ES2898148A1 - SYSTEM CONFIGURED TO DETECT A CANCERINE LESION LOCATED IN A PORTION OF HUMAN TISSUE AND METHOD - Google Patents

Abstract

System configured to detect a carcinogenic lesion located in a portion of human tissue and method. In a first aspect, a system configured to detect a carcinogenic lesion located in a portion of human tissue. The system is characterized by the fact that the system comprises: an image acquisition module, an information extraction module and a cancer discrimination module. The system also comprises a robot arm, where the image acquisition module is coupled to the robot arm, characterized by the robot arm comprises: a distance sensor configured to measure a distance between the distance sensor and a region of The portion of the human tissue and a control module of the robot arm configured to receive the distance measured by the Robot arm distance sensor and reposition the robot arm. A method configured is also described for detecting a cancerous injury located in a portion of human tissue. (Machine-translation by Google Translate, not legally binding)

Description

DESCRIPTION

SYSTEM CONFIGURED TO DETECT A CANCER LESION

LOCATED IN A PORTION OF HUMAN TISSUE AND METHOD

The object of the present application is related to a system configured to detect a cancerous lesion located in a portion of human tissue. The present application is also related to a configured method for detecting a cancerous lesion located in a portion of human tissue.

PRIOR STATE OF THE ART

Skin cancer currently accounts for one in three cancer diagnoses. The incidence of "non-melanoma" and "melanoma " skin cancers has increased in recent decades. Currently, between 2 and 3 million "non-melanoma" skin cancers and 132,000 "melanoma" skin cancers are diagnosed worldwide each year.

Melanoma, the deadliest skin cancer, forms first in the upper layers of the skin. When it metastasizes, melanoma cancer cells enter blood vessels and spread throughout the body. Melanoma is highly deadly if it is not detected in its early stages, however, it can be cured, with a survival rate of almost 100%, if it is detected and eliminated early in its development.

It is known that, during a conventional examination, dermatologists visually examine the skin for lesions or moles that fit certain predefined criteria related to a possible cancerous lesion. If an area is suspicious, the doctor may perform a biopsy, sending the tissue to a laboratory for diagnosis. Although effective, this detection method is slow, invasive, and does not provide an immediate definitive diagnosis of the suspicious lesion. In particular, performing a biopsy is an invasive, painful and possibly unnecessary procedure in cases where the lesion turns out to be benign.

In addition, this technique can lead to false positives that can lead to unnecessary biopsies and extra costs. Additionally, early detection is very difficult to performed with this technique, as developing skin cancer lesions are usually not visible.

Non-invasive procedures for detecting skin lesions are also known. These non-invasive procedures have the benefit that they can cause minimal or no pain to the patient, avoid scarring, recovery is usually immediate, and post-surgical complications can be avoided. However, for skin lesions, the diagnostic accuracy of these noninvasive techniques may be questionable.

Therefore, it is an object of the present invention to provide systems and methods for detecting a cancerous lesion located in a portion of human tissue that solve some of the aforementioned problems.

EXPLANATION OF THE INVENTION

In a first aspect, a system configured to detect a cancerous lesion located in a portion of human tissue is provided. The system is characterized in that the system comprises: an image acquisition module configured to acquire one or more images of a portion of human tissue, where the image acquisition module comprises one or more hyperspectral cameras configured to acquire a hyperspectral image of the human tissue portion, an information extraction module configured to extract an electromagnetic spectrum for each pixel comprised in each image acquired by the image acquisition module, where the electromagnetic spectrum includes a plurality of wavelengths and a cancer lesion discrimination module configured to receive the electromagnetic spectrum for each pixel extracted by the information extraction module and apply one or more detection algorithms to assess the condition of the tissue. The system also comprises a robot arm, where the image acquisition module is coupled to the robot arm, characterized in that the robot arm comprises: a distance sensor configured to measure a distance between the distance sensor and a region of the portion of human tissue and a control module of the robot arm configured to receive the distance measured by the distance sensor of the robot arm and reposition the robot arm such that the distance received by the control module remains within a range of distances relative to any region of the human tissue portion during a displacement of the robot arm.

According to this first aspect, an image acquisition module is provided comprising hyperspectral cameras configured to acquire one or more hyperspectral images of a portion of human tissue. Hyperspectral cameras allow information about a patient to be obtained that is not easily identifiable with the naked eye. For example, even if the suspected cancerous lesion may be visually identifiable, the actual extent of the lesion or what type of condition the lesion exhibits may not be discernible on visual inspection, for example, it may not be easy to identify whether the lesion is benign. or cancerous. On the contrary, the use of hyperspectral cameras is a powerful tool that allows easy and rapid identification of cancerous lesions.

Specifically, hyperspectral cameras make it possible to acquire one or more images of a portion of human tissue. The information extraction module allows to extract multiple spectral regions from each one of the pixels of the images acquired by the hyperspectral cameras. Each particular region of a patient's portion of human tissue to be analyzed has a unique spectral signature that spans across multiple bands of the electromagnetic spectrum. This spectral signature contains information about the corresponding region of the patient.

