Detection of a single layer of blood smear

Abstract

A method for detecting a single layer of red blood cells (RBCs) in a blood smear or a small section of a blood smear that is suitable for diagnosis. The method comprises: determining the coverage of RBCs in a blood smear; determining the hemoglobin distribution; determining the presence of non-RBC cells and / or abnormal RBCs; and assigning a label of diagnostic usability to the blood smear or a small section of a blood smear.

Description

[Technical Field]

[0001] The present invention relates to a method for detecting a single layer of red blood cells (RBCs) in a blood smear or a small section of a blood smear that is suitable for diagnosis, comprising: (a) determining the coverage rate of RBCs in the blood smear; (b) determining the hemoglobin distribution; (c) determining the presence of non-RBC cells and / or abnormal RBCs; and (d) assigning a label of diagnostic usability to the blood smear or small section of a blood smear. A method for detecting a single layer of red blood cells (RBCs) in a blood smear that is suitable for diagnosis is also provided, comprising: (a) acquiring a number of optical images of the blood smear; (b) classifying each image on the basis of the presence of a single layer of region, performed using an artificial neural network or according to a parameter-based method; (c) generating a heatmap of the images according to the presence of a single layer of region suitable for diagnosis; (d) identifying the region of the single layer of region suitable for diagnosis, by analyzing groups of adjacent images in the heatmap; and combining all adjacent images of the single layer of region suitable for diagnosis to form a single-layer panoramic overview image. [Background technology]

[0002] Blood smear analysis allows for the quantification and analysis of various types of blood cells, including cell counting and detection of morphological abnormalities that may indicate pathological processes. Preparing a blood smear typically involves dropping blood onto a glass slide, spreading it, and then staining it for cellular analysis. As the blood is spread, a gradual decrease in blood concentration occurs across the entire smear, eventually tapering to a narrower end. Not all smear areas are suitable for differential cell counting and diagnostic evaluation during the smear test, and this is particularly crucial for assessing and analyzing the morphology of red blood cells (RBCs).

[0003] In areas of high concentration within a blood smear, many cells may be aggregated or overlapping, making it difficult to assess cell morphology and determine the hemoglobin content of red blood cells. Conversely, near the edges of the smear, red blood cells may be too far apart, obscuring the central pallor zone and potentially leading to misdiagnosis. A diagnostically and morphologically appropriate blood smear has a monolayer morphology, meaning the red blood cells are well spread out with minimal cell overlap and aggregation, and the central pallor zone is clearly visible.

[0004] Morphological analysis of blood samples remains a crucial diagnostic starting point, especially for patients with malignant blood disorders. Such analysis is advantageous because blood smear testing is non-invasive, time-efficient, cost-effective, and allows for the suggestion of additional differential diagnoses. Therefore, cytomorphological examination remains an important part of current and future blood analysis.

[0005] The first step in such analysis is to identify the appropriate area of ​​the blood smear for screening, i.e., to identify monolayer structures. In clinical settings, specialists are often required to identify monolayers manually, an act generally considered tedious, inconsistent, subjective, and statistically biased. Automated systems, such as those commercially available from Cellavision, e.g., as described in Patent Document 1, still typically require human intervention. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] US2016078276 [Overview of the project] [Problems that the invention aims to solve]

[0007] Therefore, there is a need for an improved, automatable method for identifying and assessing monolayers in blood smears of patient samples. [Means for solving the problem]

[0008] The present invention addresses this need and, in a first embodiment, provides a method for detecting a single layer of red blood cells (RBCs) in a blood smear or sub-section of a blood smear that is suitable for diagnosis, comprising: (a) determining the coverage area of ​​RBCs in a blood smear or sub-section of a blood smear based on the spacing and distance between the RBCs and neighboring cells; (b) determining the hemoglobin distribution of RBCs in a blood smear or sub-section of a blood smear; (c) determining the presence of non-RBC cells and / or abnormal RBCs in a blood smear or sub-section of a blood smear; (d) in step (i), the coverage area is determined by the distance between the RBCs and neighboring cells being approximately 1.0 to 2.0 times the diameter of the RBC; and in step (ii), the central pale region (central pallor) occupies approximately 25 to 40% of the diameter of the RBC. A method is provided for detecting a diagnostically usable red blood cell (RBC) monolayer region in a blood smear or blood smear sub-section, comprising the steps of: (iii) identifying a region) within the RBCs; and (iii) assigning a diagnostic usability label (classification) to the blood smear or blood smear sub-section if, in step (c), non-RBC cells and / or abnormal RBCs are not identified, and / or identified non-RBC cells and / or abnormal RBCs are labeled and excluded from further analysis.Furthermore, in a further embodiment, the present invention provides a method for detecting a diagnostically usable RBC monolayer in a blood smear, comprising: (a) acquiring a number of optical images of the blood smear; (b) classifying each image on the basis of the presence of a diagnostically usable monolayer region, performed either (i) using an artificial neural network or (ii) according to a parameter-based method described herein; (c) generating a heatmap of the images according to the presence of a diagnostically usable monolayer region; (d) identifying a region of a diagnostically usable monolayer region, where adjacent images in the heatmap generated in step (c) are analyzed; (e) verifying the diagnostically usable monolayer region identified in step (d) by calculating the density configuration of the selected images; and (f) combining all adjacent images of the diagnostically usable monolayer region to form a monolayer panoramic overview image. Advantageously, these methods enable the detection of a diagnostically usable RBC monolayer region in a blood smear, which can then be used in downstream diagnostic tests. Therefore, the parameter-based and AI-based methods of the present invention enable further automation of evaluations that no longer rely on input from histology or hematology experts. Furthermore, since the two methods are advantageously complementary to each other, they enable improvement of results obtained using automated characterization techniques.

[0009] In a preferred embodiment of the present invention, the steps for determining the area coverage of RBCs in a blood smear or sub-section of a blood smear, based on the spacing and distance between the cells and neighboring cells, include: (a) assigning a tight bounding box to each cell; (b) assigning a central position to each cell; (c) determining the diameter and / or area of ​​the cell or bounding box; (d) determining the circularity index; (e) optionally determining further geometric parameters; (f) identifying one or more directly adjacent cells within a predetermined examination radius; (g) (i) a step of selecting one directly adjacent cell; (h) a step of determining the distance between the RBC and the cell directly adjacent to the RBC; (i) a step of repeating steps (g) and (h) for one or more further directly adjacent cells identified in step (f); and a step of classifying the region according to a distance criterion that the distance between the RBC and the cell directly adjacent to the RBC is 1.0 to 2.0 times the diameter of the RBC, which determines the coverage of the region usable for diagnosis, and that at least 75 to 80% of the measured distances must be within the range of 1.0 to 2.0 times the diameter of the RBC determined in step (c).

[0010] In a more preferred embodiment, the step of determining the distribution of RBCs in a blood smear or sub-section of a blood smear includes: (a) assigning a tight bounding box to each cell; (b) determining the contour lines of the cell boundaries; (c) determining the centroid of the geometric shape of the cells; (d) determining the hemoglobin distribution parameter (HDP); and (e) classifying the central clear region according to the HDP, thereby identifying a clear region where the HDP is approximately 0.5 to 2.0 and is usable for diagnosis.

[0011] In further embodiments, the step of identifying the presence of non-RBC cells and / or abnormal RBCs in the blood smear or sub-section of the blood smear is, in addition to steps (3)(a) to (3)(d): (e) a step of determining the diameter of the cell; (f) a step of determining the curvature value of the cell; (g) a step of classifying the cell as a non-RBC or abnormal cell if the diameter of the cell is >8.5 μm or <6 μm; and / or the HDP is <0.5 or >2.0; and / or the curvature value is >1 / 3 μm.

[0012] In a more preferred embodiment, cells classified as abnormal are further analyzed to diagnose a disease.

[0013] In a more preferred embodiment, the artificial neural network is trained on a training dataset generated by a pathologist, a hematologist, or a group of pathologists and hematologists.

[0014] In yet another preferred embodiment, the classification is a binary classification of "usable for diagnosis" or "unusable for diagnosis."

[0015] In further embodiments, the assignment of labels to the heatmap or diagnostic availability described above is based on the results of a binary classification, or the heatmap represents a classification according to the parameter-based method described herein. It is particularly preferable that the heatmap or classification results highlight images or regions that are available for diagnostic use.

[0016] In another preferred embodiment, the identification of a single-layer region usable for diagnosis is performed using a sliding window procedure in a group of adjacent images. It is particularly preferred that this identification is performed with a window size of 8x8 images.

[0017] It is even more preferable to eliminate single-layer regions usable for diagnosis that overlap and / or touch within a group of images.

[0018] In another preferred embodiment, the method for detecting a diagnostically usable red blood cell (RBC) monolayer site in a blood smear specimen or a small section of a blood smear specimen according to the present invention is performed using a number of optical images, and the diagnostic usability label assigned to the images included in the blood smear specimen or the small section of the blood smear specimen is compared with each image of the panoramic overview image of the monolayer obtained by the method for detecting a diagnostically usable red blood cell (RBC) monolayer site in the blood smear specimen. If there is an overlap in both images, it is identified as a preferably diagnostically usable RBC monolayer site.

[0019] In a more preferred embodiment, the blood smear specimen is obtained using any method aimed at generating a cell monolayer. The blood smear is particularly preferably prepared as a wedge smear on a microscope slide glass, by spinning the slide glass in a spinner centrifuge, or in a flow cell.

[0020] In a further aspect, the present invention relates to the use of the method according to the present invention for the selection of a diagnostically usable RBC monolayer site.

