood agglutination. The automated blood analysis method of the present invention can not only greatly reduce the error of cross-test results caused by the subjective interpretation habits of different blood bank medical examiners and improve the detection efficiency, but also based on the three-dimensional blood The hemagglutination image data can be quickly analyzed for the results of the cross-test, so as to effectively improve the interpretation accuracy of the cross-test and facilitate the implementation of subsequent medical measures, so that the automated blood analysis method of the present invention has application potential in related fields.

Various embodiments of the present invention are discussed in greater detail below. However, this embodiment can be an application of various inventive concepts and can be embodied in various specific scopes. The specific embodiments are for illustrative purposes only, and are not intended to limit the scope of the disclosure.

[Automated blood analysis system of the present invention]

Please refer to FIGS. 1 and 2. FIG. 1 is a schematic diagram of the structure of an automated blood analysis system 100 according to an embodiment of the present invention, and FIG. 2 is a processor 130 of the automated blood analysis system 100 of FIG. 1. Schematic diagram of the architecture. The automated blood analysis system 100 includes a sample loading platform 110 , a testing platform 120 and a processor 130 .

The sample loading platform 110 includes a sample loading device 111 and a blood sample loading device 112 . The sample loading device 111 is used for pre-loading a blood sample of a subject to obtain a test blood sample, and the blood sample loading device 112 is used for loading a blood sample of a donor. Sample pretreatment to obtain a target blood sample. In detail, the test blood sample may be serum or blood cells of the subject's blood sample, and the target blood sample may be serum or blood cells of the donor's blood sample. For example, before performing blood transfusion treatment on a patient, the blood bank medical examiner will draw a blood sample from the patient (ie the blood recipient) and centrifuge the patient's blood sample and the blood sample from the donor respectively, and obtain the patient's blood. The serum of the sample and the red blood cells of the blood sample of the donor are subjected to a major cross matching test (Major Cross Matching Test) to confirm the compatibility of the patient's blood and the blood of the donor, and determine whether the blood of the donor is suitable for the patient according to the results of the major cross test. lose. Similarly, in the embodiment of FIG. 1, the sample loading device 111 and the blood sample loading device 112 can centrifuge the blood sample of the subject and the blood sample of the donor respectively to obtain serum or blood cells, so as to avoid The cross-contamination of blood samples caused by the operation method during centrifugation or drawing of serum or blood cells occurs. Furthermore, the blood sample loading device 112 can automatically brew the red blood cells of the blood sample from the blood donor with 3% to 5% normal saline, so as to avoid the concentration of the red blood cell sample being too high or too low and affecting the subsequent crossover. the accuracy of the test. In addition, the sample loading device 111 and the blood sample loading device 112 of the sample loading platform 110 can separately pre-process the patient's blood sample and the donor's blood sample at the same time, so as to greatly save processing time and labor costs and improve detection rate.

The test platform 120 is adjacent to the sample loading platform 110 , and the test platform 120 includes a reagent reaction device 121 and an image capture device 122 . The reagent reaction device 121 is used for mixing the aforementioned test blood sample and the target blood sample to obtain a mixed blood sample, and performing a blood cross test on the mixed blood sample to obtain a mixed blood sample after reaction. The image capture device 122 is used to capture a three-dimensional hemagglutination image data of the mixed blood sample after the reaction, so as to facilitate subsequent interpretation of results. Specifically, the mixed blood sample is placed in a centrifuge tube for the blood cross test, and the image capture device 122 captures the three-dimensional hemagglutination image data of the mixed blood sample after the reaction in the centrifuge tube, so as to avoid the conventional use of blood type It is difficult to observe the sediment particles caused by the column card as the reagent reaction carrier.

In detail, in the automated blood analysis system 100 of the present invention, the reagent reaction device 121 will further automatically perform the manual polybrene method, the traditional antiglobulin test or other blood samples on the mixed blood samples. Crossover test, at this time, the reagent reaction device 121 will use an automatic pipette or other equipment for liquid transfer to absorb an appropriate amount of mixed blood samples and reagents for mixing, and use an automatic shaking device and other equipment to perform isokinetic shaking to make The centrifuge tube is subjected to uniform force to complete the reaction of the reagent, so as to avoid problems such as uneven reaction caused by manual shaking, which will affect the interpretation of blood agglutination valence and the interpretation of hemagglutination type.

In addition, in the embodiment of FIG. 1 , the image capturing device 122 may include a light source 123 and a camera module 124 . The light source 123 can be used to project a scattered light or direct light to the mixed blood sample after the reaction, and the camera module 124 can include a low-magnification imaging module (not shown) and a high-magnification imaging module (not shown in the figure). shown). Specifically, the light source 123 can project the scattered light to the mixed blood sample after the reaction, so that the camera module 124 can capture the three-dimensional hemagglutination image by the low-magnification imaging module or the high-magnification imaging module for interpreting the blood agglutination valence. data, and can project direct light to the mixed blood sample after the reaction, so that the camera module 124 can capture the three-dimensional hemagglutination image data for interpreting the hemagglutination type by the low-magnification imaging module or the high-magnification imaging module. Furthermore, the 3D hemagglutination image data captured by the image capture device 122 of the present invention may include at least one dynamic 3D hemagglutination image data and at least one static 3D hemagglutination image data to facilitate subsequent analysis.

In addition, it should be noted here that the sample loading platform 110 and the testing platform 120 of the present invention can be configured with different specifications of centrifuges separately or jointly depending on the number of samples to be processed by the automated blood analysis system 100 of the present invention or different testing requirements. , automatic pipettes and mechanical grippers of different specifications, or other devices such as sample conveying tracks are configured to facilitate centrifugation and reagent testing, but the present invention is not limited to the contents disclosed in the drawings.

The processor 130 is telecommunicationly connected to the image capturing device 122 , wherein the processor 130 includes a blood cross test evaluation program 131 . As shown in FIG. 2 , the blood cross test evaluation program 131 includes a first image preprocessing module 140 , a first training module 150 and a first judgment module 160 .

The first image preprocessing module 140 is used for adjusting an image size and a background value of the aforementioned three-dimensional hemagglutination image data, so as to obtain a processed three-dimensional hemagglutination image data. Specifically, the first image preprocessing module 140 will further adjust the image size of the 3D hemagglutination image data to 1920 pixels (pixel)×1080 pixels (200 times magnification), and remove the non-blood cells in the 3D hemagglutination image data. Background image information can reduce interference caused by background hemolysis or lipidemia, or patients taking special drugs such as cancer target drugs or immunomodulators when interpreting conventional cross-experiment results, thereby improving the automated blood analysis system 100 of the present invention. The results are interpreted correctly.

In addition, the first image preprocessing module 140 may be Image J image analysis software, but the present invention is not limited to this.

The first training module 150 uses a first neural network classifier to train the processed 3D hemagglutination image data to converge to obtain an image feature value. Specifically, the automated blood analysis system 100 of the present invention can use the first neural network classifier of the blood cross test evaluation program 131 to automatically analyze the image information of the processed three-dimensional hemagglutination image data, and automatically extract corresponding images The characteristic value is used to improve the evaluation efficiency of the automated blood analysis system 100 of the present invention. Furthermore, the aforementioned first neural network classifier may be a support vector machine (SVM) neural network classifier, but the present invention is not limited thereto.

