Using trained machine learning models to determine alignment characteristics

Abstract

Computer implemented methods for flow cell analysis may include, and cell analysis systems may be configured to support, flowing blood cells from a sample through a flow cell, capturing images of the cells as they flow through the flow cell, and then determining an alignment characteristic score based on the plurality of cell images using a trained machine learning model.

Description

USING TRAINED MACHINE LEARNING MODELS TO DETERMINE ALIGNMENT CHARACTERISTICS BACKGROUND

[0001] Determining the morphology of blood cells (e.g., red blood cells) in a blood sample can be useful in ascertaining patient health. However, this type of morphology assessment can be difficult to perform in an automated manner.

[0002] In traditional methods of blood analysis, indirect techniques of measurement such as light- scattering analysis, fluorescence, or impedance-based methodologies were used. These indirect techniques have use in establishing certain types of information about cells, but have limitations including in ascertaining information about cell morphology. Newer classes of analyzers may use imaging to take images of blood cells, these images can be useful to obtain additional information about blood cells beyond the capabilities of traditional technologies, such as information pertaining to cell morphology. However, it can be difficult to determine which image characteristics are useful to obtain relevant information about cell morphology, and how these characteristics can be used together to cell morphology analysis.

[0003] There is therefore a need for cell morphology assessment methods, systems, devices, and / or algorithms utilizing imaging related data. SUMMARY

[0004] Aspects of the present disclosure may be used in determining the morphology of blood cells in hematology flow imaging analyzers.

[0005] A first aspect of the disclosed technology refers to a method for flow cell analysis comprising: causing blood cells from a biological sample to be flowed through a flow cell; causing a plurality of cell images to be captured by a camera, wherein each cell image from the plurality - 1 - 4860-7089-2770v7 Confidential - Company Proprietaryof cell images depicts at least a respective cell as it is flowing through the flow cell; and determining, by a computing device, an alignment characteristic score based on the plurality of cell images using a trained machine learning model.

[0006] In particular, causing the blood cells from the sample to be flowed through the flow cell comprises flowing the blood cells from the sample through the flow cell. For instance, causing the blood cells from the sample to be flowed through a flow cell comprises instructing, e.g., by the computing device, an analyzer comprising the flow cell to flow the blood cells from the sample through a flow cell. Causing the blood cells from the sample to be flowed through the flow cell may be carried out by the computing device.

[0007] In particular, causing the plurality of cell images to be captured comprises capturing, by using the camera, the plurality of cell images. For example, causing the plurality of cell images to be captured comprises instructing, e.g., by the computing device, the camera to capture the plurality of cell images. Causing the plurality of cell images to be captured may be carried out by the computing device.

[0008] The camera may be a photo camera, a video camera or the like. For instance, the camera is a device configured to capture image, in particular a digital image, more particularly a digital image of a cell. The images of the plurality of cell images may be represented by using the RGB color format. In particular, the images of the plurality of cell images may be represented by using the Hue, Saturation, Lightness (HSL) representation or the Hue, Saturation, Brightness (HSB) representation.

[0009] The sample is in particular a biological sample. In particular, the sample may be a sample of a bodily fluid of a human or animal subject. For example, the bodily fluid may be a physiological fluid, such as blood, saliva, urine, sweat, amniotic fluid, cerebrospinal fluid, ascites fluid, or the like. The biological sample may comprise one or more components that may potentially comprise analytes of interest for performing at least one clinical test. For instance, blood serum, blood cells and blood plasma are components of a blood sample. - 2 - 4860-7089-2770v7 Confidential - Company Proprietary

[0010] In particular, determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model may comprise, for each cell image of the plurality of cell images, processing said cell image, e.g., the pixels thereof, according to a first image processing procedure. For instance, determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model comprises, for each cell image of the plurality of cell images, identifying a respective foreground in that cell image based on a lightness of each pixel in that cell image and determining the plurality of cell parameter values by using the identified foregrounds, e.g., the pixels of those foregrounds.

[0011] In particular, according to the disclosed technology, a foreground in an image is a set of pixels of the image, e.g., the set comprising the pixels of the image that depict at least a portion of the cell. According to the technology disclosed herein, identifying the foreground in an image may comprise selecting the pixels of the image that have a lightness value greater than or equal to a lightness threshold. In some cases, e.g., if the images of the plurality of cell images are represented in the RBG color format and / or in the HSV representation, identifying a foreground in a cell image may comprise converting said cell image in the HSL representation.

[0012] For example, the alignment characteristic score comprises information specifying one or more of (i) an estimate of an amount of front facing cells in the sample; and (ii) an estimate of a fraction of non-front facing cells in the sample.

[0013] Throughout the present specification, any reference to a front facing cell and front facing cells may in particular refer to a front facing Red Blood Cell (RBC).

[0014] According to aspects of the disclosed technology, the computing device may comprise one or more processors. For instance, a processor may be a CPU or a GPU. Additionally, the computing device may comprise a memory. For example, the computing device may be a computer, a laptop, a smartphone, a tablet and the like. According to the present disclosure, a computing device may be a distributed system such as a computer network and / or a cloud computing architecture. In particular, the memory of the computing device comprises computer executable instructions which, when executed by the computing device, cause the - 3 - 4860-7089-2770v7 Confidential - Company Proprietarycomputing device to determine the alignment characteristic score based on the plurality of cell images using the trained machine learning model.

[0015] According to aspects of the present disclosure, the trained machine learning algorithm is in particular a machine learning algorithm that is trained so that it is configured to determine, by processing input data, the alignment characteristic score for the plurality of cell images. In particular, the input data may consist of, comprise, or be based on the plurality of cell images. For example, the respective input data may be generated by processing the plurality of cell images via a pre-processing procedure. For example, the input data may comprise or consist of a tensor that comprises the plurality of cell images. According to the disclosed technology, determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model may comprise processing the plurality of cell images by using the pre-processing procedure thereby generating the respective input data and processing said input data by using the trained machine learning algorithm thereby determining the alignment characteristic score for the plurality of cell images. For instance, the trained machine learning model may be a Convolutional Neural Network (CNN).

[0016] The method of the disclosed technology may further comprise, for each image from the plurality of cell images, determining a plurality of cell parameter values for that cell image.

[0017] In particular, determining the plurality of cell parameter values of a cell image may comprise processing the cell image, e.g., the pixels thereof, according to a second image processing procedure and thereby obtaining the plurality of cell parameter values. For instance, determining the plurality of cell parameter values of a cell image may comprise identifying the foreground in said cell image based on a lightness of each pixel in the image and determining the plurality of cell parameter values by using the foreground of the image, e.g. of the pixels of that foreground.

[0018] A cell parameter value of the plurality of cell parameter values may comprise information indicative of the dimensions of the cell. For instance, a cell parameter value of the plurality of cell parameter values may be the cell area of the cell. In particular, a cell parameter value of - 4 - 4860-7089-2770v7 Confidential - Company Proprietarythe plurality of cell parameter values may comprise information indicative of the form or shape of the cell. For example, a cell parameter value of the plurality of cell parameter values may be the eccentricity of the cell.

[0019] For example, according to the disclosed technology, the trained machine learning algorithm is, in particular, a machine learning algorithm that is trained so that it is configured to determine, by processing input data, the alignment characteristic score for the plurality of cell images. In particular, the input data may consist of, comprise, or be based on a set of cell parameter values. In particular, the set of cell parameter values comprises, for each image from the plurality of cell images, the plurality of cell parameter values for that image. For instance, the trained machine learning model may be a feed forward neural network.

[0020] In particular, for each image from the plurality of cell images, the plurality of cell parameter values for that cell image comprise: an area of the cell depicted in that image, a solidity of the cell depicted in that image, an eccentricity of the cell depicted in that image, an equivalent diameter of the cell depicted in that image, an area of a bounding box of the cell depicted in that image, and a circular shape of the cell depicted in that image.

[0021] For instance, for each image from the plurality of cell images, determining the plurality of parameter values for that image comprises providing that image to a second trained machine learning model, wherein the second trained machine learning model is configured to provide the plurality of parameters as its output.

[0022] According to aspects of the disclosed technology, the second trained machine learning algorithm is, in particular, a machine learning algorithm that is trained so that it is configured to provide, for each cell image from the plurality of cell images, the plurality of parameters for that cell image by processing respective input data. In particular, for each cell image of the plurality of cell images, the respective input data may consist of, comprise, or be based on that cell image. For example, the respective input data may be generated by processing the plurality of cell images via a pre-processing procedure. According to the present disclosure providing an image to the second trained machine learning model, may comprise processing that image - 5 - 4860-7089-2770v7 Confidential - Company Proprietaryby using the pre-processing procedure thereby generating the respective input data and processing said input data by using the second trained machine learning algorithm thereby determining the plurality of parameters for that image. For instance, the second trained machine learning model may be a Convolutional Neural Network (CNN).

[0023] The method of any one of the previous examples, wherein determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model comprises: identifying one or more of a set of front facing cell images and a set of non- front facing cell images by, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing; and computing the alignment characteristic score by using the one or more of the set of front facing cell images and non-front facing images.