[0021] In yet another aspect, the present invention is a method for diagnosing a blood-related disease, comprising: (i) identifying at least one diagnostically usable RBC monolayer site in a blood smear of a subject sample, preferably using the method for detecting a diagnostically usable red blood cell (RBC) monolayer site in a blood smear specimen or a small section of a blood smear specimen described herein; (ii) identifying abnormal cells in the blood smear, preferably using the method for detecting a diagnostically usable red blood cell (RBC) monolayer site in a blood smear specimen or a small section of a blood smear specimen described herein; and (iii) optionally, staining the cells with a dye to visualize the intracellular molecular structure in order to identify abnormal structures, wherein the presence of abnormal cells and optionally the presence of abnormal molecular structures indicate a blood-related disease.

Brief Description of the Drawings

[0022] [Figure 1]Schematic diagram of an embodiment of the present invention. At the bottom of the drawing, a microscope slide (1) and optionally a cover (3) are shown. The slide may be labeled with a barcode (2). On that slide, a blood smear specimen (4) can be seen. The site (5) to be optically scanned of the blood smear specimen is shown as a square structure. According to the analysis according to the present invention (upper part of the drawing), a heat map of an image indicating the presence of a monolayer region (6) usable for diagnosis is generated. [Figure 2] FIG. 2A represents the placement of a monolayer region (13) usable for diagnosis, including each single image having an appropriate monolayer (12) within the scanned site (10). An inappropriate site (11) is also shown. FIG. 2B shows a site (14) of 8×8 adjacent single images having an appropriate monolayer within the scanned site (10). [Figure 3] An example of a workflow encompassing an embodiment of the present invention is shown. The exemplary workflow starts with blood sampling (20), followed by blood processing (21), preparation of a blood smear (22), imaging of the blood smear (23), image quality assessment (24), cell detection (25), classification of monolayers (26), generation of a heat map (27), selection and combination of images (27), and RBC smear analysis (28). [Figure 4]Further examples of workflows encompassing embodiments of the present invention are shown. The exemplary workflow begins with image acquisition (30), followed by cell detection (31), determination of cell properties (32), detection and determination of distances to directly adjacent neighboring cells (33), determination of cell geometric parameters (34), determination of exemplary Hu moments for handcrafted parameters and / or sets of geometric parameters (35), determination of HDP (36), classification of the smear or its subdivisions as monolayer regions usable / unusable for diagnosis (37), input from a field expert or knowledge database (38), and expert input of already acquired images or publicly available images of RBCs and monolayers. The process is followed by classification (39), preparation of a corresponding training dataset for the neural network (40), training of the neural network (41), supplying the trained neural network with test data from the acquired images (42), preparation of the neural network for decision inference (44), release of the prepared model (45), application of the acquired images based on the process model (45), classification of the smear or its subdivisions as single-layer regions usable / unusable for diagnosis based on the results of process (45) (46), generation of a heatmap (47), selection and combination of images (48), and RBC smear analysis (49). Processes (40) and (41) represent the training phase (50), processes (42) to (44) represent the inference phase (51), and process (45) represents the use phase (52). [Modes for carrying out the invention]

[0023] The present invention is described in relation to specific embodiments, but this description should not be construed as limiting.

[0024] Before describing exemplary embodiments of the present invention in detail, important definitions necessary for understanding the invention are made.

[0025] As used herein and in the appended claims, the singular forms “a” and “an” also include their respective plural forms unless the context explicitly negates them.

[0026] In the context of the present invention, the terms “about” and “approximately” represent a range of accuracy that a person skilled in the art would understand to be such that the technical effect of the configuration in question is still ensured. The terms typically indicate a deviation of ±20%, preferably ±15%, more preferably ±10%, and even more preferably ±5% from the given numerical value.

[0027] Naturally, the term "including" is not limiting. For the purposes of this invention, the terms "consisting of" or "substantially consisting of" are considered preferred embodiments of the term "comprising of." Hereinafter, when a group is defined as containing at least a certain number of embodiments, it is also preferable to include a group consisting only of these embodiments.

[0028] Furthermore, in the description or claims, terms such as "(i)", "(ii)", "(iii)", or "(a)", "(b)", "(c)", "(d)", or "first", "second", "third" are used to distinguish between similar or structural elements and do not necessarily indicate an order or chronological sequence.

[0029] Naturally, the terms used in this manner are interchangeable in the appropriate context, and embodiments of the invention described herein may be carried out in a different order than those described or explained herein. Where those terms relate to methods, procedures, or steps of use, there is no consistency in time or time intervals between steps. That is, unless otherwise indicated, the steps may be carried out simultaneously, or there may be time intervals of seconds, minutes, hours, days, weeks, etc., between such steps.

[0030] Naturally, the specific methods, protocols, etc., described herein are subject to change, and therefore the present invention is not limited thereto. Similarly, the terms used herein are for the purpose of describing specific embodiments only and do not limit the scope of the present invention, which is limited solely by the appended claims.

[0031] Drawings should be considered schematic representations, and the elements depicted in them are not necessarily shown to scale. Rather, the various elements are represented in a way that makes their function and general purpose clear to those skilled in the art.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art.

[0033] Regardless of grammatical usage, the term includes individuals who identify as male, female, or otherwise.

[0034] As described above, the present invention, in one embodiment, is a method for detecting a single layer of red blood cells (RBCs) in a blood smear or sub-section of a blood smear that is suitable for diagnosis, comprising: (a) determining the coverage of RBCs in a blood smear or sub-section of a blood smear based on the spacing and distance between the RBCs and neighboring cells; (b) determining the hemoglobin distribution of RBCs in a blood smear or sub-section of a blood smear; (c) determining the presence of non-RBC cells and / or abnormal RBCs in a blood smear or sub-section of a blood smear; and (d) in step (i), the distance between an RBC and a neighboring cell is approximately 1.0 to 2 times the diameter of the RBC. The present invention relates to a method for detecting a diagnostically usable monolayer of red blood cells (RBCs) in a blood smear or blood smear subdivision, comprising the steps of: (ii) in step (b), the area coverage is determined by a 0x ratio; (ii) in step (b), the central clear region occupying approximately 25-40% of the RBC diameter is identified within the RBC; and (iii) in step (c), non-RBC cells and / or abnormal RBCs are not identified, and / or identified non-RBC cells and / or abnormal RBCs are labeled and excluded from further analysis (also known as a parameter-based method in certain circumstances), and assigning a label (classification) of diagnostic usability to a blood smear or blood smear subdivision.

[0035] Therefore, the method involves a stepwise procedure that includes at least an analysis of geometric and physical properties, and only by combining these can a final conclusion be drawn regarding the diagnostic usability of the dependent blood smear. Thus, even if the RBC coverage is within a predetermined range limit of diagnostic usability, it may be unusable for diagnosis if there is no adequate hemoglobin distribution. Furthermore, even if the coverage and hemoglobin distribution are within a predetermined range limit of diagnostic usability, the presence of abnormally formed RBCs or non-RBC cells may render the RBC monolayer unusable for diagnosis.

[0036] The term “red blood cell monolayer region usable for diagnostic purposes,” as used herein, refers to a region of a blood smear that has a form and appearance that allows for diagnostic analysis. For example, a monolayer region contains no overlapping cells or only minimal amounts of overlapping cells, and the hemoglobin distribution of the prepared red blood cells shows a homogeneous and clearly visible pallor region in the central area of ​​each RBC, preferably in accordance with the cellular image standards of medical textbooks. While we do not wish to be bound by theory, the optimal evaluation region is considered to be between the dense region or “heel” of the smear, i.e., the zone where the blood droplet was initially placed and spread, and the extremely dilute “tapered end.” In such an ideal region, microscopically, the RBCs are uniformly distributed one by one, with only a few cells in contact or overlapping, and have a normal biconcave appearance (i.e., a central pallor). Areas of the film that are too thin, appearing to have holes or to be flat, large, and distorted RBCs, are unsuitable for diagnostic analysis. Conversely, areas with too many cells, i.e., "too dense," or where the RBCs are distorted due to cells overlapping each other in a rouleaux-like pattern, are also unsuitable for diagnostic analysis. Further details will be known to those skilled in the art or can be found in appropriate textbook references such as Rodak's Hematology: Clinical Principles and Applications, 6th edition, 2019, edited by Saunders.

[0037] The term "red blood cell" refers to erythrocytes, preferably fully grown red blood cells containing hemoglobin. It does not include other blood cells such as white blood cells or platelets.

[0038] The term “blood smear” as used herein refers to any suitable specimen of peripheral blood as known to those skilled in the art. The blood smear is preferably obtained using any method aimed at producing a cell monolayer as known to those skilled in the art. In one embodiment, the blood smear is obtained as a wedge smear on a microscope slide. Typically, a droplet of blood (e.g., 2-3 mm in diameter) obtained from a subject or patient is placed on the center line of a microscope slide (typically 75 × 25 mm in size and about 1 mm thick) about 1 cm from one end, using a stick or glass capillary tube. The blood may be obtained directly or processed (e.g., anticoagulated with EDTA). Next, a spreader slide is used to spread the blood from end to end of the contact line between the two slides. Accordingly, the spreader slide is moved along the specimen slide to produce a smear. The smear may then be air-dried or fixed with methyl alcohol. Furthermore, the smear may be hematologically stained using Romanovsky-type staining methods, including, for example, the May-Grünwald-Giemsa staining method, the Wright-Giemsa staining method, or the modified Wright staining method.