The first judgment module 160 utilizes the first neural network classifier to output an agglutination judgment result according to the aforementioned image feature values. The agglutination interpretation result includes the blood agglutination valence interpretation result and the hemagglutination type interpretation result. For details, please refer to Table 1. Table 1 is the current clinical criteria for judging blood agglutination strength, and the automated blood analysis system 100 of the present invention is based on the current clinical judgment criteria for blood agglutination strength. Interpretation of hemagglutination pattern is in line with the current clinical diagnostic basis and facilitates subsequent blood transfusion therapy. Table I blood agglutination value judgement standard 4+ A complete large clot with a clear background and no free red blood cells is judged to be strongly positive 3+ Several large clots with clear background and almost no free red blood cells were judged as strong positive 2+ Many medium-large clots with clear background and less than 1/2 free red blood cells were judged as positive 1+ Many small agglutinates with cloudy background and more than 1/2 of free red blood cells are judged to be positive +/- Agglutination barely visible to the naked eye, and small agglutination groups can be seen under the microscope, which is judged as weak positive 0 No agglutination was observed with the naked eye, and the red blood cells were free under the microscope, which was judged as negative. Mf Under the microscope, a small number of red blood cells can be seen to form a large clot, and there are a large number of free red blood cells around, which is judged as a mixed agglutination appearance PH The red blood cells partially disappeared, and the cell supernatant was clear red, which was judged to be positive. H Complete hemolysis, complete disappearance of red blood cells, and clear red cell supernatant, judged as positive

Please refer to FIG. 1 and FIG. 3 at the same time. FIG. 3 is a schematic structural diagram of another processor 130 a of the automated blood analysis system 100 of FIG. 1 . Specifically, the processor 130a includes a blood cross test evaluation program 131a, and the blood cross test evaluation program 131a includes a first image preprocessing module 140a, a first training module 150a, and a first judgment module 160a , a second image preprocessing module 170a, a second training module 180a and a second judgment module 190a, wherein the first image preprocessing module 140a, the first training module 150a and the first judgment module 160a They are the same as the first image preprocessing module 140 , the first training module 150 and the first determination module 160 in FIG. 2 , and will not be repeated here.

In the embodiment of FIG. 3 , the image capturing device 122 further captures a high-magnification three-dimensional hemagglutination image data of the mixed blood sample after the reaction for subsequent evaluation. Specifically, the image capturing device 122 captures a high-magnification three-dimensional hemagglutination image data of the mixed blood sample after the reaction by the high-magnification image capturing module of the camera module 124 . In the current clinical practice, when the blood agglutination valence interpretation result is below +1, it is necessary to further confirm and interpret the blood agglutination result by manual microscopic examination through a microscope. In addition, Rouleaux Formation also needs to be examined and interpreted through a microscope. In this way, not only is it easy to cause errors in the interpretation results of blood agglutination valence and hemagglutination type due to the subjective interpretation habits of different blood bank medical examiners, but also the detection efficiency is not ideal. In view of this, the automated blood analysis system 100 of the present invention further uses the high-magnification imaging module to capture the high-magnification three-dimensional hemagglutination image data of the mixed blood sample whose blood agglutination valence is less than +1 by the first judgment module 160a. , and based on the aforementioned high-magnification three-dimensional hemagglutination image data for more detailed analysis and interpretation.

The second image preprocessing module 170a is used for preprocessing the aforementioned high-magnification three-dimensional hemagglutination image data to obtain at least one image pixel distribution data. Specifically, the high-magnification three-dimensional hemagglutination image data of the present invention may include at least one dynamic high-magnification three-dimensional hemagglutination image data and at least one static high-magnification three-dimensional hemagglutination image data, and the second image preprocessing module 170a will further use The image processing software selects the dynamic high-magnification 3D hemagglutination image data and the static high-magnification 3D hemagglutination image data in one or more fields of view to be tested, and adds parameters such as flow velocity and exposure trajectory to analyze the pixel distribution information to output at least one. Image pixel distribution data, wherein the image pixel distribution data may further include information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between waves in pixel distribution in a single field of view, information on pixel distribution area in a single field of view, or average pixel distribution in a single field of view Peak distance information for subsequent analysis.

Specifically, if the field to be tested of the dynamic high-magnification 3D hemagglutination image data and the static high-magnification 3D hemagglutination image data is an image of a single cell without agglutination, the single-field pixel distribution peak information of the image pixel distribution data includes: Information such as the number of wave peaks, and the area under the graph of the single-field pixel distribution peak information is the single-field pixel distribution area information; if the dynamic high-magnification 3D hemagglutination image data and the static high-magnification 3D hemagglutination image data The field of view to be measured is In the image of cell aggregates, the single-view pixel distribution peak information of the image pixel distribution data also includes information such as the number of peaks, and the area under the graph of the single-view pixel distribution peak information is the single-view pixel distribution area information.

In addition, the second image preprocessing module 170a may be Image J image analysis software, but the present invention is not limited thereto.

The second training module 180a uses a regression analysis algorithm module to analyze the image pixel distribution data, and then uses a second neural network classifier to train the image pixel distribution data to converge to obtain a dual image feature value weight data.

Specifically, the image pixel distribution data output by the second image preprocessing module 170a of the present invention includes the single-field pixel distribution peak information and the single-field pixel distribution area information of the static high-magnification three-dimensional hemagglutination image data, and the dynamic high-resolution image data. The information on the number of peaks of pixel distribution in a single field of view, the information on the number of peaks of pixel distribution in a single field of view, the area information of pixel distribution in a single field of view, and the information on the average peak distance of pixel distribution in a single field of view of the magnification 3D hemagglutination image data. The group 180a will automatically use the regression analysis algorithm module to analyze the above-mentioned information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between pixels in a single field of view, information on the area of pixel distribution in a single field of view, and information on the average peak distance of pixel distribution in a single field of view. Perform analysis, and classify the image pixel distribution data according to the reference image data in the reference database of the regression analysis algorithm module (no agglutination of a single erythrocyte and agglutination of 3 to 5 blood cells) to determine whether it is under the field of view to be tested. There are microscopic agglomerations that cannot be detected by the naked eye.

Next, the second training module 180a automatically adjusts the parameter settings of the second neural network classifier according to the classification result of the image pixel distribution data by the regression analysis algorithm module, and then uses the set second neural network to classify the data. The device trains the image pixel distribution data to converge to obtain dual image eigenvalue weight data, and further outputs the dual agglutination interpretation result according to the aforementioned dual image eigenvalue weight data, and the dual agglutination interpretation result may include dual blood agglutination The valence interpretation result and the hemagglutination pattern interpretation result can improve the analysis accuracy and analysis integrity of the automated blood analysis system 100 of the present invention.

Specifically, the aforementioned regression analysis algorithm module can be a Logistic regression algorithm module, and the second neural network classifier can be a support vector machine neural network classifier.

Please refer to FIG. 1 and FIG. 4 at the same time. FIG. 4 is a schematic structural diagram of another processor 130b of the automated blood analysis system 100 of FIG. 1 . Specifically, the processor 130b includes a blood cross test evaluation program 131b, and the blood cross test evaluation program 131b includes a first image preprocessing module 140b, a first training module 150b, and a first judgment module 160a and a combined blood evaluation module 170b, wherein the first image preprocessing module 140b, the first training module 150b and the first judgment module 160b and the first image preprocessing module 140 and the first training module in FIG. 2 The group 150 is the same as the first judging module 160 , which will not be repeated here.