[0024] According to aspects of the present disclosure, the trained machine learning algorithm is, in particular, a machine learning algorithm that is trained so that it is configured to determine, for each image of the plurality of cell images, whether that cell is front facing by processing respective input data. In particular, for each cell image of the plurality of cell images, the respective input data may consist of, comprise, or be based on that cell image. For example, the respective input data may be generated by processing the plurality of cell images via a pre- processing procedure. According to the present disclosure, providing an image to the second trained machine learning model, may comprise processing that image by using the pre- processing procedure thereby generating the respective input data and processing said input data by using the trained machine learning algorithm thereby determining whether that image is front facing. For instance, the trained machine learning model may be a Convolutional Neural Network (CNN).

[0025] For instance, according to aspects of the present disclosure, the trained machine learning algorithm is, in particular, a machine learning algorithm that is trained so that it is configured to determine for each image of the plurality of cell images, whether that cell is front facing by classifying that image into one of a plurality of categories. For example, the plurality of - 6 - 4860-7089-2770v7 Confidential - Company Proprietarycategories comprises a front-facing category. If this is the case, in particular, an image is determined to be front-facing only if the trained machine learning algorithm classifies that image as front facing i.e., that image is classified by the trained machine learning as belonging to the category of front facing images. For example, an image belongs to the category of front facing images if it depicts a front-facing RBC, e.g., exactly one front facing RBC. In particular, an image is determined not to be front-facing if the trained machine learning algorithm classifies that image in a category other than the front-facing category.

[0026] In some cases, the plurality of categories further comprises the non-front facing category. In particular, an image belongs to the non-front facing category if it does not depict exactly one RBC that is front-facing. For example, an image belongs to the non-front facing category if it depicts one of: exactly one platelet, exactly one RBC that is not front facing, a plurality of blood cells and exactly one white blood cell.

[0027] In some cases, the plurality of categories may comprise the following categories: (a) front facing category, (b) platelets category, (c) tumbled category, (d) cluster category, (e) White Blood Cell (WBC) category. In particular: a. an image belongs to the front facing category if it depicts exactly one RBC that is front-facing; b. an image belongs to the platelets category if it depicts exactly one platelet; c. an image belongs to the tumbled category if it depicts exactly one RBC that is not front facing; d. an image belongs to the cluster category if it depicts a plurality of blood cells; and e. an image belongs to the WBC category if it depicts exactly one WBC.

[0028] According to aspects of the present disclosure, a front-facing cell image is an image depicting a RBC, e.g., exactly one RBC, which front faces the camera, i.e., which is oriented in a front - 7 - 4860-7089-2770v7 Confidential - Company Proprietaryfacing orientation relative to the camera. For instance, according to the present disclosure, a non-front facing cell image is an image that does not depict a RBC cell, e.g., exactly one RBC, which front faces the camera. For example, a non-front facing cell image depicts one of: exactly one platelet, exactly one RBC that is not front facing, a plurality of blood cells and exactly one white blood cell. According to aspects of the disclosed technology, using the trained machine learning model to determine whether an image is front facing comprises using the trained machine learning model to determine whether that image depicts an RBC, e.g. exactly one RBC, which front faces the camera.

[0029] Typically, an RBC has a generally biconcave shape comprising two opposite facing concave external surface regions. In particular, each of the two concave external surface regions are substantially perpendicular to a traversal axis of the RBC. According to aspects of the disclosed technology, a front-facing orientation of the RBC is in particular the orientation in which one of the two concave external surface regions faces the camera. For example, a non front-facing orientation of the RBC is an orientation such that the traversal axis of the RBC is parallel to a field of view of the camera.

[0030] For instance, computing the alignment characteristic score comprises determining one or more of a cardinality of the set of front facing cell images and a cardinality of the set of non-front facing cell images.

[0031] For example, determining the alignment characteristic score of the analyzer comprises computing the alignment characteristic score, ^, according to the following formula: ^ =^^,wherein ^ is the number of images of the plurality of cell images and ^^is the cardinality of the set of front facing cell images.

[0032] Exemplarily, determining the alignment characteristic score of the analyzer comprises computing the alignment characteristic score, ^, according to the following formula: - 8 - 4860-7089-2770v7 Confidential - Company Proprietary^ =^ − ^^^,wherein ^^is the cardinality of thefacing cell images.

[0033] In some embodiments, computing the alignment characteristic score, ^, may be carried out by using the following formula: ^ = ^^⁄ ^ . In some further embodiments, computing the alignment characteristic score, ^, may be carried out by using the following formula: ^ = ^^⁄(^^+ ^^) .

[0034] In particular, the alignment characteristic score comprises a percentage of front facing cells. For instance, computing the alignment characteristic score, ^, may be carried out by using the following formula: ^ = ^^⁄(^^+ ^^) × 100.

[0035] In particular, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing comprises: determining a morphology score (hereinafter also referred to as: “morphology score”) for that cell image, wherein the morphology score for that cell image indicates a front facing orientation for that cell image; and comparing the morphology score to a threshold determined using the trained machine learning model.

[0036] The morphology score may be a value comprised between 0 and 1. For instance, the threshold determined using the trained machine learning model is equal to 0.5. If this is the case, for instance, if the morphology score for an image is greater than or equal to 0.5, the cell depicted in this image is considered to be an RBC that front faces the camera, i.e., to be oriented in a front facing orientation relative to the camera. In particular, if, instead, the morphology score for an image is lower than 0.5, the cell depicted in this image is considered not to be an RBC that front faces the camera.

[0037] In some embodiments, for each image from the plurality of cell images, determining the morphology score for that cell image comprises determining the morphology score for that cell image based on the plurality of cell parameter values for that cell image. - 9 - 4860-7089-2770v7 Confidential - Company Proprietary

[0038] In particular, determining the morphology score for a cell image based on the plurality of cell parameter values for that cell image may comprise processing said plurality of cell parameter values according to a cell parameters processing procedure and thereby obtaining the morphology score.

[0039] Exemplarily, for each image from the plurality of images, determining the morphology score for that image comprises providing the plurality of cell parameter values for that image as input to a morphology score generation function. The morphology score generation function may comprise a function which is: (i) directly proportional to a product of first and second inputs, and (ii) inversely proportional to a product of a third and a fourth input. In particular, for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing the area of the cell depicted in that image as the first input, providing the equivalent diameter of the cell depicted in that image as the second input, providing the eccentricity of the cell depicted in that image as the third input, and providing the circular shape of the cell depicted in that image as the fourth input.

[0040] In some embodiments, the morphology score generation function, ^, reads as follows: ^(^^, ^^, ^^, ^^) =^^ × ^^^ , ^wherein ^^(^ = 1,2,3,4) is the

[0041] In some embodiments, the function is also directly proportional to one or more of a fifth input and a sixth input. If this is the case, in particular, the for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing one or more of the solidity of the cell depicted in that image as the third input and the area of the bounding box f the cell depicted in that image as the fourth input. For instance, the morphology score generation function, ^, may read as follows: - 10 - 4860-7089-2770v7 Confidential - Company Proprietary^^^^, ^^, ^^, ^^,^ , ^!" =^^ × ^^ × ^ × ^!^,^ × ^^wherein ^ and ^!are

[0042] The method of some implementations of the disclosed technology further comprises determining, for each cell image from the plurality of cell images, a plurality of additional parameter values for that cell image, the plurality of additional parameter values being other than the plurality of cell parameter values. For example, for each image from the plurality of cell images, the plurality of additional parameter values for that cell image comprises one or more of a ratio of blueness to redness in that cell image or a granularity of the nucleus of the cell depicted in that image.

[0043] In some embodiments, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing comprises providing the plurality of parameter values for that image as input to the trained machine learning model.

[0044] In particular, the method of aspects of the present disclosure further comprises presenting a graphical interface comprising a plot displaying a plurality of points, wherein, for each point from the plurality of points, that point corresponds to a respective cell image from the plurality of cell images, wherein, for each point from the plurality of points, that point is displayed at a location determined based on the plurality of cell parameter values for that cell image.

[0045] The method according to aspects of the disclosed technology may further comprise determining, based on the plot, a design change in an analyzer comprising the flow cell. Additionally, the method may further comprise obtaining an updated analyzer by implementing the design change; and repeating the cell analysis method using the updated analyzer.

[0046] In particular, obtaining the updated analyzer comprises manufacturing the updated analyzer by using an updated version of the design of the first analyzer, said updated version including the - 11 - 4860-7089-2770v7 Confidential - Company Proprietarydesign change mentioned above. Additionally or alternatively, obtaining the updated analyzer may comprise modifying the first analyzer according to the aforementioned design change.

[0047] A second aspect of the disclosed technology refers to a cell analysis system comprising: a flow cell; a camera; a computing device. This system is configured to perform a method comprising: flowing blood cells from a biological sample through the flow cell; capturing, by using the camera, a plurality of cell images, wherein each cell image from the plurality of cell images depicts at least a respective cell as it is flowing through the flow cell; and determining, by the computing device, an alignment characteristic score based on the plurality of cell images using a trained machine learning model.