For example, if the patient has a cancerous skin lesion, that lesion may have a different color, density, and/or composition than the patient's normal skin, resulting in the lesion having an electromagnetic spectrum (including a plurality of wavelengths), different from normal skin. While these differences may be difficult to detect visually, the differences may become apparent through analysis of the electromagnetic spectrum of the hyperspectral image pixels extracted by the information extraction module. This analysis is carried out using the cancer lesion discrimination module, applying specific algorithms for the discrimination and detection of cancer lesions. In this way, a possible cancerous lesion located in a portion of human tissue can be more easily identified, characterized and ultimately treated than would be possible with conventional visual inspection and/or biopsy.

Additionally, the image acquisition module (and, therefore, the hyperspectral cameras included in this module) can be located on a robot arm. In this way, the positioning of the hyperspectral cameras relative to the patient can be improved and facilitated. Specifically, inspection can be facilitated of different parts of a cancerous lesion, located in a portion of the patient's human tissue, without the need to move the patient.

Furthermore, the robot is provided with a distance sensor configured to measure a distance between the distance sensor and a region of the patient's human tissue portion and a control module of the robot arm configured to receive the distance measured by the distance sensor. distance of the robot arm and reposition the robot arm so that the distance received by the control module is maintained in a range of distances in relation to any region of the portion of human tissue, during a displacement of the robot arm along the piece of human tissue. In this way, the hyperspectral cameras can be kept at a distance within a range of distances relative to a region of the portion of the human being to be analyzed (during, for example, the movement of the camera, between two regions of the portion of human tissue, using the robot arm) and, therefore, the hyperspectral cameras can be kept correctly focused during the movement of the cameras (using the robot) along the portion of tissue to be analyzed.

In examples, the distance received by the robot arm is maintained in a range of distances between 14 centimeters and 16 centimeters in relation to any region of the portion of human tissue during a movement of the robot arm, specifically 15 centimeters.

According to an example of the system, the robot arm is an industrial robot, where the industrial robot comprises three or more axes with at least two degrees of freedom (for each joint that forms part of the robot arm).

In one example of the system, the system further comprises one or more light sources coupled to the robot arm and configured to illuminate the portion of human tissue for the hyperspectral cameras, where the light sources are configured to emit electromagnetic radiation over a range of wavelengths. of wavelength between 380 nm and 2500 nm, optionally the light sources are configured to emit electromagnetic radiation in a range of wavelengths between 380 nm and 1000 nm or in a range of wavelengths between 900 and 1700 nanometers.

In another example, the robot arm comprises a tool (or claw) configured to hold the hyperspectral cameras and light sources.

In examples, the tool comprises a housing mounted at one end of the robot arm, where the housing comprises an inner surface, two side walls, and a space formed inside the housing, where hyperspectral cameras, in use, are located in space. formed inside the housing attached to the inner surface.

In a further example, the housing further comprises a bracket with a U-shaped cross-section and configured to support the distance sensor, where the bracket for supporting the distance sensor is located in the space formed inside the housing and attached to the inner surface.

In examples, the sidewalls of the housing are configured to support the lights, where the sidewalls are configured to pivot relative to the interior surface of the housing, where, in use, the sidewalls are positioned between 40 and 50 degrees relative to the plane defined by the interior surface of the housing, optionally they are located at 45 degrees.

In other examples, the hyperspectral cameras of the image acquisition module are configured to operate in a range of wavelengths of the electromagnetic spectrum between 400 nm and 2500 nm, optionally the image acquisition module comprises two hyperspectral cameras, where the first camera hyperspectral is configured to operate in a range of wavelengths of the electromagnetic spectrum between 400 nm and 1000 nm and the second hyperspectral camera is configured to operate in a range of wavelengths of the electromagnetic spectrum between 1000 nm and 1700 nm.

In examples, the system further comprises an electromagnetic spectrum correction module configured to correct the extracted electromagnetic spectrum for each pixel comprised in each image acquired by the image acquisition module.

In other examples, the electromagnetic spectrum correction module is adapted to:

receive a first image acquired by the image acquisition module when the hyperspectral cameras do not receive light,

obtain a measurement (Rn) related to the electromagnetic spectrum for a pixel comprised in the first image, where the measurement Rn comprises a plurality of wavelengths,

receive a second image acquired by the image acquisition module of a reference material illuminated with electromagnetic radiation in a range of wavelengths between 400 nm and 2500 nm,

obtain a measure (Rb) related to the electromagnetic spectrum for a pixel comprised in the second image, where the measure Rb comprises a plurality of wavelengths,

obtain, for each measurement of the electromagnetic spectrum (M) of each pixel included in each image acquired by the image acquisition module, a measurement of the corrected electromagnetic spectrum (Rc) following the following equation:

R ^c = M-R ⁿ /R ^b -R ⁿ

In examples, the system further comprises an electromagnetic spectrum pre-processing module configured to pre-process the extracted electromagnetic spectrum for each pixel comprised in each image acquired by the image acquisition module, where the pre-processing module is configured to select one of the following algorithms for pre-processing the electromagnetic spectrum: centering and scaling by unit variance, centering on mean, Savitzky-Golay smoothing, Savitzky-Golay first derivative, Savitzky-Golay second derivative, Normal variance scaling Standard Normal Variate scaling (SNV), Multiplicative scatter/signal correction (MSC).