[0039] In further embodiments, the smear is prepared by spinning a glass slide in a spinner centrifuge. Typically, the preparation procedure begins with a drop of blood, either used undiluted or diluted in isotonic saline. The sample is then placed in a microscope slide centrifuge and spun for a short time. As a result, the slide is homogeneously coated with blood cells across its entire surface, ideally in a suitable monolayer.

[0040] In further embodiments, blood smears are prepared in a flow cell. This method involves filling a sample, such as a droplet of blood, into a fluid system, for example, a microfluidic chip or as part of a microfluidic system. The blood droplet is then transferred into an area of ​​a microfluidic system designed for optical inspection. The microfluidic cell can be manipulated by stopping the fluid flow when the blood reaches the area of ​​optical inspection, or by allowing the cells to settle to the bottom of the microfluidic cell. If there are more cells to be measured or the area to be measured is larger, the flow cell containing the settled blood cells can be mechanically scanned primarily through the field of view of the optical system. Also, because the flow rate of cells across the area of ​​optical inspection is slow, there is no movement of the relevant cells during the exposure time of the imaging system's camera, and the effects of large subject blur are avoided. Between successive exposures, a considerable portion of the cells can move out of the field of view of the optical system, and because the fluid flow is slow, new cells come into view. Once the blood cells are imaged, the blood is washed out of the flow cell, and the flow cell can then be washed and used for a different sample or another blood droplet from a different patient.

[0041] This method can be performed using blood smears under current microscopic observation, or using images of blood smears. The images should preferably have an azimuthal resolution and / or sampling of less than a micrometer in object space, preferably in the range of <200 nm, more preferably in the range of <100 nm, for the resulting image. The analysis can be performed on the entire blood smear, for example, on a blood smear on a microscope slide, or on any sub-section thereof, or on the corresponding image thereof. Thus, the term “sub-section of blood smear” means an area containing at least 5 to 15 red blood cells. A sub-section preferably contains more than 15 cells, for example, 50, 100, 150, 200, 300, 400, 500, 1000, 2000 or more, or any value between the listed values, preferably between 300 and 500 cells. The size and morphology of the sub-section may vary. For example, typically a size of 1000-5000 × 1000-5000 pixels can be used, and preferably, the subdivisions constitute the entire field of view of the camera used in the imaging process such that the number of pixels is equal to or approximates the number of pixels of the camera. It is preferable to divide the blood smear or the corresponding image thereof into a group of rectangular subdivisions that cover the entire specimen. Preferably, when imaging a blood smear in a scanning technique using multiple images, the subdivisions may constitute different fields of view of the camera. Thus, images of subdivisions that cover the entire blood smear or a portion thereof can also be used in the context of the present invention. In certain embodiments, when a blood smear is imaged in multiple camera images as subdivisions in a scanning procedure described herein, neighboring subdivisions may overlap to some extent, for example, 0.1, 0.5, 1, 2, 3, 4, 5%, or more, preferably so that all cells at the boundaries are sufficiently covered.

[0042] Images of blood smear specimens can be subjected to specific image processing. The term "image processing," as used herein, refers to general methods of converting an image into a digital format and performing operations on an image to enhance it and / or extract useful or desired information. The output of image processing may be a modified image or characteristics or values ​​associated with that image.

[0043] Image processing may include, for example, one or more of the following actions, functions, or procedures: image enhancement, including brightness or contrast adjustment; wavelet and multi-resolution processing, including image subdivision and pyramidal representation; compression, including techniques to reduce the storage required to save an image or the bandwidth required to transmit an image; morphological processing, including the extraction of image components necessary for geometric representation; segmentation, i.e., dividing an image into its constituent parts or objects; display of data obtained in the segmentation process; description, i.e., extraction of attributes from segmented data, including providing quantitative data that enables one class of objects to be distinguished from others; object recognition, i.e., assigning labels to objects based on their description; and stitching, i.e., generating larger continuous image parts from a group of subdivisions. These actions, functions, or procedures can preferably be performed automatically or in a programmed manner, for example, based on the use of a suitable computer program or AI module.

[0044] In the first step of this method, the area coverage of RBCs in a blood smear or sub-sub

[0045] The determination of area coverage can be performed using any suitable method or algorithm. The following procedure is preferable.

[0046] First, tight bounding boxes are assigned to the cells (target red blood cells) within the blood smear or its sub-sections. Preferably, the assignment is performed for each cell in the blood smear, or more preferably, in its sub-sections. The bounding boxes can be abstract rectangles that serve as reference points for object detection and generate collision boxes for those objects. Thus, the boundary coordinates of the boxes, i.e., the boundary coordinates of the enclosed cells, are stored and used in subsequent steps. In another embodiment, polygonal segmentation can be used for assigning cell coordinates. The assignment can be performed using a suitable algorithm or computer program, such as VGG image annotator, RectLable, or LabelImag.

[0047] In a further step, a central position is assigned to each cell. This central position is associated with the coordinates of a bounding box or polygonal segment.

[0048] In a further step, the cell area is determined. This can be done using geometric calculations based on a tight bounding box, preferably based on polygonal segments. Alternatively, the area of ​​the tight bounding box can be calculated as an approximation of the cell area. In a further embodiment, the cell diameter can be determined. The diameter can be determined as the minimum and maximum diameters for asymmetrically formed cells, or as the average diameter when the cell diameter is detected 360° in a measurement step of, for example, 1-5°.

[0049] In a separate step, the roundness index of the cell can be determined. The roundness index can preferably be calculated as the ratio of the tight bounding box or the long axis of the cell to the tight bounding box or the short axis of the cell.

[0050] Depending on the circumstances, further appropriate geometric parameters of the cell can be determined. Additional geometric parameters may include, among other things, classifying the cell into a given geometric object such as a circle, ellipsoid, triangular, or quadrangular shape, as described herein, the cell perimeter, the presence and number of small cell elongations, as well as the ratio of the cell diameter to area or the ratio of the cell perimeter to area, as well as the Hu moment, the integral of the cell's local cellular curvature, the absolute value of curvature, the maximum or minimum curvature value, or the range of measured curvature values.

[0051] In a further step, one or more directly adjacent cells are identified. This identification can be performed within a predetermined inspection radius. For example, any cell can be considered an "adjacent cell" if it can be found within a radius of 1 to 10 times the average cell diameter, e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9 times the average cell diameter, or any value between the listed values. The inspection radius can be defined relative to the nearest point within the cell's tight bounding box, or to the previously determined center position of the cell. For adjacent cells, information regarding the coordinates of the tight bounding box and the definition of the center position may already be available. Alternatively, for adjacent cells that have not yet been analyzed, one or more of the steps defined above for the cell of interest, preferably the assignment of a tight bounding box and the definition of the center position, can be performed.

[0052] For each direction, for example, within a range of 1 to 5° in a 360° area around the cell or bounding box, one directly adjacent cell is selected, preferably according to distance approximation.

[0053] Next, the distance between the target cell and directly adjacent cells in a specific direction is determined. This distance determination can be performed, for example, as a bounding box or as vector subtraction based on the coordinates of the cell center position. It is preferable to start the distance determination from the cell center position of the cell and perform it up to the cell center position of the second cell.

[0054] The selection and subsequent distance determination can be iterated for each direction or most directions, for example, for each of 1 to 20°, preferably for each of the ranges of 5°, 10°, 15°, or 20°, or for any value between the values ​​given within a 360° range around the cell or bounding box. It is particularly preferable to focus the distance determination on directly adjacent cells. That is, cells that are in the direct vicinity of the cell of interest and are at a radius farther than the directly adjacent cells, even if they are connected in a particular direction by an unbroken line forming a right angle, are excluded. Thus, these cells can be excluded from the determination. The iteration can be performed until all of the directly adjacent cells within the given radius, or 50 to 95% of them, for example, 50, 55, 60, 65, 70, 75, 80, 90, or 95%, are analyzed in terms of their distance to the cell of interest. In certain embodiments, for example, if a shorter inspection radius does not yield suitable directly adjacent cells or a suitable, unbiased group of directly adjacent cells, distance measurements and their repetitions may be performed for one or more different inspection radii.

[0055] In the final step, the blood smear, preferably a sub-section thereof, is classified as either within or outside a predetermined range limit of diagnostic usability for the RBC monolayer site. In one embodiment, the classification may be based on a distance parameter, i.e., the distance between the RBC of interest and the cells directly adjacent to it. Preferably, if the distance between the RBC of interest and the cells directly adjacent to it is about 1.0 to 1.5 or 2.0 times the RBC diameter determined in the diameter measurement step defined above, and at least 65% of the measured distance, e.g., 65, 70, 80, 85, 90, 95, 97, 98, 99%, is within the range of 1.0 to 1.5 or 2.0 times the RBC diameter determined in the diameter measurement step defined above, the blood smear, more preferably a sub-section thereof, can be identified as being within a predetermined range limit of diagnostic usability for the RBC monolayer site. It is particularly preferable to measure the distance from the cell center position of a second cell to the cell center position of a second cell, as defined herein.