[0048] In particular, the flow cell and the camera are arranged with respect to one another so that the flow cell is configured to convey at least a portion of the sample fluid through a viewing zone of the camera.

[0049] In particular, the method further comprises, for each image from the plurality of cell images, determining a plurality of cell parameter values for that cell image.

[0050] Exemplarily, for each image from the plurality of cell images, the plurality of cell parameter values for that cell image comprise: an area of the cell depicted in that image, a solidity of the cell depicted in that image, an eccentricity of the cell depicted in that image, an equivalent diameter of the cell depicted in that image, an area of a bounding box of the cell depicted in that image, and a circular shape of the cell depicted in that image.

[0051] For example, for each image from the plurality of cell images, determining the plurality of parameter values for that image comprises providing that image to a second trained machine learning model, wherein the second trained machine learning model is configured to provide the plurality of parameters as its output.

[0052] In particular, determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model comprises: identifying one or more of a set - 12 - 4860-7089-2770v7 Confidential - Company Proprietaryof front facing cell images and a set of non-front facing cell images by, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing; and computing the alignment characteristic score by using the one or more of the set of front facing cell images and the set of non-front facing cell images.

[0053] In some embodiments, computing the alignment characteristic score comprises determining one or more of a cardinality of the set of front facing cell images and a cardinality of the set of non-front facing cell images.

[0054] In particular, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing comprises: determining a morphology score for that cell image, wherein the morphology score for that cell image indicates a front facing orientation for that cell image; and comparing the morphology score to a threshold determined using the trained machine learning model.

[0055] For example, for each image from the plurality of cell images, determining the morphology score for that cell image comprises determining the morphology score for that cell image based on the plurality of cell parameter values for that cell image.

[0056] Exemplarily, for each image from the plurality of images, determining the morphology score for that image comprises providing the plurality of cell parameter values for that image as input to a morphology score generation function. The morphology score generation function may comprise a function which is: (i) directly proportional to a product of first and second inputs, and (ii) inversely proportional to a product of a third and a fourth input. In particular, for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing the area of the cell depicted in that image as the first input, providing the equivalent diameter of the cell depicted in that image as the second input, providing the eccentricity of the cell depicted in that image as the third input, and providing the circular shape of the cell depicted in that image as the fourth input. - 13 - 4860-7089-2770v7 Confidential - Company Proprietary

[0057] In some embodiments, the function is also directly proportional to one or more of a fifth input and a sixth input. If this is the case, in particular, the for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing one or more of the solidity of the cell depicted in that image as the third input and the area of the bounding box of the cell depicted in that image as the fourth input.

[0058] In some embodiments of a system implemented based on this disclosure, the method further comprises determining, for each cell image from the plurality of cell images, a plurality of additional parameter values for that cell image, the plurality of additional parameter values being other than the plurality of cell parameter values.

[0059] In particular, for each image from the plurality of cell images, using the trained machine learning model to identify whether that image is front facing comprises providing the plurality of parameter values for that image as input to the trained machine learning model.

[0060] The alignment characteristic score may comprise information specifying one or more of an estimate of an amount of front facing RBCs in the sample and an estimate of a fraction of cells that are not front facing RBC cells in the sample. In particular, the alignment characteristic score comprises a percentage of front facing RBCss.

[0061] In particular, the method further comprises presenting a graphical interface comprising a plot displaying a plurality of points, wherein, for each point from the plurality of points, that point corresponds to a respective cell image from the plurality of cell images, wherein, for each point from the plurality of points, that point is displayed at a location determined based on the plurality of cell parameter values for that cell image. Additionally, the method according to aspects of this disclosure may further comprise determining, based on the plot, a design change in an analyzer comprising the flow cell.

[0062] A third aspect of the disclosed technology refers to a computing device comprising a processor configured to perform the method according to the first aspect of the disclosed technology. - 14 - 4860-7089-2770v7 Confidential - Company Proprietary

[0063] A fourth aspect of the disclosed technology refers to a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method according to the first aspect of the disclosed technology.

[0064] A fifth aspect of the disclosed technology refers to a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to the first aspect of the disclosed technology.

[0065] It should be noted that any of the various features of the aspects disclosed herein can be included or combined in each of those aspects. Other aspects and implementations are described herein. Accordingly, the exemplary aspects described in this summary should be understood as being illustrative only and should not be treated as limiting. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] While the specification concludes with claims which particularly point out and distinctly claim the invention, it is believed the present invention will be better understood from the following description of certain examples taken in conjunction with the accompanying drawings, in which like reference numerals identify the same elements and in which:

[0067] FIG.1 is a schematic illustration, partly in section and not to scale, showing operational aspects of an exemplary flow cell which may be used in an analyzer configured to capture and analyze images.

[0068] FIG. 2 illustrates an example method which may be used to analyze images of particles (e.g., cells) in a sample.

[0069] FIGS. 3A-3B illustrate an architecture for a machine learning model which can be used in analyzing images.

[0070] FIG. 4 is an example of a stage such as which might be included in a machine learning model following the architecture of FIGS. 3A-3B. - 15 - 4860-7089-2770v7 Confidential - Company Proprietary

[0071] FIG.5 illustrates acts which may be performed in identifying sets of front and non-front facing cell images.

[0072] FIG. 6 illustrates how an alignment characteristic score may be used to evaluate and improve the performance of an analyzer.

[0073] FIG. 7 illustrates an example interface which could be presented to a user in connection with evaluation of an analyzer.

[0074] FIG. 8 illustrates images of cells which are front facing.

[0075] FIG. 9 illustrates images of cells which are not front facing.

[0076] The drawings are not intended to be limiting in any way, and it is contemplated that various embodiments of the invention may be carried out in a variety of other ways, including those not necessarily depicted in the drawings. The accompanying drawings incorporated in and forming a part of the specification illustrate several aspects of the present invention, and together with the description serve to explain the principles of the invention; it being understood, however, that this invention is not limited to the precise arrangements shown. DETAILED DESCRIPTION

[0077] The present disclosure relates to articles, systems, and methods for determining the morphology of cells which are imaged by an automated analyzer. In some aspects, the analyzers may be visual analyzers comprising processors to facilitate automated conversion and / or analysis of images. Such analyzers may be useful, for example, in characterizing particles in biological fluids, such as detecting and quantifying erythrocytes, reticulocytes, nucleated red blood cells, platelets, and white blood cells, including white blood cell differential counting, categorization and subcategorization and analysis. Other similar uses such as characterizing blood cells from other fluids (serum, bone marrow, lavage fluid, effusions, exudates, cerebrospinal fluid, pleural fluid, peritoneal fluid, and amniotic fluid) are also contemplated. - 16 - 4860-7089-2770v7 Confidential - Company Proprietary

[0078] Turning now to the drawings, FIG. 1 schematically shows an exemplary flow cell 22 which may be used in an analyzer for conveying a sample fluid through a viewing zone 23 of a high optical resolution imaging device 24 (e.g., a camera) in a configuration for imaging microscopic particles in a sample flow stream 32 using digital image processing. Flow cell 22 is coupled to a source 25 of sample fluid which may have been subjected to processing, such as contact with a particle contrast agent composition and heating. Flow cell 22 is also coupled to one or more sources 27 of a particle and / or intracellular organelle alignment liquid (PIOAL) / sheath fluid, such as a clear glycerol solution having a viscosity that is greater than the viscosity of the sample fluid, an example of which is disclosed in U.S. Pat. Nos.9,316,635 and 10,451,612, the disclosures of which are hereby incorporated by reference in their entirety.

[0079] The sample fluid is injected through a flattened opening at a distal end 28 of a sample feed tube 29, and into the interior of the flow cell 22 at a point where the PIOAL flow has been substantially established resulting in a stable and symmetric laminar flow of the PIOAL above and below (or on opposing sides of) the ribbon-shaped sample stream. The sample and PIOAL streams may be supplied by precision metering pumps that move the PIOAL with the injected sample fluid along a flow path that narrows substantially. The PIOAL envelopes and compresses the sample fluid in the zone 21 where the flow path narrows. Hence, the decrease in flow path thickness at zone 21 can contribute to a geometric focusing of the sample flow stream 32. The sample flow stream 32 is enveloped and carried along with the PIOAL downstream of the narrowing zone 21, passing in front of, or otherwise through the viewing zone 23 of, the high optical resolution imaging device 24 where images are collected, for example, using a Charge-Coupled Device (CCD) 48 observing the sample stream as illuminated using an illumination source 42 through a viewing port 57. Processor 18 can receive, as input, pixel data from CCD 48. The sample fluid ribbon flows together with the PIOAL to a discharge 33.

[0080] As shown here, the narrowing zone 21 can have a proximal flow path portion 21a having a proximal thickness PT and a distal flow path portion 21b having a distal thickness DT, such that distal thickness DT is less than proximal thickness PT. The sample fluid can therefore be - 17 - 4860-7089-2770v7 Confidential - Company Proprietaryinjected through the distal end 28 of sample tube 29 at a location that is distal to the proximal flow path portion 21a and proximal to the distal portion 21b. Hence, the sample fluid can enter the PIOAL envelope as the PIOAL stream is compressed by the zone 21, wherein the sample fluid injection tube has a distal exit port through which sample fluid is injected into flowing sheath fluid, the distal exit port bounded by the decrease in flow path size of the flow cell.