In some examples, the cancer lesion discrimination module is configured to select one or more of the following cancer lesion detection algorithms: decision trees, support vector machines (SVM), neural networks (NN ), convolutional neural networks (CNN), random forests (Random Forest) or projection of latent structures with discriminant analysis (PLSDA).

In a second aspect, a method is provided for detecting a cancerous lesion located in a portion of human tissue using a system according to the first aspect. The method characterized by the fact that it comprises: placing the distance sensor of the robot arm to a predetermined distance within a range of distances relative to a region of a portion of human tissue, moving the robot arm such that the distance of the distance sensor of the robot arm relative to any region of the portion of the human tissue human tissue remains within the range of distances, receive one or more hyperspectral images of the portion of human tissue, extract the electromagnetic spectrum for each pixel comprised in each image acquired by the image acquisition module, where the electromagnetic spectrum includes a plurality of wavelengths. The method further comprises for each extracted electromagnetic spectrum, applying one or more detection algorithms to evaluate a condition related to the presence of a cancerous tissue lesion so that a cancerous lesion is detected.

According to this second aspect, the robot arm (and, therefore, the image acquisition module, with the hyperspectral cameras, attached to the robot arm), is located at a predetermined distance within a range of distances in relation to a region of the human tissue portion. During the movement of the robot arm, the distance in relation to any region of the human tissue portion to be analyzed is kept within the range of distances. In this way, the hyperspectral cameras do not go out of focus and clear images of the skin region of the human being to be analyzed can be captured.

In examples, the method comprises selecting one or more wavelengths for each extracted electromagnetic spectrum depending on the type of cancerous lesion to be evaluated.

BRIEF DESCRIPTION OF THE DRAWINGS

Below, particular embodiments of the present invention will be described by way of non-limiting example, with reference to the attached drawings, in which:

Figure 1 shows a block diagram of a system configured to detect a cancerous lesion located in a portion of human tissue;

Figure 2a shows an example of a system configured to detect a cancerous lesion located in a portion of human tissue;

Figure 2b shows an example of a robot arm and a tool located at one end of the robot arm, where both elements form part of the system of Figure 2a;

Figure 2c shows an example of a detail of the tool to support lighting elements, hyperspectral cameras and a distance sensor, where the tool forms part of the system of Figure 2a;

Figure 3 shows an example of a flow chart of a method for detecting a cancerous lesion located in a piece of human tissue.

DETAILED EXPOSITION OF METHODS OF REALIZATION

Throughout this description and claims, the term "region of the human tissue patch" refers to a surface of the human tissue patch eg a point relative to which the distance sensor measures the distance. between the distance sensor of the robot arm and the portion of human tissue. The surface of this "human tissue portion region" may vary eg between 1 mm2 and 10 cm2.

Figure 1 is a block diagram of a system configured to detect a cancerous lesion located in a portion of human tissue. System 110 comprises an image acquisition module 115, an information extraction module 125, an electromagnetic spectrum correction module 126, an electromagnetic spectrum pre-processing module 127 and a cancer lesion discrimination module 130.

Image acquisition module 115 comprises one or more hyperspectral cameras 105 configured to acquire a hyperspectral image of a portion of human tissue. Hyperspectral cameras can be eg fixed optics. Hyperspectral cameras 105 may be eg connected to image acquisition module 115. Hyperspectral cameras 105 may take and store one or more hyperspectral images of a portion of human tissue that will subsequently be processed by system 110. In For examples, hyperspectral cameras 105 may be part of image acquisition module 115 or may be connected externally to image acquisition module 115. The external connection may be wired or wireless.

In order to achieve a certain level of reproducibility and in order to obtain adequate hyperspectral images, the hyperspectral cameras 105 and/or the imaging module Image acquisition 115 can be parameterized according to the requirements of a specific application eg a specific type of skin cancer lesion. Additionally, each of the hyperspectral images acquired by the camera can contain all kinds of metadata that can be useful for the correct analysis of the image. For example, patient characteristics, clinical information of the patient, time and date of image acquisition, patient age, weight, etc... This information can be used e.g. by the prediction algorithms in the phase of discrimination of cancerous lesions.

In examples, the system 110 may be provided with eg two hyperspectral cameras 105 eg a camera FX10 and a camera FX17 both from the company Specim. Hyperspectral cameras can be configured to operate between 400nm and 2500nm. In examples, a first hyperspectral camera may be configured to operate in a range of wavelengths of the electromagnetic spectrum between 400nm and 1000nm (448 spectral bands). The second hyperspectral camera can be configured to operate in a range of wavelengths of the electromagnetic spectrum between 1000nm and 1700nm (224 wavelengths). In this way, the portion of human tissue to be analyzed can be characterized in a wide spectral range in order to obtain as much information as possible about the lesion.

The image acquisition module 115 may also comprise an illumination subsystem 106. The illumination subsystem 106 may include halogen light sources, eg, four 35 Watt, 12 volt lamps. The lamps can be controlled by a voltage regulator that facilitates their switching on and off, as well as the regulation of light intensity. In examples, the light intensity can be regulated depending on the type of skin lesion.