[0056] In further embodiments, classification can be used to group more classified areas, for example, densely packed cell areas, monolayer areas within a predetermined range limit of diagnostic usability as defined above, and low-density cell areas, where densely packed and low-density cell areas are not within the predetermined range limit of diagnostic usability. A "densely packed cell area" is given if the measured distance is within or less than 0.7 to 1.0 times the RBC diameter as defined above, and if more than about 5%, about 10%, about 15%, about 20%, about 25%, about 30%, or more of the measured distance is within the range of 0.7 to 1.0 times the RBC diameter determined in the diameter measurement step as defined above. A "low-density cell region" is given if the measured distance is within 1.5 or 2.0 to 10 times the RBC diameter defined above, or greater than that, or if at least 65% of the measured distance, e.g., 65, 70, 80, 85, 90, 95, 97, 98, 99%, is within 1.5 or 2 to 10 times the RBC diameter determined in the diameter measurement step defined above. As an alternative to the binary classification of usable and unusable for diagnostic purposes, the classification may include, as appropriate, the states of monolayer, low-density cell region, and densely packed cell region, with usability following the order of monolayer > low-density cell region > densely packed cell region.

[0057] In a preferred embodiment, classification can be performed on several RBCs within a sub-section of the blood smear, for example, 50, 100, 200, 300, 400, 500, 1000, 2000 or more RBCs, or on all RBCs in the sub-section. If, in the above group analysis, at least about 75% of the RBCs are within the diagnostic usability limits, the analyzed sub-section can be classified as being within the diagnostic usability limits. In a particular embodiment, the sub-section may be expanded or moved to different areas of the blood smear. Measurement results obtained in such additional areas may be added to the results of the already analyzed sub-section, or they may be calculated individually for each sub-section. If a sub-section is identified as not being within the diagnostic usability limits, further expansion of the analysis area may be stopped. Thus, sub-sections within the range limits can be labeled as potentially diagnostically usable monolayer areas. If the expansion of the analysis area is stopped, it may preferably continue in adjacent or different sub-sections and will be expanded if it is found that the RBCs have a potentially diagnostically usable monolayer. Otherwise, the analysis can be continued in further adjacent or different sub-sections, etc. In a further embodiment, as described above, the observation that at least about 75% of the RBCs are within the limits of diagnostic usability can be used as an indicator of blood smear quality.

[0058] In yet another embodiment, the classification of a blood smear or its subdivisions can be performed according to statistical means such as a histogram, probability function, or density function. The classification may be based on the calculated numerical values ​​of the intercellular distances described above, in particular, the cumulative value of intercellular distances in a defined area, e.g., a subdivision of the blood smear or the entire blood smear. Thus, the numerical values ​​of the intercellular distances are calculated and compared with a cutoff value, and if the calculated numerical value is less than the cutoff, it indicates that the blood smear or its subdivision is not within the range limits of diagnostic usability for a monolayer site. For example, the cutoff may be that at least 65% of the measured distances between directly adjacent cells, e.g., 65, 70, 80, 85, 90, 95, 97, 98, 99%, are within the range of 1.0 to 1.5 or 2.0 times the mean RBC diameter, or the 60th, 75th, 80th, or 85th percentile of the measured RBC diameter.

[0059] Adjacent sub-sections of a blood smear can be analyzed individually, and the resulting cumulative distance values ​​can then be added together.

[0060] The steps in the above procedure for determining regional coverage can be performed in any appropriate order. Some steps may be omitted or repeated depending on the need and the condition of the blood smear specimen to be analyzed.

[0061] Following or in parallel with the elucidation of regional coverage, the hemoglobin distribution of RBCs is determined as a further important parameter for detecting monolayer RBC regions usable for diagnosis. The determination of hemoglobin distribution within RBCs is primarily based on the determination of the central luminous region. The term “central luminous region,” as used herein, refers to a less prominent, i.e., not sufficiently reddish, or less absorbent and brighter region within the central part of an RBC, as observed under standard perpendicular optical observation, in color representation or grayscale intensity representation, respectively. This central part or central luminous region typically occupies about 25–40% of the total cellular area or cell diameter; in certain pathological cases, this value may be lower, i.e., less than 25% of the total cellular area or cell diameter, or higher, i.e., more than 40% of the total cellular area or cell diameter. In addition, the morphology of the central luminous region is typically related to the typical biconcave cellular shape of the RBC. In addition, this shape can be used to identify cells that have a central clear area.

[0062] As a first step in determining the hemoglobin distribution of RBCs, a tight bounding box is assigned to the cells, preferably to each cell, as described above herein.

[0063] Furthermore, the contour lines of the cell boundaries are determined. The term "contour line," as used herein, refers to a function of at least two variables that reflect the points of the curve. The contour lines can be determined by transforming the RBCs into mathematical functions in segmentation and subsequent tight bounding boxes.

[0064] As an additional step, the centroid of the red blood cell is determined. The term "centroid," as used herein, refers to the average position of the cell's geometric properties, i.e., the cell's virtual weight. For example, the centroid can be calculated using each pixel enclosed by the cell's contour lines. In further embodiments, the calculation can be performed using the grayscale or absorbed values ​​of the pixels.

[0065] The determination of the centroid can be performed in any suitable way. It is preferable that the determination be performed as a calculation of the Hu moment of the corresponding cell image. The term "Hu moment," as used herein, refers to a value calculated in an image transformation. It is generally understood to represent the geometric properties of an object (e.g., the centroid as defined above). Typically, the Hu moment is considered to include seven values ​​calculated using a central moment that is invariant to image transformations, such as equilibrium translation, scaling, rotation, or reflection. In certain embodiments, each pixel of the image inside the contour lines defined above is assigned a value of 1, and each pixel outside the contour lines is assigned a value of 0. Further details can be found in Huang and Leng, 2010, Proceedings of 2010 2. nd This can be found in appropriate sources such as the International Conference on Computer Engineering and Technology (ICCET), pp. 476-480, Chengdu, China.

[0066] Alternatively, an algorithm that uses only the pixels along the contour lines can be used to improve the accuracy of the calculation.

[0067] In a further step, the hemoglobin distribution parameter (HDP) is determined. HDP can typically be determined based on the intracellular pixel hemoglobin distribution HDM2. Its calculation is typically given by equation (1):

number

[0068] Therefore, equation (2) I(i,j)=I 画素、計算値 (2) Based on this, a pixel-specific value representing the hemoglobin distribution can be calculated for each pixel, and this value can be zero for all pixels outside the cell.

[0069] Furthermore, the averaged hemoglobin content (HDM) of the cell is the average hemoglobin value obtained by averaging the hemoglobin levels across all pixels inside the cell. 基準値 This can be determined. The calculation is typically done using equation (3):

number

[0070] The HDP value, or hemoglobin distribution characteristic value, is given by the following equation (4): HDP = HDM2 / HDM 基準値 (4) It is calculated according to [the formula].

[0071] In a typical embodiment, a homogeneous hemoglobin distribution is indicated by an HDP value close to 1. For biconcave cells with a central light region, the HDP value can be greater than 1 and up to 2, preferably in the range of 1.5 to 2.0. Since the HDP value is particularly morphological, considering it as the sole indicator of a usable monolayer structure can lead to misjudgment. Advantageously, using the HDP values ​​defined herein, it is still possible to distinguish between cells that are biconcave and those that are not. In certain embodiments, a particular single cell within the analyzed monolayer may be hyperpigmented and, for example, may have an HDP outside the indicated range. If a directly adjacent neighboring cell shows an HDP within the above range, that RBC monolayer region can still be considered usable for diagnosis.

[0072] Advantageously, HDP values ​​can be normalized relative to the analyzed cells, thus minimizing deviation from the general hemoglobin reference range. Otherwise, the deviation would directly affect the HDP value, thus reducing HDP's ability to distinguish between different types of hemoglobin distributions.

[0073] HDP value determination can be performed on all or a subset of red blood cells in a blood smear or a sub-section of a blood smear. In certain embodiments, HDP value determination is limited to areas already identified as having area coverage within the diagnostic usability limits defined above. In further embodiments, the average HDP value of all cells in a blood smear or a sub-section can be used as an indicator of blood smear quality by reflecting the number of RBCs that retain both concave shapes.

[0074] As an optional step, the hemoglobin content in red blood cells can be determined. While we don't want to be bound by theory, considering the fact that hemoglobin absorbs light as it passes through cells, it can be assumed that lower pixel readings in the cell image may indicate a higher hemoglobin concentration.

[0075] This allows for further determination of the centroid. For example, the pixel reading scale can be inverted by subtracting the pixel reading value from the average background value, which represents values ​​without absorption. This method is given by equation (5) I 画素、計算値 =background 基準値 -I 画素、測定値 (5) It can be followed, in the formula, background 基準値 The value may be an arbitrary specific value valid only for individual, specific samples, or it may be a general value used as a background criterion.

[0076] This value will also be determined for samples from clear, i.e., non-cellular areas of a blood smear, for example. Alternatively, it may be set as a fixed parameter based on, for example, previous measurements or database values.

[0077] In another embodiment, the exponential nature of absorption can be used as the basis for calculating the Hu moment. Thus, equation (6) I(x)=I0*index(-αx) (6) The effective absorption coefficient αx can be derived from this, and this can be considered as the value of normal hemoglobin content in red blood cells.

[0078] In a preferred embodiment, the calculation of Hu moments can be used as a measure of the hemoglobin (Hb) content of cells, such as red blood cells. Thus, Hu moments with, for example, absorption-related values ​​can be used as a possible representation of the hemoglobin content of red blood cells.

[0079] In a particularly preferred embodiment, scale invariance of the Hu moment can be achieved by normalizing the Hu moment to a reference moment that is similarly affected by size.

[0080] Advantageously, the Hu moment's function of being scale-invariant with respect to the image improves the accuracy of the method because it allows us to ignore typical, unavoidable small variations between different imaging systems. For example, the effective magnification may differ slightly between systems, but no correction is required in the context of this invention.