[0081] The digital high optical resolution imaging device 24 with objective lens 46 is directed along an optical axis that intersects the ribbon-shaped sample flow stream 32. The relative distance between the objective 46 and the flow cell 33 is variable by operation of a motor drive 54, for resolving and collecting a focused digitized image on a photosensor array. Additional information regarding the construction and operation of an exemplary flow cell such as shown in FIG. 1 is provided in U.S. Patent 9,322,752, entitled “Flow cell Systems and Methods for Particle Analysis in Blood Samples,” filed on March 17, 2014, the disclosure of which is hereby incorporated by reference in its entirety.

[0082] In operation, an analyzer such as an analyzer incorporating a flow cell based imaging system as illustrated in FIG. 1 may analyze images of particles (e.g., cells) in a sample, such as by using a process of the type shown in FIG. 2. Initially, in the process of FIG. 2, cells from a sample (e.g., a blood sample or other type of biological sample) are flowed 201 through a flow cell (e.g., the flow cell 22 depicted in FIG. 1). As the cells are being flowed 201 through the flow cell, a plurality of cell images may be captured 202 (e.g., using a camera 24 as described in the context of FIG. 1), in which each cell image may depict a cell as it is flowing through the flow cell. Examples of these types of cell images are provided in FIGS. 8-9, in which FIG. 8 depicts images of cells which are front facing, and FIG. 9 depicts images of cells which are not front facing. The cell images may then be processed to obtain 203 an alignment characteristic score, or may be subject to some additional preprocessing before the alignment characteristic score of obtained 203. For example, in some cases, a camera 24 may capture a superset of cell images, and that superset of cell images may be filtered to remove images which are out of focus so that the plurality of images which are processed to obtain 203 the alignment characteristic score would include only images which are sufficiently in focus. This - 18 - 4860-7089-2770v7 Confidential - Company Proprietarymay involve determining focus values using a Laplacian transform as described in J.L. Pech- Pacheco et. al., Diatom autofocusing in brightfield microscopy: a comparative study, Proceedings 15th International Conference on Pattern Recognition. ICPR-2000, Barcelona, Spain, 2000, pp. 314-317 vol.3, available at https: / / ieeexplore.ieee.org / document / 903548; determining focus using contrast such as in “Flowcell, Sheath fluid, and Autofocus SystemsSamples”, U.S. Patent 10,705,008 entitled “Autofocus Systems and Methods for Particle Analysis in Blood Samples,” international application PCT / US2023 / 011759 entitled “measure image quality of blood cell images,” each of which is hereby incorporated by reference in its entirety, and then comparing those determined focus values with a threshold which had been previously determined for identifying images which were suitable for being further processed as described herein.

[0083] Other types of processing may also be performed prior to, or in the context of, obtaining 203 an alignment characteristic score in a method such as shown in FIG. 2. For example, in some cases, image preprocessing may include separating the foreground from the background of each image, e.g., using thresholding or by using the Otsu’s method. After any applicable preprocessing (e.g., foreground separation, image filtering, etc.) was complete, the images may be further processed to determine 204 a set of front facing images and a set of non-front facing images. This may be done using a machine learning model trained to classify an imaged cell as being front facing or not front facing. Such training may include dividing a set of training images into two disjoint sets, e.g., two sets of approximately 10,000 training images each. A first of those sets may consist of images which each depict exactly one front facing red blood cell, and each image in that set may be labeled (e.g., by a human annotator) as belonging to the front facing category. The second of those sets may consist of images which depict exactly one platelet, exactly one RBC that is not front facing, a plurality of blood cells or exactly one white blood cell, and each image in that set may be labeled (e.g., by a human annotator) as belonging to the non-front facing category. These image sets may be used to train the machine learning model by providing the images to the model without their labels, asking the model to assign a - 19 - 4860-7089-2770v7 Confidential - Company Proprietarylabels to the images, and then comparing the labels from the model with the images’ actual labels. In this case, if the model’s labels are incorrect, a loss function (e.g., mean squared error) and learning rate (e.g., 10-3) may be used to convert the difference between the labels as provided by the model and the correct labels into a numeric value that could be used to adjust the model’s parameters so that it generated more accurate predictions in the future. This may then be repeated one or more times until the some predefined training metric was achieved (e.g., completion of 200 epochs, with each epoch corresponding to a batch size of 512 images; when the model’s performance on a set of training images begins to diverge from its performance on a set of validation images; etc.).

[0084] Other approaches may also be used when training a machine learning model to identify if cells are front facing. For example, in some cases, rather than simply splitting training images in to front facing and non-front facing cells, images may be split (e.g., by a human annotator who would assign an image to a category by labeling it) into the following sets of training images: a first set of training images which each depict exactly one front facing red blood cell; a second set of training images which each depict exactly one platelet; a third set of training images which each depict exactly one red blood cell that is not front facing; a fourth set of training images which each depict exactly one white blood cell; and a fifth set of training images which each depict a plurality of blood cells. A machine learning model may then be trained to classify images into one of the above noted categories in a manner similar to that described above for two class categorization, and which category the model assigned to an image could then be used to treat it as front facing or not front facing in practice. Accordingly, the above described example of simply training a machine learning model to assign images to a front facing or not front facing category should be understood as being illustrative only, and should not be treated implying limits on the types of trained machine learning models which may be used in various implementations of the disclosed technology.

[0085] Just as a variety of approaches may be used in training a machine learning model to identify images as depicting or not depicting front facing cells, a variety of machine learning models may be used to actually make that identification. As an example, FIGS. 3A and 3B depict an - 20 - 4860-7089-2770v7 Confidential - Company Proprietaryarchitecture which could be used for a machine learning model in some systems implemented based on this disclosure. In the architecture of FIGS. 3A-3B, an input image 301 would be analyzed in a series of stages 302303304305, each of which comprises a set of layers which are illustrated in more detail in FIG. 4, discussed below.

[0086] As shown in FIG. 4, in the stages 302-305 of FIGS. 3A-3B an input 301 (which, in the initial stage 302 of FIG.3A would be a cell image and otherwise would be the output of the preceding stage) is provided to a set of layers 402 where it would be processed by a convolutional layer to generate one or more transformed images 403a-403n. This processing may include convolving the input 401, as well as (potentially) an imported input 404 provided over an incoming skip connection 405, with a set of filters 406a-407n, each of which would identify a type of feature from the underlying image and then applying an activation function (e.g., rectified linear unit, aka ReLU activation) to the output to provide a transformed image capturing the feature identified by the filter. For instance, as a simple example, convolving an image with the filter shown in table 1 −1 −1 −1#−1 8 −1%−1 −1 −1Table 1 could generate a transformed image capturing edges from the input 301.

[0087] After a set of transformed images 403a-403n had been created, those images may be combined into a single resultant image 407, in which each of the transformed images 403a-403n (or each channel of each transformed image, in a case where one or more transformed images had multiple channels) was treated as a channel of the resultant image 407. The resultant image 407 may then be provided as an output on an outgoing skip connection 409, and may also be subjected to some additional processing to create an output that could be provided to the next stage (e.g., in the case of the first stage 302, and image that could be treated as the input 401 of the second stage 303). For example, in some cases, the resultant image 407 may be provided - 21 - 4860-7089-2770v7 Confidential - Company Proprietaryas input to a pooling layer which would generate a pooled image 408. This may be done, for example, by organizing each channel of the resultant into a set of regions, and then replacing the values in that region with a single value, such as the maximum value for the region or the average of the values for the region. The result would be a pooled image 408 with channels whose resolution would be reduced relative to its corresponding transformed image channels based on the size of the regions it was split into (e.g., if a channel in the resultant image 407 had NxN dimensions, and it was split into 2x2 regions, then the corresponding channel in the pooled image 408 would have size (N / 2)x(N / 2)), which could then be used as the input 401 of the next stage.

