Measuring a presence of a target cell phenotype to predict patient response to therapy

Abstract

Techniques for measuring a presence of cells of a target phenotype in a patient tissue sample are presented. The techniques include: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample; determining respective neighborhood signatures for a plurality of locations in the patient tissue sample image, from which a plurality of neighborhood signatures are obtained; assigning respective discriminants to at least some of the plurality of neighborhood signatures; calculating a proportion of the plurality of locations, where the proportion is based on a number of locations that correspond to a neighborhood signature discriminant that is above a threshold and a number of locations that correspond to a neighborhood signature discriminant that is below the threshold; and providing the proportion as a proxy for a density of cells of the target phenotype in the patient tissue sample.

Description

Attorney Docket No. P18284-02 / 0184.0312-PCTMEASURING A PRESENCE OF A TARGET CELL PHENOTYPE TO PREDICT PATIENT RESPONSE TO THERAPYGovernment Support

[0001] This invention was made with government support under cooperative agreement No. ACI-1261715, awarded by the National Science Foundation. The government has certain rights in the invention.Related Application

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 738,573, entitled “MEASURING A PRESENCE OF A TARGET CELL PHENOTYPE TO PREDICT PATIENT RESPONSE TO THERAPY,” and filed December 24, 2024.Field

[0003] This disclosure relates generally to measurements in a biological context.Background

[0004] Tissue samples in a medical setting may be analyzed by obtaining a thin slice thereof. Such slices may be stained with any of a variety of materials to highlight various features. A slice may then be photographed using specialized equipment to obtain a whole-slide image, which is non-transiently electronically stored. A given whole-slide image may depict hundreds of thousands or even millions of cells.

[0005] In biology, each cell is associated with its phenotype, or set of observable characteristics. A given tissue sample may contain cells with a number of different phenotypes. Cells of some phenotypes may be more common than cells of other phenotypes. If cells of a particular phenotype are sufficiently rare in a tissue sample, it may be difficult to detect their presence and determine their density based on a whole-slide image, which only includes cells present along a slice through the tissue sample.Summary

[0006] According to various embodiments, a method of measuring a presence of cells of a target phenotype in a patient tissue sample is presented. The method includes: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample; determining respective neighborhood signatures for a plurality of locations in the patient tissue sample image, from which a plurality of neighborhood signatures are obtained; assigning respective discriminants to at least some of the plurality of neighborhood signatures; calculating a proportion of the plurality of locations, where the proportion is based on a number of locations that correspond to a neighborhood signature discriminant that is above a threshold and a number of locations that correspond to a neighborhood signature discriminant that is below the threshold; and providing the proportion as a proxy for a density of cells of the target phenotype in the patient tissue sample.

[0007] Various optional features of the above method embodiments include the following. The method may include predicting, based on the proportion, a patient response to a therapy. The method may include treating the patient with the therapy based on the predicting. The therapy may include at least one of an immunotherapyor a chemotherapy. The target phenotype may be a T-cell phenotype. A respective neighborhood signature may be based on at least a tally of cells in a respective region proximate to a location. A respective neighborhood signature may include a respective cell phenotype tally for a respective region proximate to a location and a respective distance to a boundary of interest. Each respective cell phenotype tally may include a number of cells of the target phenotype and a number of cells for at least one other phenotype. The boundary of interest may include a tumor boundary. Each respective region may include an area between two concentric circles. The method may include determining each respective discriminant based on a training tissue sample image. The method may include determining a pathology characterization of the patient tissue sample based on the proportion.

[0008] According to various embodiments, a system for measuring a presence of cells of a target phenotype in a patient tissue sample is presented. The system includes: a non-transitory computer readable medium including instructions; and at least one electronic processor that executes the instructions to perform operations including: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample; determining respective neighborhood signatures for a plurality of locations in the patient tissue sample image, from which a plurality of neighborhood signatures are obtained; assigning respective discriminants to at least some of the plurality of neighborhood signatures; calculating a proportion of the plurality of locations, where the proportion is based on a number of locations that correspond to a neighborhood signature discriminant that is above a threshold and a number of locations that correspond to a neighborhood signature discriminant that is below the threshold; and providing the proportion as a proxy for a density of cells of the target phenotype in the patient tissue sample.