Halogen lights produce intense broadband light that can be considered a close replica of the electromagnetic spectrum associated with daylight. Other suitable light sources can be eg a xenon lamp and/or a light emitting diode or array of diodes. Other types of light sources may also be suitable. Either way, the lighting subsystem 106 is configured to emit electromagnetic radiation in a wavelength range between 380nm and 2500nm. In examples, halogen light sources may be configured to emit electromagnetic radiation in a range of wavelengths between 380nm and 750nm. nm (visible light) or in a range of wavelengths between 0.7 micrometers and 2500 nm (infrared light).

Continuing with the example, the lighting subsystem 106 is configured to irradiate the portion of human tissue to be analyzed with light that is intense enough to allow the hyperspectral cameras 105 to obtain a quality electromagnetic spectrum of the portion of human tissue to be analyzed, that is, that is, an electromagnetic spectrum with an adequate signal/noise ratio to be able to obtain medical information about the portion of human tissue. However, in some instances, ambient light, for example fluorescent, halogen, or incandescent lighting in the room, or even sunlight, may be a suitable light source. In this way, the lighting subsystem 106 might not be activated or the system might not even include the lighting system 106.

The light from illumination subsystem 106 is configured to interact with a plurality of areas or regions within the portion of human tissue to be analyzed. The interaction between light and each area depends on the physiological structure and characteristics of that area. The interaction between the light and each irradiated area of the patient imparts a spectral signature on the light obtained from that area. This spectral signature can be used to obtain medical information about that area. Specifically, different areas interact differently with light depending on the presence of, say, a medical condition in the region. For example, fat, skin, blood, and tissue all interact with various wavelengths of light differently from each other. Similarly, a given type of cancerous lesion interacts with various wavelengths of the electromagnetic spectrum differently from normal skin, noncancerous lesions, and other types of cancerous lesions. Therefore, the light obtained from each irradiated area of the subject has a spectral signature based on the characteristics of the region, which spectral signature contains medical information about that region.

The information extraction module 125 allows the extraction of information for each pixel for each of the hyperspectral images previously captured with the image acquisition module. This information may contain an electromagnetic spectrum (for each pixel of the image provided by the hyperspectral cameras), where the electromagnetic spectrum comprises a plurality of wavelengths.

In some examples, the system may comprise an electromagnetic spectrum correction module 126. The electromagnetic spectrum correction module 126 is configured to correct the electromagnetic spectrum extracted for each pixel by the information extraction module 125. Specifically, each one of the electromagnetic spectra may be corrected relative to two references. The first reference is the darkness reference (Rn), that is, the spectrum generated by one or more hyperspectral cameras 105 when no signal enters them i.e. a signal that refers to the internal electronic noise of the cameras. In particular, the module 126 is configured to receive a first image acquired by the image acquisition module 115 when the hyperspectral cameras do not receive light and obtain a measurement (Rn) related to the electromagnetic spectrum for a pixel included in the first image, where the measurement Rn comprises a plurality of wavelengths,

The second reference is the white reference (Rb). This reference is related to a reference material that has a homogeneous response to electromagnetic radiation and that, moreover, is as stable as possible, in relation to said response, both spatially and temporally. This will make it possible to compensate for differences in light homogeneity, both spatial (that the light is not exactly the same along the measurement line) and temporal (light variations over time due to power supply variations or lamp wear, for example).

In particular, the electromagnetic spectrum correction module 126 is adapted to: receive a second image acquired by the aforementioned reference material image acquisition module, where the material is illuminated with electromagnetic radiation in a range of wavelengths between 400 nm and 2500 nm, and obtain a measurement (Rb) related to the electromagnetic spectrum for a pixel comprised in the second image, where the measurement Rb comprises a plurality of wavelengths,

Using these references (Rn, Rb), the electromagnetic spectrum correction module 126 is adapted to obtain, for each measurement of the electromagnetic spectrum (M) of each pixel comprised in each image acquired by the image acquisition module, a measurement of the corrected electromagnetic spectrum (Rc) following the following equation:

R ^c = M-R ⁿ /R ^b -R ⁿ

In examples, the system may also comprise an electromagnetic spectrum pre-processing module 127. The electromagnetic spectrum pre-processing module 127 is configured to pre-process the extracted electromagnetic spectrum for each pixel comprised in each image acquired by the module. image acquisition 115. This pre-processing includes the most appropriate pre-treatments to maximize the signal-to-noise ratio of each extracted electromagnetic spectrum for each portion of human tissue selected to be analyzed.

Specifically, the pretreatments are signal processing algorithms that are applied to each of the electromagnetic spectra extracted by the information extraction module 125, with the aim of correcting some of the problems that may arise during their acquisition, for example, noise. , drifts of the instruments, multiplicative effects, etc. The algorithms used may be related eg to electromagnetic spectrum pretreatments used in chemometrics.

Examples of algorithms to pre-process the electromagnetic spectrum can be, for example: Centered and scaled by unit variance, Centered in mean, Savitzky-Golay smoothing, First derivative of Savitzky-Golay, Second derivative of Savitzky-Golay, Standard Normal Variate scaling (SNV) or Multiplicative scatter/signal correction (MSC). In some examples, two or more of these algorithms can be combined to pre-process the electromagnetic spectrum.