In the embodiment shown in FIG. 4 , the combined blood evaluation module 170b uses a third neural network classifier to train a copy of the agglutination interpretation result output by the first judgment module 160b to converge to output a blood compatible The blood compatibility evaluation results will further indicate whether the blood coagulation results between the subject's blood sample and the blood donor's blood sample are positive or negative, so as to facilitate the subsequent blood distribution and blood transfusion treatment by the blood bank. Specifically, the third neural network classifier may be a support vector machine neural network classifier.

Thereby, the automated blood analysis system 100 of the present invention automatically performs pre-loading processing and blood cross-experiment on the blood sample of the subject and the blood sample of the blood donor through the sample loading platform 110, so as to achieve the consistency of blood product processing. The three-dimensional hemagglutination image data of the mixed blood samples after the reaction is automatically analyzed and trained by the first neural network classifier, and then the blood agglutination valence interpretation result and hemagglutination pattern are output according to the obtained image feature values. Interpreting the results can greatly reduce the error of the cross-test results caused by the subjective interpretation habits of different blood bank medical examiners, and can reduce the interference of hemolysis, lipidemia or drugs in the interpretation of the conventional cross-test results. Furthermore, the automated blood analysis system 100 of the present invention can further use the second neural network classifier and the third neural network classifier to perform the analysis on the three-dimensional hemagglutination image data of the mixed blood sample or the high-magnification three-dimensional hemagglutination image data. Training and classification can avoid errors and manpower loss caused by manual interpretation with microscopes, so as to improve the detection efficiency and quickly perform multiple batches of blood cross tests, thereby making the automated blood analysis system 100 of the present invention highly applicable On the premise of convenience and high detection efficiency, it has excellent interpretation accuracy and has application potential in related fields.

[Automated blood analysis method of the present invention]

Please refer to FIG. 5 , which is a flowchart illustrating the steps of an automated blood analysis method 200 according to another embodiment of the present invention. The automated blood analysis method 200 includes step 210 , step 220 , step 230 , step 240 , step 250 , step 260 and step 270 .

Step 210 is to provide a blood sample from a subject.

Step 220 is to provide a blood sample from a blood donor.

Step 230 is to perform a sample pre-processing step, which is to perform pre-loading processing on the blood sample of the subject to obtain a test blood sample, and perform pre-loading processing on the blood sample of the blood donor to obtain a target. blood sample. In detail, the test blood sample can be serum or blood cells of the blood sample of the subject, and the target blood sample can be serum or blood cells of the blood sample of the donor, and the automated blood analysis method 200 of the present invention can respectively The subject's blood sample and the blood donor's blood sample are centrifuged to obtain their serum or blood cells, so as to avoid cross-contamination of blood samples caused by the operation when centrifuging or drawing serum or blood cells.

Step 240 is to perform a reagent reaction step, which is to mix the test blood sample and the target blood sample to obtain a mixed blood sample, and perform a blood cross test on the mixed blood sample to obtain a mixed blood sample after reaction. Specifically, the reagent reaction step is to automatically perform manual polybrene method, traditional three-phase method or other blood cross test on the mixed blood sample, and the mixed blood sample is placed in a centrifuge tube to carry out the aforementioned blood cross test, so that To avoid the conventional problem that the sediment particles are not easy to observe due to the conventional use of blood group column card as the reagent reaction carrier.

Step 250 is to perform an image capture step, which is to capture a three-dimensional hemagglutination image data of the mixed blood sample after the reaction. Specifically, since the mixed blood sample is placed in a centrifuge tube for the blood crossover test, the image capture step is to capture a three-dimensional hemagglutination image data of the mixed blood sample in the centrifuge tube after the reaction, so as to clearly observe the inside of the centrifuge tube. Hemagglutination at different sites. Furthermore, the three-dimensional hemagglutination image data may include at least one dynamic three-dimensional hemagglutination image data and at least one static three-dimensional hemagglutination image data, so as to facilitate subsequent analysis.

Step 260 is an image preprocessing step of adjusting an image size and a background value of the three-dimensional hemagglutination image data to obtain a processed three-dimensional hemagglutination image data. Specifically, the image size of the 3D hemagglutination image data will be further adjusted to 1920 pixels (pixel) × 1080 pixels (200 times magnification) and the background image information of non-blood cells in the 3D hemagglutination image data will be removed to reduce the known Interference caused by hemolysis or lipidemia, or the patient taking special drugs such as cancer target drugs or immunomodulators in the interpretation of the cross-experiment results, further improves the accuracy of interpretation of the results of the automated blood analysis method 200 of the present invention.

Step 270 is to perform an analysis and judgment step, which is to use a first neural network classifier to train the processed 3D hemagglutination image data to converge to obtain an image feature value, and use the aforementioned first neural network classifier according to The image feature value outputs an agglutination interpretation result. Specifically, the first neural network classifier will automatically analyze the image information of the processed three-dimensional hemagglutination image data, and automatically extract the corresponding image feature values, and continue the analysis with the aforementioned image feature values to output agglutination. Interpretation results, among which, the agglutination interpretation results include blood agglutination valence interpretation results and hemagglutination type interpretation results, so as to meet the requirements of accurate interpretation under the premise of meeting the current clinical judgment criteria for blood agglutination intensity. Furthermore, the aforementioned first neural network classifier may be a support vector machine neural network classifier, but the present invention is not limited thereto.

Please refer to FIG. 6 and FIG. 7. FIG. 6 is a flowchart showing the steps of the automated blood analysis method 200a according to another embodiment of the present invention, and FIG. 7 is a flowchart of the automated blood analysis method 200a of FIG. 6. The flow chart of the steps of the double analysis judgment steps. The automated blood analysis method 200a includes step 210a, step 220a, step 230a, step 240a, step 250a, step 260a, step 270a and step 280a, wherein step 210a, step 220a, step 230a, step 240a, step 250a, step 260a and step 280a 270 a is the same as step 210 , step 220 , step 230 , step 240 , step 250 , step 260 and step 270 in FIG. 5 , and will not be repeated here.

In the embodiment shown in FIG. 6 and FIG. 7 , step 280 a is to perform a dual analysis and determination step, and the dual analysis determination step includes step 281 , step 282 , step 283 and step 284 .

Step 281 is to perform a double image capture step, which is to capture a high magnification three-dimensional hemagglutination image data of the mixed blood sample after the reaction, so as to further confirm when the first neural network classifier analyzes the three-dimensional image of the mixed blood sample. The detailed hemagglutination situation when the agglutination interpretation result obtained from the hemagglutination image data is less than +1.

