AT525533B1 - Method and system for analyzing a blood sample - Google Patents

Abstract

According to the invention, a method for analyzing a blood sample is provided, in which, in a first step, an image of a blood sample is recorded by a camera and, in a second step, the image is evaluated by a computer, with the cells visible on the image being classified. in which, based on the classification carried out in the second step, the image is re-evaluated in a third step, with the cells visible in the image being reclassified, with another image of a different region of the blood sample based on the classification carried out in the second step is captured and this image is subjected to the second and third steps.

Description

Description

The invention relates to a method for analyzing a blood sample, in which in a first step an image of a blood sample is taken by a camera, and in a second step the image is evaluated by a computer, with the cells visible on the image being classified become.

The invention further relates to a system for analyzing a blood sample, in particular for carrying out a method according to the invention, comprising a recording for the blood sample, a camera and a computer connected to the camera, which is designed to evaluate the images received from the camera , where the cells visible on the image are classified.

[0003] When analyzing blood samples, the blood sample from a human or animal is examined with regard to the quantity and/or the quality of the blood cells it contains. From such an examination, far-reaching conclusions can be drawn about the condition and the microcirculation in the tissue, the existing immune protection and the overall health of the body in general.

Several methods for analyzing a blood sample are known from the prior art in order to determine the cell composition and the cell morphology as part of a blood count.

The simplest option is for a doctor or other trained person to manually analyze a blood sample using a conventional microscope. The blood sample is provided on a slide made of glass, for example, and then optionally stained with a liquid, which makes the cell composition easier to see visually. The blood sample is dried and can then be viewed and evaluated under a microscope. The image of the blood sample is magnified approx. 400 to 1,000 times in order to make the cells visible to the human eye. During the analysis itself, the user determines the cell types based on size, shape and/or color. Disadvantages of these methods are the high amount of work involved and the high subjectivity of the result, the quality of which depends primarily on the user's work performance.

[0006] In order to eliminate these disadvantages, hematology devices can be used in order to be able to carry out automatic blood analyses. A variety of different systems exist for counting or classifying the blood cells in a blood sample. Most such devices are based on measuring the change in average electrical conductivity between two electrodes placed in a liquid, each placed in a chamber separated by a small opening (Coulter counter). The opening is slightly larger than the cells or particles to be measured. The cells or particles to be measured flow from one chamber into the other chamber by means of a flow generated, for example, by negative pressure, so that the electrical resistance between the two electrodes changes. Since the change is proportional to the size of the cell or particle, both the number and the size of the cells or particles can be recorded and evaluated.

A disadvantage of these systems is that in the case of red blood essentially only the number of red blood cells and hemoglobin can be measured directly. The other components are calculated mathematically and not measured, so the error rate is relatively high. Furthermore, the shape of the red blood cells and in particular the number of red blood cells with a changed shape cannot be determined using conventional methods. This method can also be used to determine the number and size of white blood cells.

[0008] With today's analysis devices, the flow cytometry method is used for the detection and counting of white blood cells, in which the cells are guided past an electrical voltage or a light beam. The resulting scattered light is captured and evaluated. In particular, this enables the individual types of white blood cells to be distinguished and counted accordingly. The disadvantage of this method, however, is that immature forms of granulocytes (promyelocytes, myelocytes, me-

tamyelocytes and ligaments) cannot be effectively distinguished and counted from one another. Furthermore, such analyzers do not permit the detection of nuclear and cytoplasmic changes or pathological granularity in the evaluation of leukocyte lineage. Here, too, many parameters are calculated and estimated, resulting in a high error rate.

[0009] More reliable automatic analyzers are based on automatic microscopy, in which images of the blood sample are taken by a camera and a microscope and these images are then evaluated based on appropriate algorithms. These analyzers reduce the analysis time and increase the objectivity of the measurements.

[0010] CN 110852288 A shows a method for classifying a cell image, in particular for classifying epithelial cells, in which an image of a cell is recorded and this image is roughly classified using a rough classification model.

[0011] CN 111815633 A shows a method for classifying pathological images, in particular with regard to tumor cells, the pathological image being divided into a number of blocks and the blocks being classified using a first classification model. The blocks are then reclassified using a second classification model.