[0088] As shown in FIGS. 3A-3B, these stages can be connected to one another in an interleaved fashion using their outputs and skip connections. For example, as shown in FIG.3A, the output of a first stage 302 can be provided as the input to a second stage 303, as well as being provided to the third stage 304 as an imported input 404 along the first stage’s outgoing skip connection 409. Similarly, the output of the second stage 303 can be provided as the input to the third stage 304 as well as being provided to the fourth stage 305 as an imported input 404 along the second stage’s outgoing skip connection 409. The output from the fourth stage 305 can be passed through a flattening layer to obtain a one dimensional vector 306 which can be provided as input to a series of dense layers (e.g., first dense layer 307 and second dense layer 308). Those dense layers may then provide their output (e.g., the values on the output nodes of the final dense layer) to an output layer 309 having one neuron for each class the machine learning model having the architecture of FIGS.3A-3B had been trained to identify. The values of these neurons can then be normalized (e.g., using a softmax function), and the neuron having the highest value could be treated as reflecting the classification of the initial input image 301 (e.g., if there were two neurons, then if a neuron corresponding to front facing had the higher value, then the input image could be treated as depicting a front facing RBC, whereas if the neuron corresponding to non-front facing had the higher value, then the input image could be treated as not depicting a front facing RBC). - 22 - 4860-7089-2770v7 Confidential - Company Proprietary

[0089] It should be understood that, while FIGS. 3A-3B depict an architecture which may be used in a machine learning model implemented based on this disclosure, those figures depict a simplified architecture only, and that actual architectures are likely to be significantly more complex than shown in FIGS. 3A-3B. For example, table 2 illustrates example numbers of filters, dense layer neurons and filter dimensions that may be used in an implementation following the architecture of FIGS. 3A-3B. First Stage Number of filters: 32 3 4 3 8 3 6 3 6 8 2Implementations may also vary in aspects other than numbers and dimensions as shown in Table 2. For example, in some cases, the disclosed technology may be implemented using a machine learning model which included more sets of interleaved stages (e.g., prior to the series - 23 - 4860-7089-2770v7 Confidential - Company Proprietaryof dense layers, there may be a fifth stage which receives a pooled image 408 from the fourth stage 305 as well as a resultant image 407 on the outgoing skip connection 409 of the third stage 304), different numbers of dense layers (e.g., 0, 1, 3, 4 or more dense layers) before the output layer that would provide a classification of the input image 301.

[0090] It is also possible that, rather than simply changing the number and complexity of layers, some implementations may include additional types of layers which are not depicted in FIGS. 3A- 3B. For example, in some cases one or more dropout layers may be incorporated into a machine learning model, such as by inserting a 50% dropout layer between dense layers in an architecture which includes two dropout layers such as shown in FIGS.3A and 3B. As another example of a potential type of variation, in some cases relationships between stages may differ from what is discussed above. For instance, in some cases skip connections may be omitted, and each stage may take as its sole input a single image (either the initial input image, or the pooled image from a preceding stage). Table 3 provides an illustration of layers which may be used in this type of implementation. Layer 1 Input layer: 128x128x3 (e.g., an RGB image depicting a red blood cell)- 24 - 4860-7089-2770v7 Confidential - Company ProprietaryLayer 9 Max pooling layer: 2x2 pooling area er

[0091] It is also possible that various known types of machine learning models, such as AlexNet, VGG16 or VGG19, ResNet, or MobileNet (as described, respectively in Krizhevsky, et. al., ImageNet Classification with Deep Convolutional Neural Networks, available at https: / / proceedings.neurips.cc / paper / 2012 / file / c399862d3b9d6b76c8436e924a68c45b- ImageRecognition, available at https: / / arxiv.org / pdf / 1409.1556; He, et. al., Deep Residual Learning for Image Recognition,at https: / / arxiv.org / pdf / 1512.03385; and Howard, et. al., MobileNets: Efficient Convolutional for Mobile Vision Applications,available at each of which is herein incorporated by reference in its entirety), either as available or with some modification (e.g., insertion of additional skip connections to support more accurate separation of ambiguous inputs).

[0092] It should be understood that, while the above discussion of FIGS. 3A-3B and 4, and Tables 1- 3 focused on identifying 204 front and non-front facing images using trained machine learning models, this type of identification may be made in other manners as well. For instance, in some cases, the identification 204 of front facing images may be made using a method which algorithmically analyzes attributes of an image, such as using a method such as that depicted in FIG. 5, discussed below. - 25 - 4860-7089-2770v7 Confidential - Company Proprietary

[0093] As shown in FIG. 5, identifying 204 the sets of front and non-front facing cell images may include determining 501 morphology scores for each of the captured 202 cell images. This may comprise, for each of the captured images, determining 502 a plurality of cell parameter values for that image. Exemplary cell parameter values that may be provided by this determination 502, as well as acts which may be performed in determining each of them, are provided below in Table 4. Cell Parameter Value Determination ll A f i l h i l i f i ll he e g. e of in he is of ay he al er a. a nt te e.- 26 - 4860-7089-2770v7 Confidential - Company ProprietaryBounding Box Determine the area of the bounding box of the cell, i.e. the smallest Area box which completely covers the cell in the cell image. The area of the at y ve g he he ly

[0094] In particular, one or more of the Cell Area, the Hull area, the Equivalent diameter, the Bounding Box, the Bounding Box Area, the Eccentricity mentioned above may be defined according to the definitions of the scikit-image python package, e.g. of version 0.21.0 thereof. In particular, one or more of the Cell Area, the Hull area, the Solidity, the Equivalent diameter, the Bounding Box, the Bounding Box Area, the Eccentricity mentioned above may be computed by using of the scikit-image python package, e.g. of version 0.21.0 thereof. For instance, in some embodiments, exemplary cell parameter values that may be provided by the determination of the plurality of cell parameter values 302, as well as properties of the scikit- image python package that may be used to estimate them, are provided below in Table 6. Cell Parameter Value Determination d- 27 - 4860-7089-2770v7 Confidential - Company ProprietaryEquivalent Access the value of the property “equivalent_diameter_area” of the Diameter foreground region n d he ty n ed

[0095] In a method such as shown in FIG. 5, once the plurality of cell parameter values had been determined 502 for each cell image, morphology scores may be determined 503 for the captured images using those images’ parameter values. This may be done, for example, by, for each image, providing the cell parameter values for that image as inputs to a morphology score generation function, which function may calculate the morphology score, 78, for an image using a formula such as set forth below as equation 1: 78 =9:;; <=:> × ?@ABCDBE ?@F <=:> × G@;DCDHI × JKADL>;:BH MD>N:H:=JOO:BH=DODHI × 9D=OA;>= GP>Q:The morphology score, 78, may alternatively be calculated by using a formula such as set forth below as equation 2: 78 =9:;; <=:> × JKADL>;:BH MD>N:H:=JOO:BH=DODHI × 9D=OA;>= GP>Q:Equation 2: Exemplary morphology score calculation - 28 - 4860-7089-2770v7 Confidential - Company ProprietaryHowever it is calculated, once the morphology scores had been determined 501 for the plurality of cell images, those morphology scores could be compared 504 to a threshold (e.g., set by a human, or determined using a machine learning model such as discussed previously), with the images with morphology scores above the threshold being treated as front facing cell images, and the images with morphology scores below the threshold being treated as non-front facing cell images.

[0096] It should be noted that, while the above discussion presented machine learning and parameter based approaches as alternative approaches for identifying 204 front facing cell images, it is possible that machine learning and parameter based approaches may be combined in some cases. For example, in some cases, rather than using a neural network or other machine learning model to classify images based on parameters determined using a model made up of one or more layers of the type discussed in the context of FIGS. 3A-3B and 4, a machine learning model could be trained to classify an image as either front facing or non-front facing based on the parameters shown in table 4, thereby combining the cell parameter value determination 502 described in the context of equation 1 or equation 2 with the machine learning classification described in the context of FIGS.3A-3B and 4. As another example, in some cases, a machine learning approach such as described in the context of FIGS. 3A-3B and 4 could be used to classify images as front facing cell images or non-front facing cell images, and parameters such as shown in table 4 could later be separately calculated to facilitate understanding issues which may have led to particular images being side rather than front facing. Other combinations and variations are also possible and will be immediately apparent to those of skill in the art in light of this disclosure. Accordingly, the above examples of variations, as well as the examples given previously in the context of equation 1 or equation 2 and FIGS. 3A-3B and 4, should be understood as being illustrative only, and not be treated as implying limitations on the protection provided by this document or any related document.

[0097] However the identification 204 of front and non-front facing images takes place, an alignment characteristic score may be obtained 203 by dividing 205 a cardinality of the set of front facing images by a sum of the cardinality of the set of front facing images and the set of non-front - 29 - 4860-7089-2770v7 Confidential - Company Proprietaryfacing images. The alignment characteristic score may then be used for various purposes, such as improving the design of the analyzer or troubleshooting issues relating to the flowcell or analyzer. In one example - as shown in FIG. 6, during the process of designing an analyzer, after an alignment characteristic score had been obtained 203, a determination 601 may be made as to whether that score is above a threshold. For instance, if the analyzer needed to be able to obtain front facing images of at least 80% of cells, then a comparison 601 could be performed to determine if the alignment characteristic score was above that 80% threshold. If it was, then the design could be treated as suitable, and the design process could end 602. Alternatively, if the alignment characteristic score was not above the threshold, the information determined from analyzing the sample may be presented to a user involved in the analyzer’s design to help understand issues contributing to the low characteristic score.