Step 282 is to perform a double image preprocessing step, which is to perform preprocessing on the aforementioned high-magnification three-dimensional hemagglutination image data to obtain at least one image pixel distribution data. Specifically, the high-magnification 3D hemagglutination image data of the present invention may include at least one dynamic high-magnification 3D hemagglutination image data and at least one static high-magnification 3D hemagglutination image data, and the double image preprocessing step will further use image processing The software frame selects the dynamic high-magnification 3D hemagglutination image data and one or more fields to be tested for the static high-magnification 3D hemagglutination image data, and adds parameters such as flow velocity and exposure trajectory to analyze the pixel distribution information, and outputs at least one image pixel Distribution data, wherein the image pixel distribution data may further include information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between pixel distribution in a single field of view, information on the distribution area of pixels in a single field of view, and an average peak in pixel distribution in a single field of view distance information for subsequent analysis.

Step 283 is to perform a double analysis step, which uses a regression analysis algorithm module to analyze the image pixel distribution data, and then uses a second neural network classifier to train the image pixel distribution data to converge to obtain a double image Eigenvalue weight data. Specifically, the regression analysis algorithm module will first classify the image pixel distribution data according to a threshold, and then automatically adjust the second neural network classifier according to the classification result of the regression analysis algorithm module on the image pixel distribution data. The parameters are set, and the set second neural network classifier is used to train the image pixel distribution data to converge to output dual image feature value weight data. Specifically, the aforementioned regression analysis algorithm module can be a Logis regression algorithm module, and the second neural network classifier can be a support vector machine neural network classifier.

Step 284 is to perform a double judgment step, which utilizes the second neural network classifier to output a double hemagglutination interpretation result according to the aforementioned double image feature value weight data. Wherein, the double agglutination interpretation result may include the double blood agglutination valence interpretation result and the double blood agglutination type interpretation result.

Specifically, the image pixel distribution data of the present invention includes information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between pixels in a single field of view, pixel distribution area information in a single field of view, and information on the area of pixel distribution in a single field of view of the high-magnification three-dimensional hemagglutination image data. The pixel distribution average peak distance information, and the aforementioned single-view pixel distribution peak information, single-view pixel distribution area information, and single-view pixel distribution average peak distance information will be automatically analyzed by the regression analysis algorithm module. The reference image data in the reference database of the regression analysis algorithm module (no agglutination of a single blood cell and agglutination of 3 to 5 blood cells) are used to classify the image pixel distribution data to determine whether there is any undetectable field of view that cannot be detected by the naked eye. of fine agglomeration.

Next, the second neural network classifier will automatically adjust its parameter settings according to the classification result of the image pixel distribution data by the regression analysis algorithm module, and then use the second neural network classifier that has been set to train the image pixel distribution data The double image eigenvalue weight data is obtained until convergence, and the double agglutination interpretation result is output according to the aforementioned double image eigenvalue weight data. The result of the hemagglutination pattern interpretation is used to improve the analysis accuracy and analysis integrity of the automated blood analysis method 200a of the present invention.

Please refer to FIG. 8 , which is a flowchart showing the steps of an automated blood analysis method 200b according to another embodiment of the present invention. The automated blood analysis method 200b comprises step 210b, step 220b, step 230b, step 240b, step 250b, step 260b, step 270b and step 280b, wherein step 210b, step 220b, step 230b, step 240b, step 250b, step 260b and step 280b 270b is the same as step 210, step 220, step 230, step 240, step 250, step 260 and step 270 in FIG. 5, and will not be repeated here.

In the embodiment of FIG. 8, step 280b is a step of performing a blood compatibility evaluation, which is to train a copy of the agglutination interpretation result with a third neural network classifier to convergence, so as to output a blood compatibility evaluation result , and the blood compatibility assessment result will further indicate whether the blood coagulation result between the blood sample of the subject and the blood sample of the blood donor is positive or negative, so as to facilitate the subsequent blood distribution and blood transfusion treatment by the blood bank. Specifically, the third neural network classifier may be a support vector machine neural network classifier.

Thereby, the automated blood analysis method 200, the automated blood analysis method 200a and the automated blood analysis method 200b of the present invention perform a blood cross test by mixing the blood sample of the subject and the blood sample of the donor, and agglutinating the obtained three-dimensional blood cells. The image data is automatically analyzed and trained by the first neural network classifier to output the interpretation results of blood agglutination valence and hemagglutination type, which can not only greatly reduce the cross-experiment caused by the subjective interpretation habits of different blood bank medical examiners The result error and the improvement of the detection efficiency can also be used to quickly analyze the results of the cross-test according to the three-dimensional hemagglutination image data of the blood, so as to effectively improve the interpretation accuracy of the cross-test and facilitate the implementation of subsequent medical measures, thereby enabling the automation of the present invention. The blood analysis method 200, the automated blood analysis method 200a, and the automated blood analysis method 200b have application potential in related fields. Furthermore, the automated blood analysis method 200a and the automated blood analysis method 200b of the present invention can further use the second neural network classifier or the third neural network classifier to analyze the three-dimensional hemagglutination image data or high-magnification three-dimensional image data of mixed blood samples. Hemagglutination image data are trained and classified to avoid errors and manpower loss caused by manual interpretation with microscopes, to improve detection efficiency, and to quickly perform multiple batches of blood cross tests, thereby enabling the automated blood analysis of the present invention. The method 200 , the automated blood analysis method 200a and the automated blood analysis method 200b have excellent interpretation accuracy under the premise of high convenience of use and detection efficiency, and have application potential in related fields.

[Example]

Please refer to FIG. 9 , which is a schematic diagram of an automated blood analysis system 300 according to an example of an embodiment of the present invention. The automated blood analysis system 300 includes a sample loading platform 310 , a testing platform 320 and a processor 330 .

The sample loading platform 310 includes a sample loading device 311 , a blood sample loading device 312 , a centrifuge 313 and a mechanical gripper 302 , and the test platform 320 includes a reagent reaction device 321 and an image capturing device 322 . Specifically, in the embodiment shown in FIG. 9 , the sample loading device 311 may include a sample loading track, and the blood sampling device 312 may include a blood sampling drawer, but the present invention is not based on the drawings. The content disclosed in the form is limited.

For example, when the automated blood analysis system 300 of the present invention is used to perform manual polybrene analysis on the blood sample of the subject and the blood sample of the donor stored in the blood bank, the blood collection tube containing the blood sample of the subject (not shown) will be placed on the sample loading track of the sample loading device 311, so that the sample loading track will send the blood collection tube into the automated blood analysis system 300 for processing, and the blood sample of the donor will be sent to another blood collection tube ( (not shown) is contained and put into the blood sample loading drawer of the blood sample loading device 312 . Next, the user can operate the mechanical gripper 302 of the sample loading platform 310 through the user operation interface 301 to transfer the aforementioned two blood collection tubes to the centrifuge 313 for centrifugation to obtain the serum and blood supply of the subject's blood sample Blood cells of the person's blood sample. Next, the serum of the subject's blood sample and the blood cells of the donor's blood sample will be transferred from the unopened blood collection tube to a centrifuge placed on the processed sample placing platform 314 by using the puncture automatic pipette 303 respectively. In the tube (not shown), another mechanical gripper 304 moves the aforementioned centrifuge tube into the reagent reaction device 321 and adds the brethamine reagent to obtain a mixed blood sample. Next, the aforementioned centrifuge tube containing the mixed blood sample will be reacted in a reaction tank at 37°C. After the reaction is completed, the mechanical gripper 304 will quickly move the centrifuge tube to the automatic centrifuge tube shaking device 323 to perform uniform shaking with a stable force and then centrifuge. Next, the mechanical gripper 304 will again transfer the aforementioned centrifuge tube into the field of view of the image capture device 322 to capture the three-dimensional hemagglutination image data of the mixed blood sample after the reaction, wherein the image capture device 322 continues to capture 15 second, to obtain the three-dimensional hemagglutination image data of the middle section of the centrifuge tube, and the angle between the lens of the image capturing device 322 and the centrifuge tube is about 70-90 degrees. Next, the image data of the first 10 seconds in the three-dimensional hemagglutination image data will be transmitted to the processor 330 for subsequent analysis, wherein the blood cross test evaluation program of the processor 330 may be the blood cross test evaluation program 131, The blood cross-test evaluation program 131a in the embodiment in FIG. 3 or the blood cross-test evaluation program 131b in the embodiment in FIG. 4 is described in the previous paragraph for related details, which will not be repeated here. After the image analysis and interpretation are completed, the agglutination interpretation result, the double agglutination interpretation result and the blood compatibility evaluation result will be further displayed on the user operation interface 301 to facilitate subsequent blood distribution and blood transfusion treatment by the blood bank.