[0012] A disadvantage of such known systems is that some pathological cells are optically very similar to one another and can only be distinguished from one another, for example, by their granularity or other inconspicuous details. Even if the algorithms used are designed to detect even small differences, the optical similarities often lead to incorrect determinations or incorrect classifications and thus to incorrect analysis results.

It is therefore an object of the invention to increase the analysis reliability in a method of the type mentioned at the outset and to reduce or avoid misclassifications of cells. In particular, a simple, reliable and rapid method for analyzing a blood sample is to be provided.

According to the invention, in a method of the type mentioned at the outset, it is provided that, based on the classification carried out in the second step, the image is re-evaluated in a third step, with the cells visible on the image being reclassified, and based on the classification carried out in the second step, a further image of another area of the blood sample is taken and this image is subjected to the second and the third step.

[0015] It is therefore provided that an image (photo) of a blood sample is first recorded. The image may depict the entire blood sample or, preferably, a portion of the blood sample. This image is evaluated using a first algorithm, with each recognized cell being assigned to a (predefined) group of cells, in particular on the basis of its shape, its size, its color and possibly other properties. In this way, a first classification is obtained which, based on the number of cells in each category, makes it possible to obtain a first assessment of the state of health. Based on the result of the first classification of the cells, a decision is now made as to whether there will be no further examination of the image, e.g. if the blood count indicates a good state of health, or whether a further classification (reclassification) should be carried out using a second algorithm, especially if the Blood count indicates a pathological condition, e.g. due to the presence of pathological cells or too low or too high a number of certain blood components. In order to make a decision in this step, predefined values are preferably calculated according to a predefined routine and compared with predefined threshold values or reference ranges, for example whether the number of blast cells and promyelocytes exceeds a certain concentration in the blood. If it is determined here that a pathological condition may be present, a specific mathematical algorithm tailored to the corresponding situation can be used in order to improve the classification, ie to reduce the probability of error. This step makes it possible after

the first analysis by a general mathematical algorithm for the detection of cells to select a specific special mathematical algorithm tailored to the situation detected in the first analysis and to carry out a second analysis with it, thereby reducing the probability of error in the classification and thus the result of the blood analysis in comparison is significantly improved to a method that only includes the first analysis.

[0016] This is possible because the second mathematical algorithm is specially tailored to a specific clinical picture recognized in the second step by the first algorithm and is therefore able to recognize even the smallest optical differences between the individual cells.

[0017] The classification is carried out, for example, in that the image is evaluated by image acquisition, with the number of cells, their color, size and shape, for example, being determined. These values can, for example, be transferred to numeric scales and stored. Subsequently, on the basis of these ascertained values, it is determined with the aid of the first algorithm, which further processes and, for example, weights the ascertained values, which types of cells were recognized in the image.

[0018] During the reclassification, the detected cells are re-examined, this time using a different, second algorithm, and each cell type is assigned again.

According to the invention it is provided that, based on the classification carried out in the second step, a further image of another partial area of the blood sample is recorded and this further image is subjected to the second and the third step. It is provided here, for example, that a specific predefined number of certain blood parts, for example red blood cells and/or white blood cells, must be recognized and classified accordingly by the first algorithm in order to enable a meaningful analysis. If the corresponding number of blood parts is not present in the first image and accordingly not enough blood parts are classified by the first mathematical algorithm, another image of a different area of the blood sample can be recorded by the camera, e.g. in an area adjacent to the first image the blood sample. Like the first image, this further image is also examined and evaluated according to the second step and then, if necessary, according to the third step. If too few blood parts are recognized and classified again, a further, third image and possibly further images can be recorded. The recording of further images can preferably be made possible by a mechanical displacement device, which is preferably connected to the computer, for example via a control unit, and is optionally controlled accordingly by the computer. The individual images preferably do not overlap one another and are particularly preferably directly adjacent to one another in order to enable the blood sample to be analyzed as precisely as possible. The displacement device can displace the blood sample relative to the camera and/or the camera relative to the blood sample in order to make it possible to produce an image of a recording area of the blood sample that is independent of the first image.

The camera can be any device that can capture an image and store it in the form of (electronic) data on a storage medium so that it can be evaluated by an algorithm.