[0098] In one example, the machine learning algorithm can be used to determine an alignment characteristic score for a sample, where the parameter values / equations used in the context of morphology scores (e.g., Table 6 and Equations 1 & 2) can be used to diagnose particular issues with analyzer design or sample quality which may contribute to the alignment characteristic score, or aid in redesigning or recalibrating aspects of the analyzer., For example, if the alignment characteristic score is not above a threshold, this can trigger a secondary process whereby parameter values such as described above in the context of calculating morphology scores could be calculated for the images and used for subsequent further analysis. . This may be done, for instance, by analyzing the parameters and identifying issues with the analyzer, flowcell, or blood sample (e.g., pressure or density of the sheath fluid, sheath or sample velocity, particular environmental conditions) that those parameters may reveal. For example, particular parameter values can be associated with particular conditions (e.g., particular parameter values being associated with issues involving pressure or density of the sheath fluid), to help diagnose issues related to poor alignment scores. Similarly, in some cases, the parameter values may be used to assess the analyzer design or even take some corrective action (e.g., adjusting a design element or changing an analyzer parameter) by clustering the images in an n-dimensional space where the dimensions corresponded to the cell - 30 - 4860-7089-2770v7 Confidential - Company Proprietaryparameters, and then presenting 603 an interface which included a plot showing how images were distributed according to the parameters which had the largest impact on the morphology scores (e.g., the two most impactful parameters in a two dimensional plot, or the three most impactful images in a three dimensional plot). An example of such a plot is provided in FIG. 7, which illustrates differences between cells which were classified as front facing and cells which were classified as non-front facing. As can be seen in that figure, the distribution of those cells may vary significantly by parameter, and a user may use this type of difference to indicate that there is an issue with the analyzer design that would require troubleshooting to bring the values of the errant parameter(s) more into line (e.g., the errant parameters may indicate that there is an issue with the sheath that should be enveloping the sample fluid, and the user may focus his or her troubleshooting efforts accordingly).

[0099] While the systems and methods discussed above illustrate potential implementations of the disclosed technology, it should be understood that those systems and methods are intended to be illustrative only, and that there are other systems and methods which could be implemented by those of ordinary skill in the art without undue experimentation based on this disclosure. To illustrate, consider the separation of foreground from background portions of a cell image as discussed above. While it is possible that, in some embodiments where such separation takes place, it may be performed using thresholding as described, other approaches are also possible. For example, in some cases, the foreground (cell) portion of an image may be separated form the background (non-cell) portions of an image using techniques such as described in U.S. patent 10,705,011, issued on 7 / 7 / 2020 for “Dynamic Focus System and Methods,” the disclosure of which is incorporated by reference herein in its entirety. As another example, in some cases algorithms such as the Chan-Vese algorithm, described in T. Chan and L. Vese, “Active contours without edges,” in IEEE transactions on image processing 10(2), 2001, pp. 266-277, the disclosure of which is incorporated herein in its entirety, may be used. It is also possible that, in some cases, the disclosed technology may be implemented in methods which do not include separating the foreground from background of captured images (e.g., an - 31 - 4860-7089-2770v7 Confidential - Company Proprietaryimplementation in which an image is provided as input to a machine learning model without first separating foreground and background portions of that image).

[0100] Other types of variations are also possible. For example, while the discussion of FIG. 6 focused on an alignment characteristic score being used during the design of an analyzer, such a characteristic score may also be used in the analysis of patient samples after design is complete. For example, an analyzer may be configured to determine an alignment characteristic score indicating a percentage of apparently front facing cells in a patient sample, and may use such a score to determine if it appeared the patient associated with the sample had a disorder associated with abnormal cell morphology. For instance, if the percentage of cells which appeared to be front facing was below a threshold, then the analyzer may indicate that the patient had or appeared to have a particular disease state, such as sickle cell.

[0101] Similarly, in some cases individual parameters may be used to identify conditions reflected in a patient sample. For instance, in some cases, morphology scores or constituent variables contributing to morphology scores may be used to assess particular cells associated with particular morphology characteristics, such as sickle cells and schistocytes. In some examples, particular solidity value ranges are associated with sickle cells. Other examples, such as using circular shape value ranges to identify schistocytes, or eccentricity value ranges to identify agglutinations are also possible, and will be immediately apparent to those of skill in the art in light of this disclosure. In some examples, the morphology score ranges are correlated to particular cell or disease conditions or the combination of particular variables in the morphology score (e.g., one of more of constituent variables of Equation 1) and the morphology score are used in conjunction to identify particular cell or disease conditions (e.g., sickle cells, or schistocytes). Accordingly, the examples of using the disclosed technology to identify sickle cell, schistocytes and / or agglutinations should be understood as being illustrative only, and should not be treated as limiting.

[0102] Variations also exist in parameters which may be collected and / or used in implementations which follow approaches such as described in the context of Equation 1 or - 32 - 4860-7089-2770v7 Confidential - Company Proprietaryequation 2. For example, in some cases, analyzers may be configured to determine additional parameters for a cell image other than the parameters used in the determination of whether it is a front facing or non-front facing cell image. For instance, it is possible that an analyzer may determine parameters such as ratio of blueness to redness in a cell image or granularity of the nucleus, while determining a morphology score using only the parameters listed in table 4. It is also possible that, in some cases, not all of the parameters listed in table 4 may be used in determining a morphology score. For instance, in some cases rather than combining all six of the parameters in table 4, a morphology score may be determined by using two or more of those parameters, such as placing those parameters in the numerator or denominator of an equation which is similar to equation 1 or equation 2 but omits one or more of the parameters that equations include. Still further variations are possible and will be immediately apparent to those of skill in the art in light of this disclosure. Accordingly, the examples set forth herein should be understood as illustrative only, and should not be treated as implying limitations on the scope of protection provided by this document or any related document.

[0103] It is also possible that different embodiments may vary from each other in portions of captured images which are processed, such as by providing them to a machine learning model to determine if they were or were not front facing. For example, in some cases all of the images captured by the camera may be processed to determine whether they are or are not front facing. However, in other cases, only a subset of the images captured by the camera may be so processed. One example of this type of case was described previously in the context of embodiments which would evaluate whether captured images were in focus and discard images which were not, but this is not the only example. For instance, in some cases only images of certain types of cells, such as images of RBCs, would be processed to determine whether they were front facing. It is also possible that only a subset of images of a type of cell, such as only a predefined percentage of images of RBCs would be processed as described above. It is also possible that, in some cases, an analyzer may be configured with separate modes, and that the disclosed technology may only be applied in cases when the analyzer is operating in a specific mode – e.g., a diagnostic mode used to evaluate if the analyzer was appropriately handling - 33 - 4860-7089-2770v7 Confidential - Company Proprietarycells so as to be able to capture images of the cells when they are front facing. Other variations on subsets, or sub-subsets of images which may be processed as described are also possible, and so the protection provided by this document (or any related document) should not be treated as being limited only to embodiments where all captured images, or where some specific portion of captured images, are processed using the disclosed techniques.

[0104] Each of the calculations or operations described herein may be performed using a computer or other processor having hardware, software, and / or firmware. The various method steps may be performed by modules, and the modules may comprise any of a wide variety of digital and / or analog data processing hardware and / or software arranged to perform the method steps described herein. The modules optionally comprising data processing hardware adapted to perform one or more of these steps by having appropriate machine programming code associated therewith, the modules for two or more steps (or portions of two or more steps) being integrated into a single processor board or separated into different processor boards in any of a wide variety of integrated and / or distributed processing architectures. These methods and systems will often employ a tangible media embodying machine-readable code with instructions for performing the method steps described above. Suitable tangible media may comprise a memory (including a volatile memory and / or a non-volatile memory), a storage media (such as a magnetic recording on a floppy disk, a hard disk, a tape, or the like; on an optical memory such as a CD, a CD-R / W, a CD-ROM, a DVD, or the like; or any other digital or analog storage media), or the like.

[0105] All patents, patent publications, patent applications, journal articles, books, technical references, and the like discussed in the instant disclosure are incorporated herein by reference in their entirety for all purposes.

[0106] Different arrangements of the components depicted in the drawings or described above, as well as components and steps not shown or described are possible. For example, in some cases, aspects of processing described herein (e.g., determination of a morphology score, use of such a score to evaluate a patient’s health, use of such a score to assess cell alignment in a flowcell) - 34 - 4860-7089-2770v7 Confidential - Company Proprietarymay be performed in various configurations – for instance, using a processor which is comprised by (or local to) an analyzer, a parallel-processing arrangement, or processing being performed remotely from the analyzer which captures the processed images (such as using a cloud based platform, or using a remotely linked computer or system to process the analyzer results). Similarly, some features and sub-combinations are useful and may be employed without reference to other features and sub-combinations. Embodiments of the invention have been described for illustrative and not restrictive purposes, and alternative embodiments will become apparent to readers of this patent. In certain cases, method steps or operations may be performed or executed in differing order, or operations may be added, deleted or modified. It can be appreciated that, in certain aspects of the invention, a single component may be replaced by multiple components, and multiple components may be replaced by a single component, to provide an element or structure or to perform a given function or functions. Except where such substitution would not be operative to practice certain embodiments of the invention, such substitution is considered within the scope of the invention. Accordingly, the claims should not be treated as limited to the examples, drawings, embodiments and illustrations provided above, but instead should be understood as having the scope provided when their terms are given their broadest reasonable interpretation as provided by a general-purpose dictionary, except that when a term or phrase is indicated as having a particular meaning under the heading Explicit Definitions, it should be understood as having that meaning when used in the claims.