However, it should be noted here that the detailed structure and component arrangement shown in the automated blood analysis system 300 of the present invention can be configured according to requirements. For example, the number of mechanical grippers 304 can be configured as single or multiple, centrifugal The number of the machine 313 can also be configured as single or multiple, and other equipments that can achieve sample loading, reagent reaction or analysis can be further added to meet the needs of different types of cross-examination, but the present invention is not disclosed in drawings. content is limited.

Please refer to Figure 10A and Figure 10B, Figure 10A is a static three-dimensional hemagglutination image data of the present invention, and Figure 10B is a dynamic three-dimensional hemagglutination image data of the present invention. In detail, the image capture device of the automated blood analysis system of the present invention can capture low-magnification static three-dimensional hemagglutination image data and low-magnification dynamic three-dimensional hemagglutination image data for subsequent image analysis, and can effectively remove the image data. Present the image information of the non-blood cell background in the dynamic or static 3D hemagglutination image data, and can further distinguish the different degrees of hemagglutination status (as shown by the red circle in Figure 10A and Figure 10B), so as to reduce the known Interference caused by background hemolysis, lipemia or some drugs in the interpretation of cross-test results can accurately determine blood agglutination reactions with blood agglutination valence less than +1 in the centrifuge tube.

Specifically, the static three-dimensional hemagglutination image data in Figure 10A is interpreted as the blood agglutination valence +1 and the background lipid valence +3 after being analyzed by the automated blood analysis system and automated blood analysis method of the present invention, while Figure 10B The dynamic three-dimensional hemagglutination image data is interpreted as the blood agglutination value +1 and the background hemolysis value +2 after being analyzed by the automatic blood analysis system and the automatic blood analysis method of the present invention, and the above-mentioned interpretation results are all consistent with clinical interpretation. The results are consistent, showing that the automated blood analysis system and the automated blood analysis method of the present invention have excellent blood analysis accuracy, and can avoid the conventional error and labor loss caused by manual interpretation with a microscope, so as to improve the detection efficiency and quickly Carry out multiple batches of blood cross-tests, and have application potential in related fields.

Please refer to Fig. 11, Fig. 12A and Fig. 12B. Fig. 11 is a high-magnification static three-dimensional hemagglutination image data of the present invention, and Fig. 12A is a single agglutination of the high-magnification static three-dimensional hemagglutination image data of Fig. 11 The single-field pixel distribution peak information analysis diagram of the particles, Figure 12B is the single-field pixel distribution area information analysis diagram of the multi-agglutinated particles in the high-magnification static three-dimensional hemagglutination image data of Figure 11. Specifically, when the agglutination interpretation result obtained by the first neural network classifier of the present invention by analyzing the three-dimensional hemagglutination image data of the mixed blood sample is below +1, the image capture device will further capture the static high value of the mixed blood sample. The magnification three-dimensional hemagglutination image data can be extracted to further extract the image pixel distribution data for subsequent analysis.

As shown in Fig. 11, Fig. 12A and Fig. 12B, the image processing software of the image capture device will automatically select the field of view of a single blood cell without agglutination in the static high-magnification 3D hemagglutination image data, and increase the flow rate and Exposure trajectory and other parameters to analyze the single-field pixel distribution peak information of single agglutinated particles in the field of view (Fig. 12A), and automatically select the field of view with fine agglutination of blood cells in the static high-magnification 3D hemagglutination image data, and increase the Flow velocity and exposure trajectory and other parameters to analyze the pixel distribution area information of the multi-aggregated particles in the field of view (Fig. 12B), and then calculate the reference image data in the reference database of the module according to the regression analysis (a single blood cell has no Agglutination and agglutination of 3 to 5 blood cells) classify the peak information of the pixel distribution in the single field and the distribution area information of the pixel in the single field to determine whether there is a fine agglutination that cannot be detected by the naked eye in the field of view to be tested, and then use the second The neural network classifier trains the aforementioned image pixel distribution data to converge to obtain double image feature value weight data, and further outputs a double agglutination interpretation result according to the aforementioned double image feature value weight data.

Furthermore, please refer to Fig. 13A, Fig. 13B, Fig. 14A, Fig. 14B, Fig. 14C, Fig. 15A, Fig. 15B and Fig. 15C. Fig. 13A is the high-magnification dynamic three-dimensional hemagglutination image data of the present invention without agglutination reaction, Fig. 13B is the high-magnification dynamic three-dimensional hemagglutination image data of the present invention with agglutination reaction, and Fig. 14A is Fig. 13A The high-magnification dynamic three-dimensional hemagglutination image data of the single-field image of a single-field image of agglutinated particles within 10 seconds. Figure 14B is a single-field image of the single-field image within 8 seconds corresponding to the image change map of Figure 14A. Pixel distribution peak information analysis diagram, Figure 14C is an enlarged view of the single-field pixel distribution peak information analysis diagram of the single-field image at the 4th second in Figure 14B, and Figure 15A is the high-magnification dynamic three-dimensional blood cell in Figure 13B. The image change diagram of the single-field image of multiple agglutinated particles within 10 seconds of agglutination image data. Figure 15B is the single-field image pixel distribution peak information analysis of the single-field image within 8 seconds corresponding to the image change diagram of Figure 15A. Fig. 15C is an enlarged view of the single-field pixel distribution peak information analysis graph of the single-field image at the 5th second in Fig. 15B.