Finally, the electromagnetic spectrum for each pixel extracted by the information extraction module 125 (which may have previously been corrected and/or pre-processed by models 126 and 127) can serve as input for the cancer lesion discrimination module 130. A wide variety of algorithms can be used to assess tissue condition. It may also be possible to use two or more lesion discrimination algorithms in parallel or one after the other. Examples of algorithms can be:

- Decision trees: One type of decision rule(s) that can be constructed using spectral data is a decision tree. Decision trees are used to classify lesions using spectral data sets. The goal of a decision tree is to induce a classifier (a tree) from real-world sample data. This tree can be used to classify examples that have not been used to derive the decision tree. This decision tree can be derived from eg training data. The training data Exemplars contain spectral data for a plurality of injuries (the training injuries), each of which has a particular medical condition.

- Support Vector Machines (SVM): The Support Vector Machine (SVM) technique is used to classify lesions using predetermined values of “observables”. “Observables” may include, but are not limited to, specific pixel values, patterns of values of specific groups of pixels, values of specific measured wavelengths, or any other form of observable data that is directly present in the spectral data and/or that can be derived from the spectral data.

- Neural Networks (NN): A basic approach to using neural networks is to start with an untrained network, provide a training pattern to the input layer, pass signals through the network, and determine the output at the input layer. exit. These outputs are compared to default values; any difference corresponds to an error. This error or criterion function is a scalar function of the weights and is minimized when the network outputs match the desired outputs. Therefore, the weights are adjusted to reduce this measure of error. In some examples, the neural networks used may be Convolutional Neural Networks (CNN).

Some other algorithms that can be used to discern skin conditions can be: “ random forest” type algorithms or algorithms related to projection of latent structures with discriminant analysis (PLSDA). These algorithms classify electromagnetic spectra in such a way as to distinguish portions of a patient's skin as normal or as portions of a patient's skin containing a particular medical condition.

In relation to the technique of "latent structures with discriminant analysis ( PLSDA)", this technique seeks the best combination of wavelengths to establish classes where to classify the electromagnetic spectra extracted from the portion of the skin to be analyzed, with the least possible error. in the classification. This technique is also configured to generate coefficients that, applied to the electromagnetic spectra extracted for each pixel of each of the images acquired by the hyperspectral cameras (as explained previously), allows each of them to be classified in each of the previously established classes. For each electromagnetic spectrum (with a plurality of wavelengths), the membership of one or another class will be given by the values of its quality indicators and its proximity to the values of the centroid established for each class.

To know the accuracy of the generated classification models, two methods can be used, one statistical and the other visual. Two statistical variables can be used in the mathematical method: sensitivity and specificity. For example, a classification in 2 categories. If we have human tissue samples belonging to 2 classes (Positive in relation to a cancerous lesion and Negative in relation to a cancerous lesion) and a classification made, 4 possible results can be given:

Sensitivity can represent how well human tissue samples from a class rank in that class. In examples, it would be defined as:

Sensitivity = True Positives / (True Positives False Negatives)

Specificity can represent how well human tissue samples that do not belong to a class are classified in the other classes that are not that class. Namely:

Specificity = True Negatives / (True Negatives False Positives)

Therefore, a class that has a high sensitivity means that most of its samples are classified as corresponding to that class. If, in addition, that class has a high specificity, it means that very few of the samples of the other classes are incorrectly classified as belonging to that class.

In some examples, it is possible to select the wavelengths that are most related to the parameter that identifies a lesion and discard those that are not related to the parameter that identifies a lesion. In this way, more robust models can be obtained and generally with a lower number of latent structures, which also results in their reliability.

Following the example, the following tables show classes for the discrimination of tissue affected by carcinoma vs healthy tissue, where the classes have been obtained through algorithms related to projection of latent structures with discriminant analysis (PLSDA). Class 1 represents affected tissue and Class 2 healthy tissue.

In relation to the first table, the term "None" refers to a type of statistical pre-treatment applied to the data provided by the hyperspectral cameras. In the specific example of this first table, no pretreatment has been applied. Instead, the data discrimination method is applied directly to the data obtained by the cameras. The term “N” refers to the number of principal components that the “Latent Structures with Discriminant Analysis ( PLSDA)” technique needs to be able to differentiate the classes with the “sensitivity” and “specificity” that appear in the values in the table. under. The terms “Sens. CV' and “Esp. CV' refer respectively to the 'sensitivity' and 'specificity' obtained after carrying out a cross-validation using the available data.

In relation to the second table, the term “1st Derivative 5 pts. MC” refers again to a type of statistical pre-treatment as described above, applied to the data provided by hyperspectral cameras. In this specific case, the term refers to the first derivative with a window of 5 amplitude points and, subsequently, a centered mean. The rest of the terms in the second table have the same meaning as the data in the first table discussed above.

In examples, it is possible to determine classes that may be able to identify nevi (class 1) vs. healthy tissue (class 2) vs. melanoma (class 3) with high Sensitivity and Specificity values.