[0107] Explicit Definitions

[0108] It should be understood that, in the above examples and the claims, a statement that something is “based on” something else should be understood to mean that it is determined at least in part by the thing that it is indicated as being based on. To indicate that something must be completely determined based on something else, it is described as being “based EXCLUSIVELY on” whatever it must be completely determined by.

[0109] It should be understood that, in the above examples and claims, the term “set” should be understood as one or more things which are grouped together. The cardinality of a set is - 35 - 4860-7089-2770v7 Confidential - Company Proprietarydefined as the number of elements of the set. For example, a set with one element will have cardinality 1, a set with five elements will have cardinality 5, etc. etc. Similarly, “superset” and “subset” should be each be understood as being synonymous with “set,” with the alternative terms “superset” and “subset” being used only for ease of reading. For the avoidance of doubt, a “superset” of a type of item should not be understood as necessarily referring to all items of that type, unless such is explicitly stated. For example, a reference to capturing a “superset” of cell images should not be understood as requiring that the “superset” of cell images are all of the captured cell images. While the “superset” may be all of the captured cell images, this is not necessary, and should not be treated as being implied by the term “superset.” - 36 - 4860-7089-2770v7 Confidential - Company Proprietary

Claims

What is claimed is:

1. A method for flow cell analysis comprising: capturing a plurality of cell images, each cell image depicting a respective blood cell of a blood sample as it flows through a flow cell; and determining an alignment characteristic score based on the plurality of cell images using a trained machine learning model.

2. The method of claim 1, further comprising, for each image from the plurality of cell images, determining a plurality of cell parameter values for that cell image.

3. The method of claim 2 wherein, for each image from the plurality of cell images, the plurality of cell parameter values for that cell image comprise: an area of the cell depicted in that image, a solidity of the cell depicted in that image, an eccentricity of the cell depicted in that image, an equivalent diameter of the cell depicted in that image, an area of a bounding box of the cell depicted in that image, and a circular shape of the cell depicted in that image.

4. The method of claims 2 or 3, wherein, for each image from the plurality of cell images, determining the plurality of parameter values for that image comprises providing that image to a second trained machine learning model, wherein the second trained machine learning model is configured to provide the plurality of parameters as its output.

5. The method of any of claims 2-4, wherein the method comprises assessing performance of an analyzer which comprises the flow cell based on comparing the alignment characteristic score with an alignment threshold.

6. The method of claim 5, wherein determining the plurality of cell parameters for each image from the plurality of cell images is performed based on comparing the alignment characteristic score with the alignment threshold. - 37 - 4860-7089-2770v7 Confidential - Company Proprietary7. The method of any of claims 5-6, wherein the method further comprises: based on the comparison of the alignment characteristic score with the alignment threshold, presenting a graphical user interface comprising a plot; and updating a design of the analyzer based on the plot.

8. The method of claim 7, wherein the plot displays a plurality of points, wherein, for each point from the plurality of points, that point corresponds to a cell image from the plurality of cell images, wherein, for each point from the plurality of points, that point is displayed at a location determined based on the plurality of cell parameter values for that cell image.

9. The method of any one of the previous claims, wherein determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model comprises: identifying a set of front facing cell images by, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing; and computing the alignment characteristic score by using the set of front facing cell images.

10. The method of claim 9, wherein computing the alignment characteristic score comprises determining a cardinality of the set of front facing cell images.

11. The method of claims 9 or 10, wherein, for each image from the plurality of cell images, using the trained machine learning model to identify whether that image is front facing comprises: determining a morphology score for that cell image, wherein the morphology score for that cell image indicates a front facing orientation for that cell image; and comparing the morphology score to a threshold determined using the trained machine learning model. - 38 - 4860-7089-2770v7 Confidential - Company Proprietary12. The method of claim 11 when taken in combination with any one of the claims 2 to 4, wherein, for each image from the plurality of cell images, determining the morphology score for that cell image comprises determining the morphology score for that cell image based on the plurality of cell parameter values for that cell image.

13. The method of claim 12, wherein: for each image from the plurality of images, determining the morphology score for that image comprises providing the plurality of cell parameter values for that image as input to a morphology score generation function; wherein the morphology score generation function comprises a function which is directly proportional to a product of first and second inputs, and which is inversely proportional to a product of a third and a fourth input, wherein, for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing the area of the cell depicted in that image as the first input, providing the equivalent diameter of the cell depicted in that image as the second input, providing the eccentricity of the cell depicted in that image as the third input, and providing the circular shape of the cell depicted in that image as the fourth input.

14. The method of any one of claims 2-13, wherein the cell analysis method comprises determining, for each cell image from the plurality of cell images, a plurality of additional parameter values for that cell image, the plurality of additional parameter values being other than the plurality of cell parameter values.

15. The method of any one of claims 2 to 14, wherein, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing comprises providing the plurality of parameter values for that image as input to the trained machine learning model. - 39 - 4860-7089-2770v7 Confidential - Company Proprietary16. The method of any one of the preceding claims, wherein the alignment characteristic score comprises information specifying one or more of an estimate of an amount of front facing cells in the sample and an estimate of a fraction of front facing cells in the sample.

17. The method of any one of the preceding claims, wherein the alignment characteristic score comprises a percentage of front facing cells.

18. The method of claim 1, wherein the plurality of cell images comprises a subset of a larger set of cell images.

19. The method of claim 1, wherein the plurality of cell images comprises a subset of a larger set of cell images, wherein the larger set of cell images are associated with a focus quality, and wherein the plurality of cell images meet at least a particular focus quality threshold.

20. A cell analysis method comprising: capturing a plurality of cell images, each cell image depicting a respective blood cell of a blood sample as it flows through a flowcell; and calculating a score for each cell image; wherein the score represents a front facing orientation assessment.

21. The method of claim 20, wherein, for each image from the plurality of cell images, the cell depicted in that cell image is a red blood cell.

22. The method of any of claims 20-21, wherein calculating the score for each cell image comprises providing each cell image to a machine learning model trained to determine front facing scores.

23. The method of any of claims 20 or 21, wherein, for each image from the plurality of cell images, determining the score for that cell image comprises determining a plurality of cell - 40 - 4860-7089-2770v7 Confidential - Company Proprietaryparameter values for that image, and providing the plurality of cell parameter values for that image as input to a morphology score generation function.

24. The method of claim 23, wherein, for each image from the plurality of cell images, the plurality of cell parameters values for that cell image comprises two or more of: an area of the cell depicted in that image, a solidity of the cell depicted in that image, an eccentricity of the cell depicted in that image, and an equivalent diameter of the cell depicted in that image.

25. The method of claim 24 wherein, for each image from the plurality of cell images, the plurality of cell parameter values for that cell image comprises: the area of the cell depicted in that image, the solidity of the cell depicted in that image, the eccentricity of a cell depicted in that image, the equivalent diameter of the cell depicted in that image, an area of a bounding box of the cell depicted in that image, and a circular shape of the cell depicted in that image.

26. The method of any of claims 23-25, wherein for each image from the plurality of cell images, determining the plurality of parameter values for that image comprises providing that image to a parameter determination machine learning model, wherein the parameter determination machine learning model is configured to provide the plurality of parameters as its output.

27. The method of any of claims 25-26, wherein the morphology score generation function comprises a function which is directly proportional to a product of first, and second inputs and is inversely proportional to a product of a third input and a fourth input, wherein, for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing the area of the cell depicted in that image as the first input, providing the equivalent diameter of the cell depicted in that image as the second input, providing the eccentricity of the cell depicted in that image as the third input, and providing the circular shape of the cell depicted in that image as the fourth input. - 41 - 4860-7089-2770v7 Confidential - Company Proprietary28. The method of any of claims 23-26 wherein, for each image from the plurality of cell images, providing the plurality of parameter values for that image to the morphology score generation function comprises providing the plurality of parameter values for that image to a morphology score generation machine learning model.

29. The method of any of claims 23-28, further comprising determining, for each cell image from the plurality of cell images, a plurality of additional parameter values for that cell image, the plurality of additional parameter values being other than the plurality of cell parameter values.

30. The method of any of claims 23-29, further comprising: presenting a graphical interface comprising a plot displaying a plurality of points, wherein, for each point from the plurality of points, that point corresponds to a cell image from the plurality of cell images, wherein, for each point from the plurality of points, that point is displayed at a location determined based on the plurality of cell parameter values for that cell image.

31. The method of any of claims 20-30, further comprising determining, based on the scores for the plurality of cell images, an alignment characteristic score of an analyzer comprising the flow cell.

32. The method of claim 31, wherein the alignment characteristic score of the analyzer comprises a percentage of front facing cells.

33. The method of any of claims 31-32, further comprising: assessing performance of an analyzer which comprises the flowcell based on comparing the alignment characteristic score with an alignment threshold.

34. The method of claim 33 in combination with claim 30, wherein the method further comprises: - 42 - 4860-7089-2770v7 Confidential - Company Proprietarybased on the comparison of the alignment characteristic score with the alignment threshold, presenting the graphical user interface comprising the plot; and updating a design of the analyzer based on the plot.

35. The method of any of claims 20-34, wherein the method comprises flowing blood cells from the blood sample through the flowcell.