As shown in Fig. 13A, Fig. 14A, Fig. 14B and Fig. 14C, when the first neural network classifier of the present invention analyzes the three-dimensional hemagglutination image data of the mixed blood sample, the agglutination interpretation result obtained is +1 or less , in order to more accurately assess the blood agglutination situation, the image capture device will further capture the dynamic high-magnification 3D hemagglutination image data (Fig. 14A) of the mixed blood sample that changes continuously for 15 seconds, and the image processing software of the image capture device The field of view of a single blood cell without agglutination in the dynamic high-magnification 3D hemagglutination image data will be automatically framed, and parameters such as flow velocity and exposure trajectory will be added to analyze the single-field pixel distribution of different single agglutinated particles within 10 seconds in the field of view. The peak number information (Fig. 14B) and the single-view pixel distribution inter-wave peak number information (the sum of the numbers indicated by the circles in Fig. 14C represents the single-view pixel distribution inter-wave peak number information of the single-view image in the 4th second, The average of the distances marked by any two adjacent circles in Figure 14C is the information on the average peak distance of the single-view pixel distribution), and the single-view pixel distribution area information under the curve is automatically calculated, and then the module is calculated according to the regression analysis. The reference image data in the reference database categorizes the information on the number of peaks of pixel distribution in a single field of view, the information on the number of peaks between pixels in a single field of view, the area information of pixel distribution in a single field of view, and the information on the average peak distance of pixel distribution in a single field of view. , to determine whether there is a fine agglutination that cannot be detected by the naked eye in the field of view to be tested, and then use the second neural network classifier to train the image pixel distribution data to converge to obtain dual image eigenvalue weight data, and further according to the aforementioned two The double agglutination interpretation result is output from the weighted data of the eigenvalues of the double image.

Furthermore, as shown in Fig. 13B, Fig. 15A, Fig. 15B and Fig. 15C, when the first neural network classifier of the present invention analyzes the three-dimensional hemagglutination image data of the mixed blood sample, the agglutination interpretation result obtained is + 1 or less, but when the 3D hemagglutination image data shows fine agglutination, in order to evaluate the blood agglutination more accurately, the image capture device will further capture the dynamic high-magnification 3D hemagglutination image data of the mixed blood sample that lasts for 15 seconds (No. 15A), and the image processing software of the image capture device will automatically select the field of view where blood cells have fine agglutination in the dynamic high-magnification 3D hemagglutination image data, and increase the parameters such as flow rate and exposure trajectory to analyze the field of view within 10 seconds The information on the number of peaks in the pixel distribution of different polyagglutinated particles (Fig. 15B) and the information on the number of peaks in the pixel distribution in a single field of view (the sum of the numbers indicated by the circles in Fig. 15C represents the single-field image in the 5th second). The information on the number of peaks in the single-view pixel distribution between waves, and the average of the distances marked by any two adjacent circles in Figure 15C is the single-view pixel distribution average peak distance information), and the single-view pixel under the curve is automatically calculated. Distribution area information, and then according to the reference image data in the reference database of the regression analysis algorithm module, the single-view pixel distribution peak number information, the single-view pixel distribution peak number information, the single-view pixel distribution area information, and the single-view image The average peak distance information of the pixel distribution is classified separately to determine whether there are other fine agglutinations that cannot be detected by the naked eye in the field of view to be tested, and then use the second neural network classifier to train the image pixel distribution data to converge to obtain a double image. Eigenvalue weight data, and further output a dual agglutination interpretation result according to the aforementioned dual image eigenvalue weight data.

Please refer to Fig. 16, Fig. 17 and Fig. 18. Fig. 16 shows the receiver operating characteristic curve (ROC) analyzed by the automated blood analysis system of the present invention using single-view pixel distribution area information. ), Figure 17 is a graph showing the receiver operating characteristic curve of the automated blood analysis system of the present invention using the single-field image pixel distribution peak number information of the single-field image to analyze, and Figure 18 is an automated blood analysis system of the present invention. The blood analysis system analyzes the receiver operating characteristic curve based on the average peak distance information of the pixel distribution of the single field of view.

As shown in Fig. 16, when the automated blood analysis system of the present invention is used to perform a cross-experiment between the blood sample of the subject and the blood sample of the donor and analyze the information of the pixel distribution area of the single field of view, the receiver operates The Area Under the Receiver Operating Characteristic curve (AUROC) of the characteristic curve can be as high as 0.993, its evaluation sensitivity (Sensitivity) is 96.67% to 100.00%, and its evaluation specificity (Specificity) is 80.00% to 100.00%. The associated criterion value for analyzing the pixel distribution area information of the single field of view is greater than 9519 to 12738.

As shown in Fig. 17, when the automated blood analysis system of the present invention is used to perform a cross-experiment between the blood sample of the subject and the blood sample of the donor and analyze the information on the peak number between the single-field pixel distributions, it receives The area under the curve of the operator's operating characteristic curve can be as high as 0.962, its evaluation sensitivity is 96.67% to 100.00%, its evaluation specificity is 63.33% to 93.33%, and the relevant standard value of the single-field pixel distribution peak number information for analysis is less than 68 to 75.

Furthermore, as shown in FIG. 18, when the automated blood analysis system of the present invention is used to perform a cross-experiment between the blood sample of the subject and the blood sample of the donor, and analyze the average peak distance information of the pixel distribution of the single field of view. , the area under the curve of the receiver operating characteristic curve can be as high as 0.964, its assessment sensitivity is 86.67% to 93.33%, and its assessment specificity is 90.00% to 96.67%. The relevant normative value is greater than 14.6 to 15.2.

Furthermore, the automated blood analysis system of the present invention is further based on information on the number of peaks of pixel distribution in a single field of view, information on the distribution area of pixels in a single field of view, information on the number of peaks between pixels in a single field of view, and information on the average peak distance of pixel distribution in a single field of view. The presented results are calculated to obtain a decision parameter A (Cut score A) and a decision parameter B (Cut score B) to further evaluate the accuracy of hemagglutination interpretation of the automated blood analysis system of the present invention, wherein the decision parameter A is the following The formula (I) is calculated, and the decision parameter B is calculated by the following formula (II). The decision parameter A = (0.00075 × information on the pixel distribution area of a single field of view) + (2.27 × information on the number of peaks of the pixel distribution in a single field of view) + (14.88 × information on the average peak distance of the pixel distribution in a single field of view)… Equation (I). Decision parameter B = (0.00075 × single-view pixel distribution area information) + (2.27 × single-view pixel distribution peak number information) + (14.88 × single-view pixel distribution average peak distance information) – (4.11 × single-view pixel distribution information) information on the number of peaks in the prime distribution)… Equation (II).

Please refer to Fig. 19 and Fig. 20. Fig. 19 shows the receiver operating characteristic curve of the decision parameter A output by the automated blood analysis system of the present invention, and Fig. 20 depicts the automated blood analysis system of the present invention. The analysis system analyzes the receiver operating characteristic curve graph of the output decision parameter B. As shown in Figure 19, the area under the curve of the receiver operating characteristic curve analyzed with the decision parameter B can be as high as 0.991, its assessment sensitivity is 96.67% to 100.00%, its assessment specificity is 73.33% to 100.00%, and its The relevant standard value is greater than 385 to 394, and as shown in Figure 20, the area under the curve of the receiver operating characteristic curve analyzed with the decision parameter B can be as high as 0.991, and its evaluation sensitivity is 90.00% to 100.00%. The specificity is 80.00% to 100.00%, and the relative standard value is greater than 381 to 390, indicating that the automated blood analysis system and the automated blood analysis method of the present invention have excellent hemagglutination interpretation accuracy.

In addition, when the automated blood analysis system and automated blood analysis method of the present invention completes the analysis, the blood bank can further output information on the number of peaks in the pixel distribution of the single field of view, the relevant standard value of the pixel distribution area information in the single field of view, and the pixel distribution in the single field of view. The relevant standard value of the information on the number of peaks between waves, the relevant standard value of the information on the average peak distance of the single-field pixel distribution, and the value of the aforementioned decision parameter A or decision parameter B are for further analysis and judgment by the medical examiner.