The computer is preferably a digital computer and can, for example, be designed as a workstation computer or tablet computer.

The mathematical algorithms (potentially) used in the third step are preferably stored in a database. Depending on the requirements or the result in the second step, the appropriate algorithm is selected and used to carry out the evaluation in the third step.

In the classification or reclassification, for example, the red blood cells and the blood platelets are classified according to their size, color and shape and their concentration in the blood is determined. The greater the proportion of morphologically altered and destroyed blood cells, the more pronounced are tissue hypoxia and microcirculation disorders.

In a preferred embodiment, it is provided that the reclassification in the third method step is carried out only with regard to a specific cell type, with, for example, only the white blood cells being reclassified. In particular, this makes it possible to use an even more specific algorithm that is specialized for this type of cell and is designed or trained accordingly, and can therefore deliver an even more precise result. Alternatively, all cell types are reclassified.

[0025] After the third step, the result of the blood analysis, ie in particular the number of individual recognized blood cells, is preferably stored on a data medium and/or output on an output unit, for example a screen. This enables the data to be stored, for example, for later use or to be displayed to a user in order to draw appropriate conclusions from the data.

The mathematical algorithms are preferably stored in a local data medium of the computer, but can also be stored, for example, on a central data medium (for example a server), which can be accessed by a number of computers to carry out the method. This allows the algorithms to be updated quickly and easily.

The method is (partially) computer-implemented, with the necessary algorithms and other program parts for carrying out the method being stored in particular on a data carrier.

Before the first method step, the blood sample is optionally prepared accordingly, e.g. applied to a carrier and dried and then, e.g.

[0029] White blood cells are differentiated according to their respective characterizing properties of size, color and shape. For example, the white blood cells are divided into the following groups during the analysis: * basophils (involved in the development of allergic reactions) * eosinophils (main task is to fight multicellular parasites) * segmented neutrophils (largest group of granulocytes) * banded neutrophils (young, immature neutrophils) * Lymphocytes (Provide humoral and cellular immunity and regulate the activity of other cell types) * Monocytes (Provide innate immunity and bear pattern-recognizing receptors) * Large granular lymphocytes (proportion of approximately 5% to 20% of the total number of lymphocytes in the blood)

In addition to these normal blood components, pathological cells can also be contained in the blood, for example the following: * Promyelocytes (myeloid cells, the precursors of granulocytes during granulopoiesis, developing from myeloblasts into myelocytes) * Myelocytes (young cells of the granulocytic series, come normally found in the bone marrow but not in the peripheral blood) * Metamyelocytes (myeloid cells undergoing granulopoiesis and derived from myelocytes) * Blasts (young, immature cells produced in the bone marrow) * Plasma cells (immunocompetent cells that produce antibodies. Normally, their proportion of white blood cells is not more than 0.5%) * Atypical lymphocytes (the presence of these cells in the blood indicates an acute infection and thus the need for antibody synthesis) * Reactive lymphocytes (cytotoxic lymphocytes, which due to the stimulated by an antigen) * prolymphocytes (progenitor cells of lymphopoiesis, the direct precursors of lymphocytes, which arise from lymphoblasts)

It is preferably provided that a first neural network is used for classification in the second step. The first mathematical algorithm used in the second step includes or is formed by a first (artificial) neural network. Neural networks are highly efficient when complex patterns need to be recognized and classified during image processing. The first neural network is preferably designed or trained to be able to analyze blood samples of all types, in particular normal, healthy blood samples with at most slight deviations from the norm, such as reactive lymphocytosis or a slight left shift, in particular to be able to analyze all possible cells capture and categorize. In order to make this possible, it is preferably provided that the first neural network is trained with a large number of blood counts which are healthy or deviate only slightly from the norm. With this training, the first neural network is better at recognizing and classifying healthy blood counts, that is, at least substantially normal, free from malignant cells and serious pathological conditions.