[0009] Various optional features of the above system embodiments include the following. The operations may further include predicting, based on the proportion, a patient response to a therapy. The patient may be treated with the therapy based on the predicting. The therapy may include at least one of an immunotherapy of a chemotherapy. The target phenotype may be a T-cell phenotype. A respective neighborhood signature may be based on at least a tally of cells in a respective region proximate to a location. A respective neighborhood signature may include a respective cell phenotype tally for a respective region proximate to a location and a respective distance to a boundary of interest. Each respective cell phenotype tally may include a number of cells of the target phenotype and a number of cells for at least one other phenotype. The boundary of interest may include a tumor boundary. Each respective region may include an area between two concentric circles. The operations may further include determining each respective discriminant based on a training tissue sample image. The operations may further include determining a pathology characterization of the patient tissue sample based on the proportion.

[0010] According to various embodiments, a method of predicting a response to immunotherapy is presented. The method includes: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample stained to identify: tumor cells, cyto-toxic T cells, macrophages, and regulatory T cells; identifying automatically, in the patient tissue sample image, incidences of tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells; determining, based on spatial arrangements of the tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells, a score that correlates with a likelihood of response to and immunotherapy; and providing an indication of the score.

[0011] Various optional features of the above method embodiments include the following. The immunotherapy may include an anti-PD1 therapy. The method may include treating the patient with the anti-PD1 therapy based on the score. The patient tissue sample may be stained with at least one of a fluorogenic stain or a chromogenic stain. The patient tissue sample may be stained based on antibodies for CD8, CD163, Foxp3, cytokeratin, melanocytes, or a cancer cell type. The spatial arrangements may be based on a presence of identified cells within annular regions. The spatial arrangements may be based on a presence of identified cells within geometric regions. The patient tissue sample may be stained based on RNA associated with at least one of: tumor cells, cyto-toxic T cells, macrophages, or regulatory T cells. The method may include including segmenting the patient tissue sample image. The identifying automatically may use machine learning. The method may include parameterizing and indexing spatial data regarding the patient tissue sample image in a database.

[0012] According to various embodiments, a system for predicting a response to immunotherapy is presented. The system includes: a non-transitory computer readable medium including instructions; and at least one electronic processor that executes the instructions to perform operations including: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample stained to identify: tumor cells, cyto-toxic T cells, macrophages, and regulatory T cells; identifying automatically, in the patient tissue sample image, incidences of tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells; determining, based on spatial arrangements of the tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells, a score that correlates with a likelihood of response to and immunotherapy; and providing an indication of the score.

[0013] Various optional features of the above system embodiments include the following. The immunotherapy may include an anti-PD1 therapy. The patient may be treated with the anti-PD1 therapy based on the score. The patient tissue sample may be stained with at least one of a fluorogenic stain or a chromogenic stain. The patient tissue sample may be stained based on antibodies for CD8, CD163, Foxp3, cytokeratin, melanocytes, or a cancer cell type. The spatial arrangements may be based on a presence of identified cells within annular regions. The spatial arrangements may be based on a presence of identified cells within geometric regions. The patient tissue sample may be stained based on RNA associated with at least one of: tumor cells, cyto-toxic T cells, macrophages, or regulatory T cells. The operations may further include segmenting the patient tissue sample image. The identifying automatically may use machine learning. The operations may further include parameterizing and indexing spatial data regarding the patient tissue sample image in a database.