In relation to the first table below, the term 1 Derivative 9 pts. AS” refers to a type of statistical pre-treatment as described above, applied to the data provided by hyperspectral cameras. In the specific example of this first table, the pretreatment refers to the Savitsky-Golay First Derivative with a 9-point-width window and autoscaling (mean and variance).

In relation to the second table below, the term “1st Derivative 31 pts. MC refers to a pretreatment related to the first derivative with a window of 31 amplitude points and, subsequently, a centered mean.

Figure 2a shows an example of a system configured to detect a cancerous lesion located in a portion of human tissue. System 300 comprises a robot arm 301. The robot arm may be a collaborative robot arm configured to work in an environment with people, where the robot arm does not need an enclosed area for use. Figure 2b shows the robot arm in detail. The robot comprises a base 301a, three arm sections 301b-301d, and four link members 301e-301h. Joining element 301e joins base 301a with section 301b. Joining element 301f joins section 301b to section 301c. Joining element 301g joins section 301c with section 301d. The connecting element 301h is located at one end of the section 301d.

The robot arm 301 may be mounted eg on a platform or cart (not shown). The platform or cart may be provided with wheels. These wheels allow the robotic arm, located on the platform or cart, to move easily relative to the patient, thus allowing the patient to obtain hyperspectral images of different parts of the subject's body without requiring the subject to move.

Turning to Figure 2a, system 300 includes an illumination subsystem 106 for irradiating the portion of the skin to be analyzed with light and hyperspectral cameras 105a and 105b, as previously described. The system also includes a subsystem of visualization (not shown) that includes a screen to show a hyperspectral image of the area of the patient to be analyzed, in real time, as well as a processing subsystem (shown). System 300 also includes a distance sensor 306 configured to measure the distance between the distance sensor and a region of the patient's skin portion to be tested. The distance sensor can be eg a laser photocell or a camera with "time of flight" (TOF) technology.

The processing subsystem is in communication with each of the illumination 106, distance sensor 306, display, and hyperspectral camera 105a, 105b subsystems and coordinates the operations of these subsystems to irradiate the patient, obtain spectral information from the patient, construct an image based on the spectral information and display the image.

Specifically, the illumination subsystem 106 irradiates the portion of the skin of the subject to be analyzed with light. The light interacts with a plurality of regions of the patient portion. The hyperspectral cameras 105a, 105b collect light from each of the plurality of regions of the skin portion of the subject to be analyzed and resolve the light from each region into a corresponding electromagnetic spectrum. The hyperspectral cameras 105a, 105b further generate a digital signal representing the spectra of all regions of the selected slice. The processor subsystem obtains the digital signal from the cameras, and processes the digital signal to generate a hyperspectral image.

System 300 also includes a robot arm control module (not shown) configured to receive the distance measured by the robot arm distance sensor and reposition the robot arm such that the distance received by the control module is maintained at a distance measured by the robot arm. range of distances in relation to any region of the portion of human tissue during a movement of the robot arm. It is important to highlight that the robot arm maintains its distance in relation to a region of the human tissue portion in any situation, even when the robot is not moving and the robot is statically positioned in relation to a region of the tissue portion. due to e.g. patient movements.

As shown in the detail of Figure 2c, a tool 340 may be attached to the end of the robot's attachment element 301h. Tool 340 may be made using 3D printing techniques. The hyperspectral cameras 105a and 105b, the distance sensor (not shown in this figure), and the lighting subsystem (not shown in this figure). shown in this figure) can be coupled to the tool 340. In this way, the robot arm 300 with the tool 340 can scan the portion of the tissue of the human being to be analyzed, illuminating it with the light provided by the lighting subsystem and can position the tool (and thus the hyperspectral cameras) at a suitable working distance relative to a region of the human tissue slice that is being analyzed. It is particularly important to position the cameras 105a, 105b within a range of distances (or a substantially constant distance) relative to a region of the human tissue portion being analyzed in the case of cameras with fixed optics. In the case of not keeping the distance substantially constant (eg within a range of /- one centimeter from the default distance of eg 15 centimeters), the cameras could go out of focus and the acquired hyperspectral image would not have adequate quality .

Specifically, the tool 340 comprises a housing 340a mounted on one end of the robot arm. Housing 304a comprises an interior surface 340b, two side walls 340c, 340d, and a space 340e formed within the housing. The hyperestral cameras, in use, are located in the space formed within the housing attached to the interior surface 340a. In examples, hyperspectral cameras are attached to the inner surface 340 by eg screws. In other examples, the hyperspectral cameras can be attached to the surface by a rail. In this manner, the cameras can move along the rail relative to the interior surface 340a so that the cameras are positioned to take full advantage of the light emitted by the lighting subsystem.

The housing further comprises a U-shaped cross-sectional support 340f and configured to hold the distance sensor (not shown in Figure 2c but visible in Figure 2a). This U-shaped bracket is located in the space formed inside the housing and attached to the inner surface, where the U-shaped bracket is located radially inward relative to one of the side walls.

In use, sidewalls 340c, 340d may be positioned at an angle between 40 and 50 degrees (specifically 45 degrees) relative to the surface defining interior surface 340a.