Furthermore, it is preferably provided that a second neural network is used for the classification in the third step. The mathematical algorithm used in the third step includes or is formed by a second (artificial) neural network, which differs from the first mathematical algorithm or from the first neural network and is specifically intended and designed for a specific blood count classified in the second step . The second neural network is thus preferably designed or trained in order to be able to distinguish pathological cells from one another and to be able to categorize them accordingly. In order to make this possible, it is preferably provided that the second neural network is trained (exclusively) with a large number of sick blood counts, ie blood counts that deviate from the norm, in particular malignant cells, rare and/or pathological clinical pictures. With this training, the second neural network is better at recognizing, distinguishing and classifying very immature, malformed and deformed cell groups, as it has a high sensitivity to small deviations in nuclei and cytoplasms of immature cells.

Thus, the first algorithm is better at detecting immature cells and classifying normal cells, while the second algorithm is better at distinguishing and classifying individual groups of immature cells.

Neural network training involves estimating the parameters of the neural network to enable the most accurate classification of blood cells possible. In this case, the parameters of the neural network are adjusted on the basis of training data, ie blood counts that have already been classified and are fed to the neural network, which leads to more precise results.

If only the first algorithm is used for classification, only a low accuracy in distinguishing between immature cell groups is achieved, since the first algorithm does not (reliably) recognize the fine details in distinguishing these cell groups. On the other hand, if only the second algorithm is used to classify a normal blood sample, an overly sensitive analysis is output which is susceptible to false positive detections and is therefore also inaccurate.

According to the invention, a system for analyzing a blood sample, in particular for carrying out a method according to the invention, is also provided, comprising a receptacle for the blood sample, a camera and a computer connected to the camera, which is designed to evaluate the images received from the camera , wherein the cells visible on the image are classified, the computer being further configured to re-evaluate the image based on the classification of the cells, the cells visible on the image being reclassified, the system being further configured to taking another image of another area of the blood sample based on the classification performed in the second step and subjecting this image to the second and third steps.

[0037] A microscope is preferably also provided here in order to enlarge the image of the blood sample. The magnification of the microscope is preferably adjustable and particularly preferred

the microscope is connected to the computer and can be controlled by it, for example to change the magnification.

Furthermore, the system preferably includes a displacement device in order to move the receptacle for the blood sample relative to the camera, in particular to translate it.

The displacement device can alternatively or additionally preferably be designed to move the camera relative to the blood sample, in particular to translate it. The invention is explained in more detail below with reference to an exemplary embodiment shown schematically in the drawing. 1 shows a method according to the invention in a schematic representation, FIG. 2 shows a schematic representation of a system according to the invention, and FIG. 3 shows some exemplary image sections of blood cells.

In Fig. 1, a method according to the invention is shown schematically. At 1 the method is started after a blood sample to be examined has been placed in a photograph of a microscope. At 2, an image of part of the blood sample is now made using a camera (step 1). The image is then fed to a computer and evaluated by it using a first mathematical algorithm, in particular a first neural network, with the cells recognized in the image being classified, for example depending on the color, size and shape of the individual cells. At 3 it is determined whether a first cell type (e.g. red blood cells) has been recognized in a sufficient quantity on the image. If not, the classification of the image is first completed at 4 and the recording area of the camera is then changed at 5 so that another partial area of the blood sample can be captured by the camera, creating another image that is also analyzed and the detected cells classified become. This is continued until it is determined at 3 that sufficient cells of one type (e.g. red blood cells) have now been recorded and classified on the images recorded up to now. It is then checked at 6 whether a second cell type (e.g. white blood cells) was recognized in a sufficient quantity on the images recorded up to now. If not, the detected cells are analyzed at 7 and the field of view of the camera is changed at 5 in order to obtain an image of another partial area of the blood sample, which is also examined again. If sufficient cells of the second cell type were detected, the classified cells are checked with regard to several criteria at 8, 9, 10, 11 and 12, for example as follows:

At 8: number of blasts and promyelocytes greater than 9

At 9: number of lymphocytes, reactive lymphocytes, atypical lymphocytes, large granular lymphocytes and plasma cells greater than 0 and the quotient of reactive lymphocytes, atypical lymphocytes, large granular lymphocytes and plasma cells and lymphocytes, reactive lymphocytes, atypical lymphocytes, large granular lymphocytes and plasma cells is greater than 0.2

At 10: Banded neutrophil to leukocyte ratio is greater than 0.1

At 11: ratio of myezolytes and metamylocytes (IMGRA) and leukocytes is greater than 0.04

At 12, banded neutrophil to leukocyte ratio is greater than 0.07 and myezolyte to metamylocyte (IMGRA) to leukocyte ratio is greater than 0.02.