[0081] In the final step of determining the hemoglobin distribution, the central clear region is classified according to the determined HDP. This classification allows for labeling of cells as being within or outside the limits of diagnostic usability. Preferably, if the measured HDP is approximately 0.5–2, the cells are considered to be within the limits of diagnostic usability. If the HDP value is less than 0.5 or greater than 2.0, the cells are considered to be outside the limits of diagnostic usability. If the HDP value is greater than 1, preferably greater than 1.3, greater than 1.5, or greater than 2.0, the cells can be characterized as biconcave and not hyperpigmented. On the other hand, RBCs with an HDP of less than 1 can be characterized as hyperpigmented.

[0082] The steps of the above procedure for determining hemoglobin distribution and hemoglobin content can be performed in any suitable order. Some steps may be omitted or repeated depending on the need and the condition of the blood smear to be analyzed.

[0083] Following, or in parallel with, the determination of regional coverage and hemoglobin distribution and, if applicable, hemoglobin content, the presence of non-RBC cells and / or abnormal RBCs in the blood smear or its sub-sections is determined as a further important parameter for the detection of RBC monolayer sites usable for diagnosis.

[0084] As a first step in determining the presence of non-RBC cells and / or abnormal RBCs in a blood smear or subdivision thereof, a tight bounding box is assigned to the cells, preferably to each cell, as described above herein.

[0085] Furthermore, as described above in this specification, the contour lines of the cell boundaries are determined.

[0086] In addition, as described above in this specification, the hemoglobin distribution parameter (HDP) can be determined.

[0087] As an additional parameter to be determined, the present invention preferably assumes the determination of the cell diameter, as described above herein.

[0088] Further parameters to be determined are the curvature of cells in the blood smear or its subdivisions. The term “cell curvature,” as used herein, refers to the curvature of the cell membrane to accommodate various cell morphologies, particularly in the outer contour lines of cells in the smear, and is expressed as a geometric measure. Curvature is typically measured as two principal curvatures c1 and c2 that characterize the shape of the cell membrane in three-dimensional space. The principal curvatures are typically given by equations (7) and (8). c1 = 1 / R1 (7) c² = 1 / R² (8) It is determined according to the formula, where R1 is the radius of the corresponding circular segment 1 and R2 is the radius of the second corresponding circular segment 2.

[0089] The circular segments can be determined by locally fitting a quadratic function or a sphere to the cell's contour. The curvature of each segment of the contour is then determined. Alternatively, a spline function can be locally fitted to the cell's contour. The curvature is then determined as the second derivative of the spline function at a given location. Further details will be known to those skilled in the art or can be obtained from a suitable geometry textbook.

[0090] Based on the above parameters or a subset thereof, classification of any cells in a blood smear or its subdivisions can be performed. Classification makes it possible to distinguish between normal, and therefore diagnostically usable, red blood cells and either non-RBCs or abnormal RBCs. Typically, if the cell diameter is >8.5 μm or <6 μm, the cells are considered to be in the non-RBC or abnormal RBC group. A further criterion that can be used as an addition or alternative is that the HDP is <0.5 or >2.0. Yet another criterion that can be used as an addition or alternative is that the curvature value is >1 / 2 μm, >1 / 2.5 μm, or preferably >1 / 3 μm. In preferred embodiments, cells are considered to be in the non-RBC or abnormal RBC group if the cell diameter is >8.5 μm or <6 μm and the HDP is <0.5 or >2.0; or if the cell diameter is >8.5 μm or <6 μm and the curvature is >1 / 3 μm; or if the HDP is <0.5 or >2.0 and the curvature is >1 / 3 μm. In further particularly preferred embodiments, cells are considered to be in the non-RBC or abnormal RBC group if the cell diameter is >8.5 μm or <6 μm and the HDP is <0.5 or >2.0 and the curvature is >1 / 3 μm.

[0091] According to the above classification, it becomes possible to identify the morphology of certain abnormal cells. For example, teardrop cells, arkantocytes, or urchin-like cells can be identified, and the areas where these cells are found can be classified as RBC monolayer areas that are unsuitable for diagnosis in blood smear specimens.

[0092] In certain embodiments, additional parameters such as an indentation index can be determined. For example, if cells show an increase in the number of indented or convex contour segments, these are also considered non-RBCs or abnormal RBCs. Typically, indented areas in sickle cells or arcantoid cells, for example, have negative curvature, i.e., they indicate concave regions. Therefore, if the curvature is determined along the cell contour, the curvature value can be integrated along the cell contour, and the integral of the absolute value of the curvature can be determined as an additional step. The integral of the signed curvature value can be divided by the absolute value integral. Thus, a resulting value of 1 indicates that the cell has no indented areas. The further the calculated value is from 1, the greater the proportion of indented areas in the cell. The exact value may vary depending on, for example, the shape of the cell. Preferably, a cell can be considered indented if the value of the indentation index calculated as described above is less than 1, preferably less than 0.95, less than 0.9, or more preferably less than 0.8.

[0093] This parameter can be used in addition to one or more of the parameters defined above, such as diameter, HDP, and curvature.

[0094] The steps in the above procedure for the presence of non-RBCs or abnormal RBCs can be performed in any appropriate order. Some steps may be omitted or repeated depending on the need and the condition of the blood smear to be analyzed.

[0095] The present invention assumes that a monolayer RBC site in a blood smear or sub-section of a blood smear can be classified as diagnostically usable if there are no non-RBC cells and / or abnormal RBCs, or if the number of non-RBC cells and / or abnormal RBCs is small, for example, less than 15%, less than 10%, less than 20%, preferably 5% or less. Accordingly, in certain embodiments, the ratio of non-RBCs and / or abnormal RBCs to RBCs or normal RBCs can be determined and compared to a cutoff value, and if the calculated value is less than the cutoff, it indicates that the blood smear or sub-section is not within the limits of the range of diagnostic usability of the monolayer site. For example, the cutoff could be that at least 80% of the cells, for example, 80, 85, 90, 95, 97, 98, or 99%, are RBCs and / or non-abnormal RBCs. The presence of non-RBCs and / or abnormal RBCs, preferably a large number of non-RBCs and / or abnormal RBCs, for example, more than 10%, preferably more than 15%, or more, can lead to classifying a monolayer of RBCs in a blood smear or sub-section of a blood smear as unsuitable for diagnosis.

[0096] In certain embodiments, a particular sub-section or region of a blood smear may be excluded from further analysis if a certain percentage of non-RBC cells or abnormal cells are found within it, and may be correspondingly labeled, for example, in an image metafile or annotation. The cutoff value for excluding a sub-section of a blood smear is preferably greater than 15%, more preferably greater than 10%, of cells that are non-RBC and / or abnormal RBCs.

[0097] Based on the results of the above procedure regarding regional coverage, hemoglobin distribution and, if applicable, hemoglobin content, as well as the presence of non-RBCs or abnormal RBCs, a blood smear or its sub-sections can be labeled for diagnostic usability (classification). Labels may be applied to the entire smear or to a portion thereof. For example, if a particular sub-section contains unusable coverage, unusable hemoglobin distribution, or non-RBCs or abnormal RBCs, it may be labeled accordingly.

[0098] In further embodiments, the present invention envisions the specific exclusion of non-RBC cells and / or cells identified as abnormal RBCs, as well as the corresponding regions in which these cells are located. This exclusion can be implemented as a digital storage of cell coordinates or region coordinates, which are then used as exclusion zones for further diagnostic analysis.

[0099] In further embodiments, the presence of areas found to be unusable for diagnosis can also be digitally labeled, for example, by their coordinates, and excluded from diagnostic analysis as prohibited zones.

[0100] In yet another embodiment, the presence of regions found to contain cells with hemoglobin distributions deemed unsuitable for diagnostic use can also be digitally labeled, for example, by their coordinates, and excluded from diagnostic analysis as a prohibited zone.

[0101] All prohibited zones or explicitly excluded cells or regions are incorporated into the overall summary of, for example, the entire blood smear or its sub-sections.

[0102] In further embodiments, cells identified as non-RBCs and / or abnormal RBCs are digitally stored based on their coordinates within a blood smear or sub-section thereof, or preferably, they are labeled in an image of the blood smear or sub-section thereof. These cells can then be used for specific analytical or diagnostic approaches, namely, for diagnosing diseases, for example, based on the counting of such cells, based on the precise determination of the morphology of these cells, and optionally based on comparison with controls or database entries or literature data or expert opinions regarding the quantity or morphology of these cells. The term “disease,” as used herein, refers to any disease relating to blood cells and detectable using them, e.g., anemia, cancer, e.g., leukemia, infection, thrombocytopenia.

[0103] In a further embodiment, the present invention relates to a method for detecting a diagnostically usable RBC monolayer in a blood smear, comprising: (a) acquiring a number of optical images of the blood smear; (b) classifying each image on the basis of the presence of a diagnostically usable monolayer region, performed either (i) using an artificial neural network or (ii) according to a parameter-based method described herein; (c) generating a heatmap of the images according to the presence of a diagnostically usable monolayer region; (d) identifying a region of a diagnostically usable monolayer region, where adjacent images in the heatmap generated in step (c) are analyzed; (e) verifying the diagnostically usable monolayer region identified in step (d) by calculating the density configuration of the selected images; and (f) combining all adjacent images of the diagnostically usable monolayer region to form a monolayer panoramic overview image. This method should be considered as a partially alternative approach to detecting a diagnostically usable RBC monolayer in a blood smear. Consequently, it places more emphasis on image recognition and the evaluation of appropriate training data (sometimes referred to as an image-based method).