Figure 3 shows an example flow chart of a method for detecting a cancerous lesion located in a portion of human tissue.

Initially, a patient is examined. The examination may include visual examination and/or touching of the patient, as is conventionally done in this type of medical examination. The physician eg a dermatologist may focus on a particular area of the patient's skin, based on observations or comments made by the patient. Finally, the doctor selects a portion of the patient's human tissue to be analyzed.

In step 401, a robot arm (and thus hyperspectral cameras coupled to a robot arm) is positioned at a predefined and suitable distance from a region of the human tissue portion to be analyzed. As described above, the robot arm comprises a distance sensor configured to measure a distance between the distance sensor coupled to the robot arm and a region of the human tissue portion. The suitable distance between the sensor and a region of the surface to be analyzed can be eg between 14 and 16 centimeters, specifically 15 centimeters.

In step 402, a robot arm control module receives the distance measured by the robot arm distance sensor and repositions the robot arm such that the distance received by the control module remains within a range of distances relative to any region of the human tissue portion during a movement of the robot arm along the human tissue portion, i.e., the distance of the robot arm is maintained within a range of distances relative to any region of the human tissue portion eg in a range between 14 centimeters and 16 centimeters, specifically 15 centimeters. In this way, the cameras can be correctly focused throughout the procedure. This is especially important on fixed lens cameras.

In step 403, one or more hyperspectral images of the human portion are received using the image acquisition module. Initially, the selected human tissue portion may be irradiated with light via the illumination subsystem. Light is then obtained from the selected portion of human tissue using the hyperspectral cameras. Depending on the interactions between the portion of human tissue selected and the electromagnetic spectrum of the light used to irradiate this region, the light may be reflected, refracted, absorbed, and/or scattered relative to this portion of the patient.

In step 404, the electromagnetic spectrum may be extracted for each pixel comprised in each image acquired by the image acquisition module, where the electromagnetic spectrum includes a plurality of wavelengths. From light obtained from each region is obtained in a corresponding electromagnetic spectrum including a plurality of wavelengths.

Hyperspectral images can represent spectral information in various ways. For example, the image may include a two-dimensional map that represents the intensity of one or more selected wavelengths within each region of the patient. Such an image may be monochrome, with the intensity of the map in a given region based on the intensity of the selected wavelengths (eg, the intensity of the image directly proportional to the intensity of light at the selected wavelengths). Alternatively, the image can be in color, with the color of the map in a given region depending on the intensity of the selected wavelengths.

In some examples, one or more wavelengths of the electromagnetic spectrum can be selected for each pixel, taking into account the type of cancerous lesion to be evaluated. This selection may be based on one or more of several different types of information. For example, wavelengths can be selected based on a spectral signature library, which contains information about the spectral characteristics of one or more predetermined medical conditions. These spectral characteristics may include, for example, predetermined wavelengths that must be selected to determine if the subject has a particular medical condition.

In embodiments where the spectral image is displayed on a video screen, the image can be inspected, optionally while the subject is being examined, facilitating the obtaining of information that is useful in diagnosing and treating a medical condition. In some embodiments, a conventional visible light image of the subject's regions is displayed along with the image containing spectral information to aid in the correlation of the spectral characteristics with the physical characteristics of the subject.

In step 405, for each extracted electromagnetic spectrum, one or more detection algorithms may be applied to assess a condition related to the presence of a cancerous tissue lesion so that a cancerous lesion is detected. Representing the image in 2 dimensions, a color can be applied that allows the lesion to be identified and the affected area to be delimited. In this way, tissue lesions can be detected simply and efficiently.

Additionally, taking into account the information on the distance and displacement of the robot or collaborative robot, it is possible to obtain a 3D model in which the XY axis represents the area of the skin scanned and the area affected by the lesion, while the axis Z represents the height.

Although only a few particular embodiments and examples of the invention have been described herein, it will be understood by those skilled in the art that other alternative embodiments and/or uses of the invention are possible, as well as obvious modifications and equivalent elements. Furthermore, the present invention encompasses all possible combinations of the specific embodiments that have been described. Numerical signs relative to the drawings and placed in parentheses in a claim are intended only to increase the understanding of the claim, and should not be construed as limiting the scope of protection of the claim. The scope of the present invention should not be limited to particular embodiments, but should only be determined by a proper reading of the appended claims.

Claims (15)