If none of the conditions 8, 9, 10, 11 or 12 are met, the blood sample is classified as "normal" at 13, the results are stored at 14 and the method is terminated at 15.

If one of the conditions 8, 9, 10, 11 or 12 should be met, the blood sample is classified at 16 as "pathological". Based on which of the conditions is met, a second mathematical algorithm is then selected, e.g. from a database, and the recognized cells are analyzed and classified at 17 again. This ensures that the cells are recognized and assigned in the best possible way according to the condition of the blood sample,

so that incorrect classifications are avoided or reduced as far as possible. After the reclassification at 17, the result is stored at 14 and the method ends at 15.

The cells are divided using the classification or the reclassification, for example, as follows:

White Red Blood Cells Platelets Blood Cells Basophils Size Color Shape Microplatelets Eosinophils Microcytes Polychromy Poikilo Macrocytes Platelets Promyelo Macrocytes Hypochromia Codo Normal cytes cytes Platelets Myelocytes Normal Hyperchromia Schisto Size cytes Band Normochromia Helm Neutrophil Cells Segmented Sickle Neutrophil Cells Lymphocytes Spherocytes Monocytes Elliptocytes Plasma Cells Ovalocytes Reactive Tears Lymphocytes Droplet cells Large granular stomato lymphocytes cells Prolympho acanthocytes cells Atypical echino lymphocytes cells Blasts Normal form

2 shows a system according to the invention, which is designed to carry out a method according to FIG. The system includes a microscope 18, a camera 19, a receptacle 20 for a blood sample, a computer 21, an output unit 22 designed as a screen, a control unit 23 and a displacement device 24. The computer 21 controls the method and receives the data from the camera 19 Images of the blood samples to subsequently evaluate them and classify the detected cells and, if necessary, select the appropriate algorithms and reclassify the cells. The displacement device 24 is designed to move the receptacle 20 relative to the camera 19 or the camera 19 relative to the receptacle 20 in all spatial directions, as required, in order on the one hand to enable the generation of images of several partial areas of the blood sample and on the other hand to focus the camera 19 to improve the blood test. The result of the blood analysis, in particular the classification, can be displayed on the output unit 22 .

[0050] In FIG. 3 some image sections of blood cells are shown as examples, each of which has a high degree of visual similarity to one another. In Fig. 3a are lymphocytes 25, prolympho-

cytes 26 and atypical lymphocytes 27 shown. In Fig. 3b monocytes 28, blasts 29 and reactive lymphocytes 30 are shown. In Fig. 3c, promyelocytes 31, myelocytes 32 and metamyelocytes 33 are shown.

Claims (6)

patent claims 1. A method for analyzing a blood sample, in which in a first step an image of a blood sample is recorded by a camera, and in a second step the image is evaluated by a computer, the cells visible on the image being classified, characterized in that that, based on the classification carried out in the second step, the image is re-evaluated in a third step, with the cells visible in the image being reclassified, with another image of a different area of the blood sample being recorded based on the classification carried out in the second step and this image is subjected to the second and third steps. 2. Method according to claim 1, characterized in that in the second step a first neural network is used for the classification. 3. The method according to claim 1 or 2, characterized in that a second neural network is used for classification in the third step. 4. System for analyzing a blood sample, in particular for carrying out a method according to one of claims 1 to 3, comprising a receptacle for the blood sample, a camera and a computer connected to the camera, which is designed to evaluate the images received from the camera , wherein the cells visible on the image are classified, characterized in that the computer is further configured to re-evaluate the image based on the classification of the cells, the cells visible on the image being reclassified, the system being further configured is to take a further image of another area of the blood sample based on the classification carried out in the second step and to subject this image to the second and the third step. 5. System according to claim 4, characterized in that a microscope is further provided to magnify the image of the blood sample. 6. System according to claim 4 or 5, characterized in that a displacement device is provided in order to move the recording for the blood sample relative to the camera, in particular to translate it. 2 sheets of drawings