Please refer to Fig. 21 and Fig. 22. Fig. 21 shows the pixel distribution area information of a single visual field without agglutination, the average peak distance information of the single visual field pixel distribution, and the decision parameter output by the automated blood analysis system of the present invention. The three-dimensional correlation distribution diagram of B, Fig. 22 shows the information of the single-field pixel distribution area information of agglutination, the average peak distance information of the single-field pixel distribution and the three-dimensional determination parameter B output by the automated blood analysis system of the present invention. Correlation distribution map. As shown in Fig. 21 and Fig. 22, the blood cross-test group without agglutination and the blood cross-test group with agglutination can be clearly distinguished (please refer to the coordinate distribution range of Fig. 21 and Fig. 22), It further shows that the automated blood analysis system and the automated blood analysis method of the present invention have excellent hemagglutination interpretation accuracy, and have application potential in the relevant market.

As shown by the above results, the automated blood analysis system and the automated blood analysis method of the present invention are used for cross-experiment of blood samples of subjects and blood samples of donors, and the accuracy, sensitivity and specificity of the results are excellent. , it shows that the automated blood analysis system and the automated blood analysis method of the present invention can accurately interpret blood agglutination valence and hemagglutination pattern according to the three-dimensional hemagglutination image data of the mixed blood samples after the reaction in the cross-experiment, so as to achieve blood processing. It can greatly reduce the error of cross-test results caused by the subjective interpretation habits of different blood bank medical examiners and improve the detection efficiency.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention The scope shall be determined by the scope of the appended patent application.

100,300: Automated blood analysis system 110,310: Loading platform 111,311: Sample loading device 112,312: Blood sample loading device 120,320: Testbed 121,321: Reagent reaction device 122,322: Image capture device 123: light source 124: Camera Module 130, 130a, 130b, 330: Processors 131, 131a, 131b: Blood Cross-Test Evaluation Procedures 140, 140a, 140b: The first image preprocessing module 150, 150a, 150b: The first training module 160, 160a, 160b: The first judgment module 170a: Second image preprocessing module 170b: Hybrid Assessment Module 180a: Second training module 190a: Second Judgment Module 200, 200a, 200b: Automated Blood Analysis Methods Steps 301: User interface 302, 304: Mechanical gripper 303: Puncture automatic pipette 313: Centrifuge 314: Sample placement platform 323: Automatic centrifuge tube shaking device

FIG. 1 is a schematic diagram showing the structure of an automated blood analysis system according to an embodiment of the present invention; FIG. 2 is a schematic structural diagram of the processor of the automated blood analysis system of FIG. 1; FIG. 3 is a schematic structural diagram of another processor of the automated blood analysis system of FIG. 1; FIG. 4 is a schematic structural diagram of another processor of the automated blood analysis system of FIG. 1; FIG. 5 is a flowchart showing the steps of an automated blood analysis method according to another embodiment of the present invention; FIG. 6 is a flowchart showing the steps of an automated blood analysis method according to another embodiment of the present invention; Fig. 7 is a flow chart showing the steps of the double analysis and determination steps of the automated blood analysis method of Fig. 6; FIG. 8 is a flow chart showing the steps of an automated blood analysis method according to still another embodiment of the present invention; FIG. 9 is a schematic diagram illustrating an automated blood analysis system according to an example of an embodiment of the present invention; Figure 10A is a static three-dimensional hemagglutination image data of the present invention; Figure 10B is a dynamic three-dimensional hemagglutination image data of the present invention; Figure 11 is a high-magnification static three-dimensional hemagglutination image data of the present invention; Fig. 12A is an analysis of the peak information of the pixel distribution of the single-field pixel distribution of the single agglutinated particle in the high-magnification static three-dimensional hemagglutination image data of Fig. 11; Figure 12B is an analysis of the single-field pixel distribution area information of polyagglutinated particles in the high-magnification static three-dimensional hemagglutination image data of Figure 11; Figure 13A is a high-magnification dynamic three-dimensional hemagglutination image data of the present invention without agglutination; Figure 13B is a high-magnification dynamic three-dimensional hemagglutination image data of an agglutination reaction of the present invention; Fig. 14A is a graph of image changes within 10 seconds of the single-field image of a single agglutinated particle in the high-magnification dynamic three-dimensional hemagglutination image data of Fig. 13A; Fig. 14B is an analysis diagram of peak information of single-view pixel distribution of a single-view image within 8 seconds corresponding to the image change chart of Fig. 14A; Fig. 14C is an enlarged view of the single-view pixel distribution peak information analysis graph of the single-view image at the 4th second in Fig. 14B; Fig. 15A is a graph of the image change within 10 seconds of the single-field image of polyagglutinated particles in the high-magnification dynamic three-dimensional hemagglutination image data of Fig. 13B; Fig. 15B is an analysis diagram of peak information of single-view pixel distribution of a single-view image within 8 seconds corresponding to the image change chart of Fig. 15A; Fig. 15C is an enlarged view of the single-view pixel distribution peak information analysis graph of the single-view image at the 5th second in Fig. 15B; FIG. 16 is a receiver operating characteristic curve (ROC) diagram of the automated blood analysis system of the present invention that analyzes the pixel distribution area information in a single field of view; Fig. 17 is a graph showing the receiver operating characteristic curve of the automated blood analysis system of the present invention, which analyzes the peak number information of the single-view pixel distribution of the single-view image; Fig. 18 is a graph showing the receiver operating characteristic curve of the automated blood analysis system of the present invention that analyzes the average peak distance information of the pixel distribution in a single field of view; Fig. 19 is a graph showing the receiver operating characteristic curve of the decision parameter A (Cut score A) output by the automated blood analysis system of the present invention; Fig. 20 is a graph showing the receiver operating characteristic curve of the decision parameter B (Cut score B) output by the automated blood analysis system of the present invention; FIG. 21 is a three-dimensional correlation distribution diagram of the single-field pixel distribution area information without agglutination, the single-field pixel distribution average peak distance information, and the decision parameter B output by the automated blood analysis system of the present invention; and FIG. 22 is a three-dimensional correlation distribution diagram of the single-field pixel distribution area information of agglutination, the single-field pixel distribution average peak distance information and the decision parameter B output by the automated blood analysis system of the present invention.