1. A system configured to detect a cancerous lesion located in a portion of human tissue, characterized in that it comprises: - an image acquisition module configured to acquire one or more images of a portion of human tissue, where the image acquisition module comprises one or more hyperspectral cameras configured to acquire a hyperspectral image of the portion of human tissue, - an information extraction module configured to extract an electromagnetic spectrum for each pixel included in each image acquired by the image acquisition module, where the electromagnetic spectrum includes a plurality of wavelengths, - a cancer lesion discrimination module configured to receive the electromagnetic spectrum for each pixel extracted by the information extraction module and apply one or more detection algorithms to assess the condition of the tissue, - a robot arm, where the image acquisition module is coupled to the robot arm, characterized in that the robot arm comprises: or a distance sensor configured to measure a distance between the distance sensor and a region of the human tissue portion, or a robot arm control module configured to receive the distance measured by the robot arm distance sensor and reposition the robot arm such that the distance received by the control module remains within a range of distances relative to any region of the portion of human tissue during a movement of the robot arm. 2. A system according to claim 1, characterized in that the distance received by the control module is maintained in a range of distances between 14 centimeters and 16 centimeters in relation to any region of the portion of human tissue during a displacement of the robot arm, specifically 15 centimeters. A system according to claim 1-2, characterized in that the robot arm is an industrial robot, where the industrial robot comprises three or more axes with at least two degrees of freedom. 4. A system according to any one of claims 1-3, characterized in that the system further comprises one or more light sources coupled to the robot arm and configured to illuminate the portion of the human tissue for the hyperspectral cameras, where the light sources are configured to emit electromagnetic radiation in a wavelength range between 380nm and 2500nm, optionally the light sources are configured to emit electromagnetic radiation in a wavelength range between 380nm and 1000nm or in a wavelength range between 900 nm and 1700 nm. 5. A system according to claim 4, characterized in that the robot arm comprises a tool configured to support the hyperspectral cameras and light sources. 6. A system according to claim 5, characterized in that the tool comprises a housing mounted at one end of the robot arm, where the housing comprises an inner surface, two side walls and a space formed inside the housing, where the hyperspectral cameras, in use, are located in the space formed inside the housing attached to the inner surface. 7. A system according to claim 6, characterized in that the housing further comprises a support with a U-shaped cross-section and configured to support the distance sensor, where the support for supporting the distance sensor is located in the space formed inside the housing and attached to the inner surface. A system according to any one of claims 4-7, characterized in that the side walls are configured to support the light sources, where the side walls are configured to pivot relative to the interior surface of the housing, where In use, the side walls are located between 40 degrees and 50 degrees relative to a plane defined by the interior surface of the housing, optionally they are located at 45 degrees. 9. A system according to any one of claims 1-8, characterized in that the hyperspectral cameras of the image acquisition module are configured to operate in a range of wavelengths of the electromagnetic spectrum between 400 nm and 2500 nm, optionally the image acquisition module comprises two hyperspectral cameras, where the first hyperspectral camera is configured to operate in a range of wavelengths of the electromagnetic spectrum between 400 nm and 1000 nm and the second hyperspectral camera is configured to operate in a range of wavelengths of the electromagnetic spectrum between 1000 nm and 1700 nm A system according to any one of claims 1-9, characterized in that the system further comprises an electromagnetic spectrum correction module configured to correct the extracted electromagnetic spectrum for each pixel comprised in each image acquired by the acquisition module of pictures. 11. A system according to claim 10, characterized in that the electromagnetic spectrum correction module is adapted to: receive a first image acquired by the image acquisition module when the hyperspectral cameras do not receive light, obtain a measurement Rn related to the electromagnetic spectrum for a pixel included in the first image, where the measurement Rn comprises a plurality of wavelengths, receive a second image acquired by the image acquisition module of a reference material illuminated with electromagnetic radiation in a range of wavelengths between 400 nm and 2500 nm, obtaining an Rb measurement related to the electromagnetic spectrum for a pixel comprised in the second image, where the Rb measurement comprises a plurality of wavelengths, obtain, for each measurement of the electromagnetic spectrum M of each pixel included in each image acquired by the image acquisition module, a measurement of the corrected electromagnetic spectrum Rc following the following equation: R ^c = M-R ⁿ /R ^b -R ⁿ 12. A system according to any one of claims 1-11, characterized in that the system further comprises a module for pre-processing the electromagnetic spectrum configured to pre-process the electromagnetic spectrum extracted for each pixel comprised in each image acquired by the image acquisition module, where the pre-processing module is configured to select one of the following algorithms to pre-process the electromagnetic spectrum : Centered and Scaled by Unit Variance, Centered on Mean, Savitzky-Golay Smoothing, Savitzky-Golay First Derivative, Savitzky-Golay Second Derivative, Standard Normal Variance Scale, Multiplicative Dispersion/Signal Correction. 13. A system according to any one of claims 1-12, characterized in that the cancer lesion discrimination module is configured to select one or more of the following cancer lesion detection algorithms: decision trees, support machines of vectors, neural networks, convolutional neural networks, random forests or projection of latent structures with discriminant analysis. 14. Method for detecting a cancerous lesion located in a portion of human tissue using a system according to any one of claims 1-13, the method characterized in that it comprises: - placing the distance sensor of the robot arm at a predetermined distance within a range of distances in relation to a region of a portion of human tissue, - moving the robot arm so that the distance of the distance sensor of the robot arm in relation to any region of the portion of the human tissue remains within the range of distances, - receiving one or more hyperspectral images of the portion of human tissue, - extracting the electromagnetic spectrum for each pixel comprised in each image acquired by the image acquisition module, where the electromagnetic spectrum includes a plurality of wavelengths, - for each extracted electromagnetic spectrum, applying one or more detection algorithms to evaluate a condition related to the presence of a cancerous tissue lesion so that a cancerous lesion is detected. 15. A method according to claim 14, characterized in that it comprises selecting one or more wavelengths for each extracted electromagnetic spectrum depending on the type of cancerous lesion to be evaluated.