[0014] Combinations, (including multiple dependent combinations) of the above-described elements and those within the specification have been contemplated by the inventors and may be made, except where otherwise indicated or where contradictory.Brief Description of the Drawings

[0015] Various features of the examples can be more fully appreciated, as the same become better understood with reference to the following detailed description of the examples when considered in connection with the accompanying figures, in which:

[0016] Fig. 1 illustrates a tissue sample and slice thereof according to various embodiments;

[0017] Fig. 2 depicts segmentation and phenotypical ascription of cells in a whole-slide image, according to various embodiments;

[0018] Fig. 3 schematically illustrates a non-limiting example neighborhood signature for a location in an image of a tissue sample, according to various embodiments;

[0019] Fig. 4 is a flow diagram of method of determining respective neighborhood signature discriminants for a plurality of neighborhood signatures in an image of a tissue sample, according to various embodiments; and

[0020] Fig. 5 is a flow diagram of a method of measuring a presence of cells of a target phenotype in a patient tissue sample, according to various embodiments.Description of the Examples

[0021] Reference will now be made in detail to example implementations, illustrated in the accompanying drawings. Wherever convenient, the same reference numbers will be used throughout the drawings to refer to the same or like parts. In the following description, reference is made to the accompanying drawings that form a part thereof, and in which is shown by way of illustration specific exemplary examples in which the invention may be practiced. These examples are described in sufficient detail to enable those skilled in the art to practice the invention and it is to be understood that other examples may be utilized and that changes may be made without departing from the scope of the invention. The following description is, therefore, merely exemplary.

[0022] Various embodiments provide techniques for measuring a presence of cells of a particular phenotype, referred to as a “target phenotype,” in a biological tissue sample. Some embodiments quantify a presence of target phenotype cells in a tissuesample by providing a value that represents a proxy for a density of the target phenotype cells in the tissue sample. Some embodiments may quantify a presence of target phenotype cells in a tissue sample even though the target phenotype cells are relatively rare in the tissue sample. In particular, some embodiments may quantify a presence of target phenotype cells in a tissue sample, even though no target phenotype cells appear in a whole-slide image of the tissue sample, by detecting a presence of target phenotype cell niches. For example, each target phenotype cell may have a local neighborhood that has different properties in comparison to tissue that is more remote from target phenotype cells. Such a local neighborhood of a target phenotype cell represents a niche that is detectable by way of a neighborhood signature, even though the target phenotype cell itself might not be detectable, or might not be as detectable as the niche. In particular, some embodiments overcome the prior art lack of ability to detect rare phenotype cells by determining neighborhood signatures for locations in a tissue sample image, assigning to the neighborhood signatures discriminants, and then calculating a proportion of the locations based on comparing their discriminants to a threshold, where the proportion measures a presence of the rare phenotype cells.

[0023] Embodiments have a variety of use cases. Some embodiments may be used to detect a presence of target phenotype cells in a tissue sample, where the target phenotype is of an immune cell, such as a T-cell. According to some embodiments, a proxy for density of particular immune cells may be used to predict a patient’s response to a corresponding immunotherapy and / or chemotherapy treatment. According to some embodiments, a quantification of a presence of target phenotype cells may be used as a pathology characterization of a tissue sample.

[0024] These and other features and advantages are shown and described herein in reference to the accompanying figures.

[0025] Fig. 1 illustrates a tissue sample 102 and slice 112 thereof according to various embodiments. In particular, Fig. 1 illustrates target phenotype cells (e.g., 104) that are relatively rare in the tissue sample 102. By way of non-limiting example, three target phenotype cells (e.g., 104) are shown in the tissue sample 102 in Fig. 1. None of the target phenotype cells (e.g., 104) intersect the slice 116. Therefore, none of the target phenotype cells (e.g., 104) would appear in a whole-slide image of the slice 112.