100: Automated Blood Analysis System 110: Loading platform 111: Sample loading device 112: Blood sample loading device 120: Testbed 121: Reagent reaction device 122: Image capture device 123: light source 124: Camera Module 130: Processor

Claims (20)

An automated blood analysis system, comprising: a sample loading platform, comprising: a sample loading device for pre-processing a blood sample of a subject to obtain a test blood sample; a blood sample a sample device, which is adjacent to the sample loading device, wherein the blood sample loading device is used for pre-processing a blood sample of a blood donor to obtain a target blood sample; and a sample loading mechanical gripper, The sample loading device and the blood sample loading device are adjacent to the sample loading device, wherein the sample loading mechanical gripper is used to move the blood sample of the subject, the blood sample of the subject, the blood sample of the blood donor and the the target blood sample; a test platform adjacent to the sample loading platform, wherein the test platform includes: a test drug reaction device for mixing the test blood sample and the target blood sample to obtain a mixed blood sample, and perform a blood cross test on the mixed blood sample to obtain a mixed blood sample after reaction; a test mechanical gripper is adjacent to the reagent reaction device, wherein the test mechanical gripper is used to move the mixed blood sample after the reaction a blood sample; and an image capture device adjacent to the reagent reaction device, wherein the image capture device is used to capture a three-dimensional hemagglutination image data of the mixed blood sample after the reaction; a centrifuge, adjacent to the reaction device the sample loading platform and the test platform; and a processor telecommunicationly connected to the image capture device, wherein the processor includes a blood cross test evaluation program, and the blood cross test evaluation program It includes: a first image preprocessing module for adjusting an image size and a background value of the three-dimensional hemagglutination image data to obtain a processed three-dimensional hemagglutination image data; a first training module for using A first neural network classifier trains the processed 3D hemagglutination image data to converge to obtain an image feature value; and a first judgment module using the first neural network classifier according to the image feature The value outputs an agglutination judgment result; wherein, the agglutination judgment result includes a blood agglutination valence judgment result and a blood cell agglutination type judgment result. The automated blood analysis system of claim 1, wherein the mixed blood sample is placed in a centrifuge tube to perform the blood crossover test, and the image capture device captures the reacted mixed blood sample in the centrifuge tube The three-dimensional hemagglutination imaging data. The automated blood analysis system of claim 1, wherein the image capture device comprises: a light source for projecting a scattered light or direct light to the mixed blood sample after the reaction; and a camera module comprising a low magnification An imaging module and a high-magnification imaging module. The automated blood analysis system of claim 1, wherein the three-dimensional hemagglutination image data includes at least one dynamic three-dimensional hemagglutination image data and at least one static three-dimensional hemagglutination image data. The automated blood analysis system of claim 1, wherein the first neural network classifier is a support vector machine (SVM) neural network classifier. The automated blood analysis system according to claim 1, wherein the image capture device further captures a high-magnification three-dimensional hemagglutination image data of the mixed blood sample after the reaction, and the blood cross test evaluation program further comprises: a second an image preprocessing module for preprocessing the high-magnification three-dimensional hemagglutination image data to obtain at least one image pixel distribution data; a second training module for analyzing the image pixels using a regression analysis algorithm module After distributing the data, a second neural network classifier is used to train the image pixel distribution data to converge to obtain a dual image feature value weight data; and a second judgment module, which uses the second neural network The path classifier outputs a double agglutination interpretation result according to the double image feature value weight data; wherein, the double agglutination interpretation result includes a double blood agglutination valence judgment result and a double blood agglutination type judgment result. The automated blood analysis system as claimed in claim 6, which The high-magnification three-dimensional hemagglutination image data includes at least one dynamic high-magnification three-dimensional hemagglutination image data and at least one static high-magnification three-dimensional hemagglutination image data, and the image pixel distribution data includes a single field of view pixel distribution peak number information, a single Information on the number of peaks between the pixels of the visual field distribution, information on the area of the pixel distribution in a single visual field, and information on the average peak distance of the pixel distribution in a single visual field. The automated blood analysis system according to claim 6, wherein the regression analysis calculation module is a Logistic regression calculation module. The automated blood analysis system of claim 6, wherein the second neural network classifier is a support vector machine neural network classifier. The automated blood analysis system as claimed in claim 1, wherein the blood cross-test evaluation program further comprises: a combined blood evaluation module for training a copy of the agglutination interpretation result with a third neural network classifier until convergence, to output a blood compatibility evaluation result. The automated blood analysis system of claim 10, wherein the third neural network classifier is a support vector machine neural network classifier. An automated blood analysis method, comprising the following steps: providing an automated blood analysis system as described in claim 1; providing the blood sample of the subject; providing the blood sample of the blood donor; performing a sample preprocessing step , which uses the sample loading platform to perform pre-loading pre-processing on the blood sample of the subject to obtain the subject blood sample, and uses the sample-loading platform to perform pre-loading processing on the blood sample of the blood donor , to obtain the target blood sample; perform a test drug reaction step, which is to mix the test blood sample and the target blood sample with the test platform to obtain the mixed blood sample, and perform the blood crossover on the mixed blood sample testing to obtain the mixed blood sample after the reaction; performing an image capture step, wherein the image capture device captures the three-dimensional hemagglutination image data of the mixed blood sample after the reaction; performing an image preprocessing step, which Adjusting the image size and the background value of the three-dimensional hemagglutination image data by the first image preprocessing module to obtain the processed three-dimensional hemagglutination image data; and performing an analysis and judgment step, which uses the first The training module uses the first neural network classifier to train the processed three-dimensional hemagglutination image data to convergence, so as to obtain the image feature value, and uses the first judgment module to use the first neural network classifier according to the first neural network classifier. The image feature value outputs the agglutination interpretation result; wherein, the agglutination interpretation result includes the blood agglutination valence interpretation result and the blood cell agglutination type interpretation result. The automated blood analysis method of claim 12, wherein the mixed blood sample is placed in a centrifuge tube to perform the blood crossover test, and the three-dimensional hemagglutination image data is the result of the reaction mixed blood sample in the centrifuge tube The three-dimensional hemagglutination image data. The automated blood analysis method of claim 12, wherein the three-dimensional hemagglutination image data includes at least one dynamic three-dimensional hemagglutination image data and at least one static three-dimensional hemagglutination image data. The automated blood analysis method of claim 12, wherein the first neural network classifier is a support vector machine neural network classifier. The automated blood analysis method according to claim 12, further comprising: performing a double analysis and judging step, the double analysis judging step includes the following steps: performing a double image capturing step, which is after capturing the reaction A high-magnification three-dimensional hemagglutination image data of mixed blood samples; performing a double image preprocessing step, which is to preprocess the high-magnification three-dimensional hemagglutination image data to obtain at least one image pixel distribution data; The analysis step is to use a regression analysis algorithm module to analyze the image pixel distribution data, and then use a second neural network to classify training the image pixel distribution data to convergence to obtain a dual image feature value weight data; and performing a dual determination step, which utilizes the second neural network classifier to determine the dual image feature value weight data according to the dual image feature value weight data A double agglutination interpretation result is output; wherein, the double agglutination interpretation result includes a double blood agglutination valence interpretation result and a double blood agglutination type interpretation result. The automated blood analysis method of claim 16, wherein the high-magnification three-dimensional hemagglutination image data includes at least one dynamic high-magnification three-dimensional hemagglutination image data and at least one static high-magnification three-dimensional hemagglutination image data, and the image pixel distribution data includes Information on the number of peaks of pixel distribution in a single field of view, information on the number of peaks between waves in pixel distribution in a single field of view, information on the distribution area of pixels in a single field of view, and information on an average peak distance of pixel distribution in a single field of view. The automated blood analysis method of claim 16, wherein the regression analysis algorithm module is a Logis regression algorithm module, and the second neural network classifier is a support vector machine neural network classifier. The automated blood analysis method as claimed in claim 12, further comprising: performing a combined blood evaluation step of training a copy of the agglutination interpretation result to convergence with a third neural network classifier to output a blood Consistency assessment results. The automated blood analysis method of claim 19, wherein the third neural network classifier is a support vector machine neural network classifier.

Images