[0104] In the first step, an optical image of the blood smear specimen described above in this specification is acquired. The optical image may be an image of a small section of the blood smear. Therefore, in order to cover the entire blood smear specimen, a number of optical images are acquired. The sizes of the optical images vary. It depends especially on the size of the blood smear, the microscope equipment or optical equipment used, and the resolution of the camera. It is preferable that one optical image contains or shows about 100 to 2000 cells. The entire blood smear specimen can cover an area of, for example, about 50 to 100 mm 2 and may be divided into a number of small sections and corresponding images, preferably 40 to 200 small sections and corresponding images. The number of images may depend on the equipment used, such as the magnification, resolution or field of view of the optical equipment.

[0105] According to a certain specific embodiment of the present invention, the image should preferably have a resolution of 80 to 100 nm per pixel, which is considered suitable for digital scanning. Based on this value and the nature of the optical equipment, the number of images can be determined. As an example, it can be calculated as follows: Assuming the pixel size or pixel pitch is about 4 μm, a magnification of 50× is required. For an analysis of about 1 μm 2 , the number of pixels should be given as 100 (˜150), while for an analysis of 100 mm 2 , the number of pixels should be given as 10 (˜15) Gpxl. Depending on the optical equipment, the number of photos can be estimated. For example, if the optical equipment has a resolution of 5 Mpxl, a number of about 2000 or more images are required, while at a resolution of 20 Mpxl, about 500 or more images are required.

[0106] In a typical embodiment, an optical device with a camera sampling or resolution of 5 to 25 Mpxl can be used.

[0107] Images can be acquired by a scanning process that moves a camera or optical device along a predetermined row-and-column coordinate system on the blood smear. Images may overlap in any direction, or preferably only in two directions, such as the right and top sides of a rectangle. The overlap may be, for example, 0.05, 0.1, 0.5, 1, 2, 3, 4, 5, 10, 15, 20, 40, 50%, or more. Alternatively, the images may not overlap. Non-overlapping images may preferably have continuous boundaries to cover all elements of the blood smear specimen.

[0108] It is even more preferable that the images are supplied in a sequential system. For example, starting from the upper left corner of the blood smear on the microscope slide, moving to the upper right corner, then to the second row from the left of the blood smear on the microscope slide, moving to the right, and so on, the final image being the image of the lower right corner of the blood smear on the microscope slide. In further embodiments, different appropriate order or sequence may be contemplated. Furthermore, the morphology, size, or resolution of the images, etc., can be adapted to the optical instrument, microscope, or laboratory equipment used.

[0109] In a further preferred embodiment, the acquired image can be controlled based on quality parameters. The corresponding image quality evaluation may include determining one or more of the following parameters: sharpness, i.e., the amount of detail in the image; noise, i.e., irregular fluctuations in image density; dynamic range, i.e., the range of light levels captured by the optical instrument; gradation reproduction, i.e., the ratio of luminance to emission; contrast, i.e., the slope of gradation reproduction; color accuracy; distortion aberration; vignetting; chromatic aberration; or color moiré, i.e., artificial color banding. The quality evaluation can be performed using any suitable method, for example, a full-reference method in which the image is compared to a reference image of a predetermined quality, or a metric-based method in which the extracted metric configuration is compared, for example, based on a reference image of a predetermined quality. The full-reference method is preferred.

[0110] Numerous images intended for use in the next evaluation step can preferably be processed as described above herein.

[0111] In the first evaluation step, each image is classified in relation to the presence of a single layer region usable for diagnosis. This classification can typically be based on a morphological comparison of the image with a repository of images of single layers regions classified independently by experts. The term "morphological comparison," as used herein, refers to the act of comparing and contrasting with respect to one or more of the characteristics of cells or groups of cells present in the image, such as the presence of structures, shapes, forms, sizes, patterns, optical / visual representation, contrast, color density, or staining density.

[0112] The comparison typically involves comparing and contrasting images, preferably processed as described herein, with images or data previously classified by experts as either usable or unusable for diagnostic purposes and stored in an image repository along with classification information. These stored images may have undergone further image processing similar to or identical to that performed on images of blood smear specimens obtained according to the method of the present invention.

[0113] The morphological comparison described above preferably involves applying the acquired images of the blood smear specimens to a machine learning-based method. When the concept of “machine learning” is used within the context of the present invention, it typically relies on a two-step approach: firstly, a training phase; and secondly, a prediction phase. In the training phase, the values ​​of one or more parameters of a machine learning model (MLM) are set using training techniques and training data. In the prediction phase, the trained MLM operates based on measurement data. Examples of MLM parameters include: the weights of neurons in a given layer of an artificial neural network (ANN), such as a convolutional neural network (CNN); the kernel values ​​of a classifier kernel, and so on.

[0114] The construction of the MLM may include a training phase to determine the values ​​of each parameter. Training is particularly preferably performed using images from a blood smear image repository, especially single-layer images identified in the blood smears. The images in the repository are typically independently classified by experts. These experts may be, for example, pathologists, histologists, or hematologists. In a further embodiment, the experts may be a group of experts who coordinate or harmonize their subjective classifications, for example, a group of pathologists, histologists, or hematologists, or a mixed group of various experts. The corresponding results, namely labeling classifying blood smear region images into groups of RBC single-layer regions usable for diagnosis or groups of RBC single-layer regions / blood smear regions unusable for diagnosis, are subsequently stored in the image repository, for example, along with the images. This information can be retrieved and used by the MLM as a training set. In certain embodiments, the training dataset is optimized based on, or supplemented with, the parameters and / or thresholds of the parameter-based methods described herein. This may, advantageously, lead to further optimization of the overall classification process.

[0115] Building an MLM generally involves determining the values ​​of one or more hyperparameters. Typically, the values ​​of one or more hyperparameters in the MLM are set and remain unchanged during the training phase. Therefore, the values ​​of hyperparameters can be changed in the iterations of the outer loop, while the values ​​of the MLM parameters can be changed in the iterations of the inner loop. In some cases, there may be multiple training phases, for which multiple values ​​of one or more hyperparameters can be tested, or even optimized. The performance and accuracy of most MLMs are strongly dependent on the values ​​of the hyperparameters.

[0116] Examples of hyperparameters include: the number of layers in the convolutional neural network; the kernel size of the classifier kernel; the input neurons of the ANN; the output neurons of the ANN; the number of neurons per layer; and the learning speed.

[0117] In the context of the present invention, various types and kinds of MLMs can be used. For example, novelty detector MLMs / anomalous detector MLMs, or classifier MLMs, such as binary classifiers, can be used. For example, deep learning (DL) MLMs can be used: in this case, the configuration detected by the DL MLM may not be predefined but rather can be set by the values ​​of each parameter of the model that can be learned during training.

[0118] Multiple techniques can be used to build an MLM. For example, the type of training may vary depending on the type of MLM. Furthermore, the type of training used may vary across different implementations. For instance, iterative optimization might be used, which employs optimization functions defined for one or more error signals.

[0119] The use of a pre-trained ResNet50 convolutional neural network, which has undergone initial transfer learning and subsequent fine-tuning, is particularly preferred.

[0120] Based on the classification of RBC monolayer regions into those usable for diagnosis and those unusable for diagnosis (RBC monolayer regions / blood smear regions), a heatmap is generated for each image according to the classification result in a further step, and two possible results are shown using appropriate encoding, such as color coding.

[0121] In the next step, a region of the RBC monolayer available for diagnosis is identified. This identification step includes analyzing heatmaps of adjacent images. For example, after analyzing an image, a neighboring or adjacent image to its right or left is analyzed. If the heatmap of the further analyzed image matches that of the initially analyzed heatmap, the region of the RBC monolayer available for diagnosis may be extended to the further analyzed image. In a preferred embodiment, this procedure can be performed as a sliding window procedure across all adjacent images. The window size of the sliding window can have any appropriate value. It is preferable to use a window size of 5×5, 6×6, 7×7, 8×8, 9×9, or 10×10 images. The use of an 8×8 image is particularly preferred.

[0122] In certain embodiments, the sliding window heatmap analysis procedure is primarily designed to select and order a group of images, for example, 8x8 images = 64 images, according to their classification status. For example, a group where all are classified as blood smear regions usable for diagnosis is ordered to the top of the list, followed by a group containing heatmaps of 63 images classified as blood smear regions usable for diagnosis, and so on. This procedure is performed for all images derived from the blood smear. Images at the same position in the ordering list are preferably grouped into groups or buckets. Next, using a non-maximal suppression algorithm, overlapping areas are typically removed individually for each bucket, retaining only the non-overlapping areas. The final assortment of non-overlapping areas can be grouped if it includes consecutive images. The resulting number of grouped areas can be adapted to the user's needs. For example, the user can pre-define how many areas within the blood smear should be defined.

[0123] In further specific embodiments, additional analytical steps can be performed to filter out contacting, locally concentrated monolayer regions that may cause duplicate cell counts during subsequent analysis. This step involves using a graph interpretation-based algorithm, where images classified as usable for diagnosis are considered nodes in a graph, and edges represent the Euclidean distance between their centers. The presence of contacting monolayer regions can be determined using a conversion to an adjacency matrix calculation. Identifying contacting regions may lead to the exclusion of images. Alternatively, identified images may be labeled with data, preferably containing information about the precise division of the contact.