36. The method of claim 20, wherein the plurality of cell images comprises a subset of a larger set of cell images.

37. The method of claim 20, wherein the plurality of cell images comprises a subset of a larger set of cell images, wherein the larger set of cell images are associated with a focus quality, and wherein the plurality of cell images meet at least a particular focus quality threshold.

38. A computing device comprising a processor configured to perform the method of any one of claims 1 to 37.

39. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of any one of the methods of claims 1 to 37.

40. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of any one of the methods of claims 1 to 37.

41. A cell analysis system comprising: a flow cell; a camera; and a computing device; wherein the cell analysis system is configured to: - 43 - 4860-7089-2770v7 Confidential - Company Proprietaryflow blood cells from a biological sample through a flow cell; capture, by using the camera, a plurality of cell images, each cell image depicting a respective blood cell of the blood sample as it flows through the flow cell; and determine, by the computing device, an alignment characteristic score based on the plurality of cell images using a trained machine learning model.

42. The system of claim 41, wherein the method further comprises, for each image from the plurality of cell images, determining a plurality of cell parameter values for that cell image.

43. The system of claim 42, wherein, for each image from the plurality of cell images, the plurality of cell parameter values for that cell image comprise: an area of the cell depicted in that image, a solidity of the cell depicted in that image, an eccentricity of the cell depicted in that image, an equivalent diameter of the cell depicted in that image, an area of a bounding box of the cell depicted in that image, and a circular shape of the cell depicted in that image.

44. The system of claims 42 or 43, wherein, for each image from the plurality of cell images, determining the plurality of parameter values for that image comprises providing that image to a second trained machine learning model, wherein the second trained machine learning model is configured to provide the plurality of parameters as its output.

45. The system of any of claims 42-44, wherein the method the system is configured to perform comprises assessing performance of the cell analysis system based on comparing the alignment characteristic score with an alignment threshold.

46. The system of claim 45, wherein, in the method the system is configured to perform, determining the plurality of cell parameters for each image from the plurality of cell images is performed based on comparing the alignment characteristic score with the alignment threshold. - 44 - 4860-7089-2770v7 Confidential - Company Proprietary47. The system of any of claims 45-46, wherein the method the system is configured to perform comprises, based on the comparison of the alignment characteristic score with the alignment threshold, present a graphical user interface comprising a plot.

48. The system of claim 47, wherein the plot displays a plurality of points, wherein, for each point from the plurality of points, that point corresponds to a cell image from the plurality of cell images, wherein, for each point from the plurality of points, that point is displayed at a location determined based on the plurality of cell parameter values for that cell image. .

49. The system of any one of claims 41 to 48, wherein determining the alignment characteristic score based on the plurality of cell images using the trained machine learning model comprises: identifying one or more of a set of front facing cell images and a set of non-front facing cell images by, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing; and computing the alignment characteristic score by using the one or more of the set of front facing cell images and the set of non-front facing cell images.

50. The system of claim 49, wherein computing the alignment characteristic score comprises determining one or more of a cardinality of the set of front facing cell images and a cardinality of the set of non-front facing cell images.

51. The system of claims 49 or 50 wherein, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing comprises: determining a morphology score for that cell image, wherein the morphology score for that cell image indicates a front facing orientation for that cell image; and comparing the morphology score to a threshold determined using the trained machine learning model. - 45 - 4860-7089-2770v7 Confidential - Company Proprietary52. The system of claim 51 when taken in combination with any one of the claims 17 to 19, wherein, for each image from the plurality of cell images, determining the morphology score for that cell image comprises determining the morphology score for that cell image based on the plurality of cell parameter values for that cell image.

53. The system of claim 52, wherein: for each image from the plurality of images, determining the morphology score for that image comprises providing the plurality of cell parameter values for that image as input to a morphology score generation function; wherein the morphology score generation function comprises a function which is directly proportional to a product of first and second inputs, and which is inversely proportional to a product of a third and a fourth input, wherein, for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing the area of the cell depicted in that image as the first input, providing the equivalent diameter of the cell depicted in that image as the second input, providing the eccentricity of the cell depicted in that image as the third input, and providing the circular shape of the cell depicted in that image as the fourth input.

54. The system of any one of the claims 41 to 53, wherein the method further comprises determining, for each cell image from the plurality of cell images, a plurality of additional parameter values for that cell image, the plurality of additional parameter values being other than the plurality of cell parameter values.

55. The system of any one of claims 42 to 54, wherein, for each image from the plurality of cell images, using the trained machine learning model to determine whether that image is front facing comprises providing the plurality of parameter values for that image as input to the trained machine learning model. - 46 - 4860-7089-2770v7 Confidential - Company Proprietary56. The system of any one of the claims 41 to 55, wherein the alignment characteristic score comprises information specifying one or more of an estimate of an amount of front facing cells in the sample and an estimate of a fraction of front facing cells in the sample.

57. The system of any one of the claims 41 to 56, wherein the alignment characteristic score comprises a percentage of front facing cells.

58. A cell analysis system comprising: a flow cell; a camera; and a computing device ; wherein the cell analysis system is configured to: flow blood cells from a biological sample through a flow cell; capture, by using the camera, a plurality of cell images, each cell image depicting a respective blood cell of the blood sample as it flows through the flow cell; and calculate a score for each cell image, wherein the score represents a front facing orientation assessment.

59. The system of claim 58, wherein, for each image from the plurality of cell images, the cell depicted in that cell image is a red blood cell.

60. The system of any of claims 58-59, wherein calculating the score for each cell image comprises providing each cell image to a machine learning model trained to determine front facing scores.

61. The system of any of claims 58-59, wherein, for each image from the plurality of cell images, determining the score for that cell image comprises determining a plurality of cell parameter values for that image, and providing the plurality of cell parameter values for that image as input to a morphology score generation function. - 47 - 4860-7089-2770v7 Confidential - Company Proprietary62. The system of claim 61, wherein, for each image from the plurality of cell images, the plurality of cell parameters values for that cell image comprises two or more of: an area of the cell depicted in that image, a solidity of the cell depicted in that image, an eccentricity of the cell depicted in that image, and an equivalent diameter of the cell depicted in that image.

63. The method of claim 62 wherein, for each image from the plurality of cell images, the plurality of cell parameter values for that cell image comprises: the area of the cell depicted in that image, the solidity of the cell depicted in that image, the eccentricity of a cell depicted in that image, the equivalent diameter of the cell depicted in that image, an area of a bounding box of the cell depicted in that image, and a circular shape of the cell depicted in that image.

64. The system of any of claims 61-63, wherein for each image from the plurality of cell images, determining the plurality of parameter values for that image comprises providing that image to a parameter determination machine learning model, wherein the parameter determination machine learning model is configured to provide the plurality of parameters as its output.

65. The system of any of claims 63-64, wherein the morphology score generation function comprises a function which is directly proportional to a product of first, and second inputs and is inversely proportional to a product of a third input and a fourth input, wherein, for each image from the plurality of cell images, providing the plurality of cell parameter values for that cell image as input to the morphology score generation function comprises providing the area of the cell depicted in that image as the first input, providing the equivalent diameter of the cell depicted in that image as the second input, providing the eccentricity of the cell depicted in that image as the third input, and providing the circular shape of the cell depicted in that image as the fourth input.

66. The system of any of claims 62-65 wherein, for each image from the plurality of cell images, providing the plurality of parameter values for that image to the morphology score - 48 - 4860-7089-2770v7 Confidential - Company Proprietarygeneration function comprises providing the plurality of parameter values for that image to a morphology score generation machine learning model.

67. The system of any of claims 61-66, wherein the system is further configured to determine, for each cell image from the plurality of cell images, a plurality of additional parameter values for that cell image, the plurality of additional parameter values being other than the plurality of cell parameter values.

68. The system of any of claims 61-67, wherein the system is further configured to: present a graphical interface comprising a plot displaying a plurality of points, wherein, for each point from the plurality of points, that point corresponds to a cell image from the plurality of cell images, wherein, for each point from the plurality of points, that point is displayed at a location determined based on the plurality of cell parameter values for that cell image.

69. The system of any of claims 58-68, wherein the system is further configured to determine, based on the scores for the plurality of cell images, an alignment characteristic score of an analyzer comprising the flow cell.

70. The system of claim 69, wherein the alignment characteristic score of the analyzer comprises a percentage of front facing cells.

71. The system of any of claims 69-70, wherein the system is further configured to: assess performance of an analyzer which comprises the flowcell based on comparing the alignment characteristic score with an alignment threshold.

72. The system of claim 71 in combination with claim 30, wherein the system is further configured to: based on the comparison of the alignment characteristic score with the alignment threshold, present the graphical user interface comprising the plot. - 49 - 4860-7089-2770v7 Confidential - Company Proprietary73. The system of claim 58, wherein the plurality of cell images comprises a subset of a larger set of cell images.

74. The system of claim 58, wherein the plurality of cell images comprises a subset of a larger set of cell images, wherein the larger set of cell images are associated with a focus quality, and wherein the plurality of cell images meet at least a particular focus quality threshold. - 50 - 4860-7089-2770v7 Confidential - Company Proprietary

Images