[0124] In a further step, the already identified monolayer regions available for diagnosis are validated by calculating the density configuration of the selected images. The term "density configuration," as used herein, refers to a probability function representing the density of a continuous random variable between a specific range of values. The analysis may be based on calculations using segmented field images belonging to a single layer. Furthermore, the results of the machine learning heatmap generation disclosed above are validated using a support vector machine classifier trained on the same dataset as the ANN described above. The validation step is preferably used to eliminate any errors that may have occurred during the machine learning phase. Further details can be found in appropriate literature sources, such as Jahanifar et al., 2016, 23rd Iranian Conference on Biomedical Engineering and 2016, 1st International Iranian Conference on Biomedical Engineering (ICBME), 2016, pp. 128–133.

[0125] In the final step of this procedure, all adjacent images of a single-layer region available for diagnosis are combined to form a single-layer panoramic overview image. The term "adjacent images," as used herein, refers to nearby or adjacent images in any direction. However, there are no gaps or voids in the panoramic overview image. If gaps or breaks exist between images, a new panoramic overview image must be supplied.

[0126] The panoramic overview can be used for further diagnostic analysis, for example, by pathologists or histologists.

[0127] In a particularly preferred embodiment of the present invention, the classification used in the above method is a binary classification of two classes: (i) available for diagnosis and (ii) not available for diagnosis. Availability can refer to various sizes of the analysis, for example, the entire blood smear, any appropriately sized sub-section thereof, or that sub-section as reflected by the optical image defined above. Furthermore, within these sub-sections, cells or groups of cells can be classified as available for diagnosis or not available for diagnosis. When providing a binary classification, highlighting or labeling can be performed simultaneously, for example, the availability information can be stored in metadata or in a database.

[0128] In further specific embodiments of the present invention, the parameter-based method and the image-based method described herein are performed using the same number of optical images, and the results of the parameter-based method and the image-based method are compared in terms of diagnostic usability. The comparison determines, if there is overlap in results, whether the corresponding image or combination of images is either usable for diagnosis or not. If there is difference in results, the image or combination of images is reclassified as potentially usable or not usable, preferably being excluded from the group of single-layer regions usable for diagnosis. The comparison of results may also be performed using further single-layer classification methods, e.g., methods known to those skilled in the art.

[0129] In further embodiments, the detected red blood cell (RBC) monolayer region, which is available for diagnosis, is further examined by a specialist, such as a histologist or hematologist, to confirm the result.

[0130] In exemplary embodiments, a workflow for the preparation and evaluation of RBC blood smears may include the following steps, as shown in Figure 3: (1) A blood sample is taken from a subject or patient; this step is optional, and the workflow may also be started with blood already taken or blood supplied separately. (2) The blood sample is processed, for example, as described herein. (3) A blood smear is then prepared, for example, as described herein. (4) In a further step, the blood smear is scanned, and images, e.g., images of various sizes, different resolutions and / or diverse overlaps, are taken and optionally stored in a database or server. (5) Image quality evaluation is further performed, as described herein. (6) Cells are then detected, and their geometric parameters, distance from corresponding adjacent cells, and hemoglobin distribution values, etc., as described in detail above herein, are determined. (7) Based on the cell determination step (6), the blood smear site is classified as a single layer usable for diagnosis or a single layer unusable for diagnosis. (7) The classified images are analyzed to generate a heatmap showing the population or image of cells that are considered usable or unusable for diagnosis. (8) Images of adjacent single-layer regions usable for diagnosis are selected as an array to obtain a single-layer panoramic overview image. (9) The single-layer regions obtained in step (8) are analyzed for diagnostic purposes.

[0131] The workflow described above may be modified or altered in several respects. For example, the workflow may begin with images that have been previously acquired, processed, and evaluated. In a further example, the workflow may include alternative steps (6) and (7), which involve classifying images for the presence of monolayer regions using an artificial neural network trained on a suitable set of training data described herein. The workflow may further include two alternatives to steps (6) and (7), where the classification results are compared and only regions found to be consistent between the two classification procedures are used in each step of the further workflow. Alternatively, in case of inconsistency, the deviation results may be provided to a field expert, such as a histologist or hematologist. If the field expert modifies the classification, this information can be reintroduced into the current method, for example, as an improved training dataset, enabling a continuous improvement process. This step can be repeated. The inclusion of more than one field expert, for example, a group of experts with different scientific backgrounds, is also conceivable.

[0132] In further exemplary embodiments, a workflow for the preparation and evaluation of RBC blood smears may include the following steps, as shown in Figure 4: (1) an image is acquired with an optical instrument; (2) cells are detected and bounding boxes are defined as described herein; (3) cell diameters are determined; (4) immediately adjacent neighboring cells are determined and the distance to these cells is measured; (5) the geometric parameters of the cells are determined; (6) the Hu moment is determined; (7) the HDP value is determined. For steps (3) to (7), input from a field expert or a corresponding knowledge repository or database may be used for the purpose of defining, optimizing, comparing and / or verifying the configuration; (8) a classification of the image / blood smear is provided, the output of which may be binary (available or unavailable for diagnosis) or graded, for example, certain regions or sub-regions are excluded from further analysis. As an alternative process, (2a) the acquired images can be used for expert classification based on previously analyzed training data, becoming a trained neural network, which is then used for inference operations based on a test dataset derived from the images acquired in step (1). (3a) The inferred decisions are translated into image / blood smear classification, the output of which can be binary (usable or unusable for diagnosis) or stepwise, for example, certain regions or sub-regions are excluded from further analysis. (9) The classified images are analyzed to generate a heatmap showing populations of cells that are considered usable or unusable for diagnosis. (10) Adjacent images of single-layer regions usable for diagnosis are selected as an array to obtain a single-layer panoramic overview image. (11) The single-layer regions obtained in step (10) are analyzed for diagnosis.

[0133] This workflow may also be modified or altered in several respects. For example, comparisons by field experts may be omitted. The classification results can be compared, and only areas found to be consistent between the two classification procedures can be used in subsequent steps of the workflow. Alternatively, in cases of disagreement, the deviations may be provided to a field expert, such as a histologist or hematologist. The option of combining both methods to potentially improve the quality of the classification is considered a particularly advantageous configuration of the present invention.

[0134] Advantageously, the configurations relating to predetermined geometric properties / hemoglobin / cell distances described herein allow for an improved quality of information used to select blood smear sites for final analysis, thus enabling optimized selection. In addition, this information can also be used and visualized as supplementary or additional information in a heatmap display of the smear, generating a high-granularity heatmap display. Such additional data can, in certain embodiments, be incorporated into the process with full resolution, or incorporated in a classified format after being compared to threshold levels as described above herein.

[0135] In another embodiment, the present invention relates to the use of the methods described herein for selecting RBC monolayer sites usable for diagnosis. Further conceivable is the use of a combination of methods for selecting RBC monolayer sites usable for diagnosis, for example, a combination of the parameter-based and image-based methods described herein.

[0136] In yet another embodiment, the present invention relates to a method for diagnosing a blood-related disorder, comprising (i) identifying at least one RBC monolayer region available for diagnosis in a blood smear of a sample, preferably using a method according to the present invention, preferably a parameter-based method and / or an image-based method as defined herein, and (ii) identifying abnormal cells in the blood smear, preferably using a parameter-based method as defined herein. “Blood-related disorder” as used herein can be any disorder relating to an abnormal morphology of blood cells, an unusual cell in the blood, or an abnormal number of blood cells, or a combination thereof. Examples include anemia, cancer, e.g., leukemia, lymphoma, infection, and thrombocytopenia.

[0137] In further embodiments, the method may include, as an additional step or as an option, a step of staining cells with a dye to identify molecular structures within the cells. The staining step can be performed using appropriate dyes such as acridine orange, 7-AAC, calcein, CFSE, DAPI, Hoechst dye, propidium iodide, resazurin, trypan blue, or tetrazolium salts. The staining results can be compared to a control or database entry of healthy or normal cells to indicate the presence of abnormal cells or a pathological condition. Thus, the presence of abnormal cells, and optionally the presence of abnormal molecular structures, can be considered indicators of blood-related diseases. Further consideration may include providing corresponding reports for the abnormalities found and the expected diagnostic outcomes.

[0138] In a further embodiment, the present invention relates to a computer-implemented method for detecting a monolayer of red blood cells (RBCs) in a blood smear or sub-section of a blood smear that is suitable for diagnosis. The method comprises steps (a) to (d) and optionally in addition thereto, the corresponding sub-steps of the parameter-based method or steps (a) to (f) of the image-based method. The method can be implemented on any suitable storage or computer platform, such as cloud-based, internet-based, intranet-based, or on a local computer or mobile phone.

[0139] In another embodiment, the present invention relates to a data processing device including means for performing a computer-implemented method as defined above. This device includes means for performing any one or more steps of the computer-implemented methods of the present invention described above. Thus, any of the computer-implemented methods described herein can be performed in whole or in part using a computer system including one or more processors that can be configured to perform each step. Accordingly, some embodiments of the present invention relate to a computer system configured to perform a step of any of the computer-implemented methods described herein, and potentially having various components for performing each step or each group of steps. Each corresponding step of each method can further be performed simultaneously or in a different order. In addition, parts of these steps can be used together with parts of other steps of other methods. Furthermore, all or part of the steps can be optional. In addition, any step of any of the methods can be performed using modules, circuits or other means for performing these steps.

[0140] A computer program is also conceivable that, when executed by a computer, includes instructions causing the computer to perform any of the computer-implemented methods of the present invention as defined herein, or any one or more computerizable steps of the methods of the present invention described herein.

[0141] The provision of computer-readable storage media containing the computer program products defined above is also envisioned. The computer-readable storage media may be connected to a server element, reside in a cloud structure, or be connected to one or more database structures or client databases via the internet or an intranet.

[0142] Any software component or computer program or function described herein can be implemented, for example, using prior art or object-oriented techniques, as software code for execution by a processor using any suitable computer language such as Java, Python, Javascript, VB.Net, C++, C#, C, Swift, Rust, Objective-C, Ruby, PHP, or Perl. The software code can be stored as a set of instructions or commands in a computer-readable medium for storage and / or transmission, suitable media including random access memory (RAM), read-only memory (ROM), magnetic media such as hard drives, or optical media such as compact discs (CDs) or DVDs (digital multipurpose discs) and flash memory. The computer-readable medium can be any combination of such storage devices or transmission devices. Such programs can also be encoded and transmitted using carrier signals adapted for transmission over wired networks, optical networks, and / or wireless networks compliant with various protocols, including the Internet. For this reason, a computer-readable medium according to the present invention can be made using data signals encoded by such a program. Computer-readable media encoded in program code may be included with a compatible device or provided separately from another device (e.g., via internet download). Any such computer-readable media may reside on or within a single computer program product (e.g., a hard drive, CD, or an entire computer system), or on or within different computer program products within a system or network. A computer system may include a monitor, printer, or other suitable display for providing the user with any of the results described herein.

[0143] The following drawings are provided for illustrative purposes only. Therefore, they should not be construed as limiting. Those skilled in the art will clearly be able to envision further modifications of the principles described herein. [Explanation of Symbols]

[0144] 1. Microscope slide glass 2 barcodes 3 Cover 4 Blood smear 5. Areas that are scanned by light 6. Single-layer regions usable for diagnosis 10 Scanned areas 11. Inappropriate body parts 12 Appropriate single layer 13. Single-layer regions usable for diagnosis 14. Regions of adjacent 8x8 single images with appropriate monolayers. 20 blood samples collected 21 Blood processing 22 Preparation of blood smear 23 Imaging of blood smears 24 Image Quality Rating 25 Cell detection 26. Classification of single layers 27 Heatmap generation / Image selection and combination 28 Analysis of RBC smears 30 Image Capture 31 Cell detection 32 Determining the properties of cells 33 Detection and determination of distance to directly adjacent neighboring cells 34. Determination of the geometric parameters of cells 35. Determination of exemplary Hu moments for a set of custom-defined parameters and / or geometric parameters. 36 HDP decisions 37 Classification of smear specimens or their subdivisions as single-layer regions usable / unusable for diagnosis 38. Input from experts in specific fields or from knowledge databases. 39 Expert classification of already acquired or publicly available images of RBCs and monolayers. 40. Preparing training datasets for neural networks. 41 Training a Neural Network 42. Feed the trained neural network test data from the captured images. 44. Preparing for decision inference using neural networks 45. Release of the preparation model / Application of captured images based on the process model. 46 Classification of the smear or its sub-sections as single-layer regions usable / unusable for diagnosis, based on the results of step (45). 47. Generating a Heatmap 48. Image Selection and Combination 49 RBC smear analysis 50 Training Phases 51 Inference Phase 52 Usage Phases

Claims

1. A method for detecting a single layer of red blood cells (RBCs) suitable for diagnosis in a blood smear or a sub-section of a blood smear, comprising: (a) A step of determining the coverage rate of RBCs in a blood smear or sub-section of a blood smear based on the spacing and distance between the cells and neighboring cells; (b) A step of determining the hemoglobin distribution of RBCs in a blood smear or a sub-section of a blood smear; (c) A step of determining the presence of non-RBC cells and / or abnormal RBCs in a blood smear or a sub-section of a blood smear; (d) In step (a) of (i), the area coverage rate is determined by the distance between the RBC and the cells in the vicinity of the RBC being approximately 1.0 to 2.0 times the diameter of the RBC; (ii) In step (b), a central clarified region accounting for approximately 25-40% of the RBC diameter is identified within the RBC; and (iii) In step (c), if no non-RBC cells and / or abnormal RBCs are identified, and / or identified non-RBC cells and / or abnormal RBCs are labeled and excluded from further analysis, The process of assigning a label (classification) of diagnostic usability to a blood smear or a sub-section of a blood smear. A method for detecting a single layer of red blood cells (RBCs) suitable for diagnosis in a blood smear or a sub-section of a blood smear.

2. The process of determining the coverage rate of RBCs in a blood smear or sub-section of a blood smear, based on the spacing and distance between the RBCs and neighboring cells, is as follows: (a) The step of assigning a tight bounding box to each cell; (b) The step of assigning a central position to each cell; (c) the step of determining the diameter and / or area of ​​the cell or bounding box; (d) A step to determine the circularity index; (e) a step of determining further geometric parameters, if applicable; (f) The step of identifying one or more directly adjacent cells within a predetermined inspection radius; (g) the step of selecting one directly adjacent cell; (h) A step of determining the distance between an RBC and a cell directly adjacent to the RBC; (i) Repeating steps (g) and (h) for one or more further directly adjacent cells identified in step (f); (j) A step of classifying regions according to a distance criterion, wherein the distance between an RBC and a cell directly adjacent to the RBC is 1.0 to 2.0 times the diameter of the RBC, which identifies the coverage of a region usable for diagnosis, and at least 75 to 80% of the measured distances must be within the range of 1.0 to 2.0 times the diameter of the RBC determined in step (c). The method according to claim 1, including the method described in claim 1.

3. The process for determining the distribution of RBCs in a blood smear or sub-section of a blood smear is: (a) The step of assigning a tight bounding box to each cell; (b) A step of determining the contour lines of cell boundaries; (c) A step of determining the centroid of the geometric shape of the cell; (d) A step to determine the hemoglobin distribution parameter (HDP); (e) A step of classifying the central clear area according to the HDP, the step of identifying a clear area that is usable for diagnosis and has an HDP of approximately 0.5 to 2.0 The method according to claim 1 or 2, including the method according to claim 1 or 2.

4. In addition to steps (a) to (d) of claim 3, the step of identifying the presence of non-RBC cells and / or abnormal RBCs in a blood smear or sub-section of a blood smear is: (e) A step to determine the diameter of the cell; (f) A step in which the curvature value of the cell is determined; (g) A step of classifying cells as non-RBC or abnormal cells if the cell diameter is >8.5 μm or <6 μm; and / or the HDP is <0.5 or >2.0; and / or the curvature value is >1 / 3 μm. The method according to claim 3, including the method described in claim 3.

5. The method according to claim 4, wherein cells classified as abnormal are further analyzed to diagnose a disease.

6. A method for detecting a single layer of red blood cells (RBCs) in a blood smear suitable for diagnosis, comprising: (a) A step of acquiring multiple optical images of a blood smear; (b) (i) Using an artificial neural network, or (ii) A step of classifying each image based on the presence of a single layer region usable for diagnosis, performed according to the method of any one of claims 1 to 5; (c) A step of generating an image heatmap according to the presence of a single layer region usable for diagnosis; (d) A step in which groups of adjacent images in the heatmap generated in step (c) are analyzed to identify a single layer region that can be used for diagnosis; (e) A step of verifying the single-layer regions available for diagnosis identified in step (d) by calculating the density configuration of the selected images; (f) The process of combining all adjacent images of single-layer regions that can be used for diagnosis to create a single-layer panoramic overview image. A method for detecting a monolayer of red blood cells (RBCs) in a blood smear that is suitable for diagnosis, including the above.

7. The method according to claim 6, wherein the artificial neural network is trained on a training dataset generated by a pathologist, a hematologist, or a group of pathologists and hematologists.

8. The method according to any one of claims 1 to 7, wherein the classification is a binary classification of "usable for diagnosis" or "not usable for diagnosis".

9. The method of claim 8, wherein the assignment of labels for a heatmap or diagnostic availability is based on the results of a binary classification, or the heatmap represents a classification according to any one of claims 6(ii), preferably highlighting images or regions that are available for diagnosis.

10. The method according to any one of claims 6 to 9, wherein the identification of a single layer area usable for diagnosis is performed in a group of adjacent images using a sliding window procedure, preferably with a window size of 8 × 8 images.

11. The method according to claim 10, wherein single-layer regions usable for diagnosis that overlap and / or touch within a group of images are excluded.

12. The method according to any one of claims 1 to 4, wherein the diagnostic usability labels assigned to images contained within a blood smear or sub-section of a blood smear are compared with each image in a single-layer panoramic overview image obtained according to any one of claims 6 to 11, and if there is overlap between the two images, it is identified as a preferred RBC single-layer region usable for diagnosis.

13. The method according to any one of claims 1 to 12, wherein the blood smear is obtained using any method aimed at generating a cell monolayer, preferably as a wedge smear on a microscope slide, as a monolayer provided by spinning the slide in spinner centrifugation, or as a monolayer prepared in a flow cell.

14. Use of the method according to any one of claims 1 to 13 for selecting a monolayer RBC site that can be used for diagnosis.

15. A diagnostic method for blood-related diseases, (i) Identifying at least one RBC monolayer region in the blood smear of the target sample that is suitable for diagnosis, preferably using the method described in any one of claims 1 to 5 or 7 to 13; (ii) A step of identifying abnormal cells in the blood smear, preferably using the method described in any one of claims 1 to 5; (iii) A step in which cells are stained with a dye to identify abnormal molecular structures in the cells, the presence of abnormal cells and, if applicable, abnormal molecular structures indicates a blood-related disease. A method for diagnosing the aforementioned blood-related disease, including the above.

Images