[0026] As shown in Fig. 1 , each target phenotype cell (e.g., 104) is included in a respective niche (e.g., 106) that includes the cells of the tissue sample in a small surrounding neighborhood. As described in detail herein, niches may have certain properties, referred to as neighborhood signatures, that are detectable even if the enclosed target phenotype cell is not detected. Fig. 1 illustrates that although none of the target phenotype cells (e.g., 104) intersect the slice 112, a portion of a target phenotype cell niche 116 does intersect the slice 112. Therefore, the target phenotype cell niche may be detectable in the slice 112.

[0027] Thus, as shown in Fig. 1 , a target phenotype cell (e.g., 104) may be sufficiently rare in a three-dimensional tissue sample 102 such that it does not appear in a given two-dimensional slice 116 of the tissue sample. Nevertheless, some embodiments are able to measure a presence of target phenotype cells in a three- dimensional tissue sample based on detecting a portion of target phenotype cell niches (e.g., 106) in a two-dimensional slice of the three-dimensional tissue sample, even if the two-dimensional slice itself does not include any target phenotype cells.

[0028] Fig. 2 depicts segmentation and phenotypical ascription of cells in a whole-slide image 202, according to various embodiments. The whole-slide image202 shown in Fig. 2 is a whole-slide multiplex immunofluorescence (mIF) scan. The whole-slide image 202 includes thousands of constituent tiles, e.g., 204. Some embodiments utilize whole-slide images (e.g., 202) in which each individual cell is segmented and ascribed with a phenotype in a process referred to as single-cell mapping, as illustrated by 206. Single-cell mapping may be performed using, for example, the AstroPath platform, which is capable of generating spatially resolved mIF datasets. In general, the single-cell mapping process is performed by executing software that inputs a whole-slide image and outputs a data set that represents the segmented and phenotype-ascribed cells within the whole-slide image.

[0029] Various embodiments may operate on image datasets for whole-slide images that have been segmented and have had cellular phenotypes ascribed using a computer, as shown in Fig. 2. For example, Fig. 3 illustrates using such an image dataset to determine a neighborhood signature.

[0030] Fig. 3 schematically illustrates a non-limiting example neighborhood signature 360 for a location 320 in an image 310 of a tissue sample, according to various embodiments. The tissue sample may be a training tissue sample (e.g., as described herein in reference to Fig. 4) or a patient tissue sample (e.g., as described herein in reference to Fig. 5). The image 310 may be a whole-slide image, for example. The image 310 may depict a particular region of interest 302, which may be a tumor, by way of non-limiting example. Note that the neighborhood signature 360 is for a particular location 320 in the image 310. The location 320 may be the location of a particular cell, or may be any other location, without respect to whether it corresponds to a cell.

[0031] The neighborhood signature 360 illustrated in Fig. 3 includes two parts, by way of non-limiting example. The first part of the example neighborhood signature360 is a tally 350 of nearby cells according to phenotype, and the second part is a measure of proximity of the location 320 to a boundary of the region of interest 302. Note that the parts within the neighborhood signature 360 may be presented in any order subject to consistent convention.

[0032] The tally 350 part of the example neighborhood signature 360 may be determined as follows. First, two concentric circles around the location 320 are considered such that the outer concentric circle includes the centers (e.g., centroids) of multiple cells. By way of non-limiting example, the diameter of the inner circle may be 5pm and the diameter of the of the outer circle may be 10pm. Cells in the image 310 whose centers (e.g., centroids) lie between the two circles are then tallied according to phenotype. The generated tally 350 is then used as the first part of the neighborhood signature 360. As shown in Fig. 3, the tally may be represented as a list (1 , 2, 2, 1 ) that includes the total numbers of cells of each of the four phenotypes that appear inside of the larger circle and outside of the smaller circle. Note that the value of 5pm may be chosen for the diameter of the inner circle so as to exclude from the tally 350 a cell at the center of the inner circle. According to some embodiments, neighborhood signatures may be determined at locations that correspond to cell centers (e.g., centroids); according to such embodiments, the cell or cells whose centers (e.g., centroids) are within the smaller concentric circle may be excluded from the tally 350.

[0033] The second part of the example neighborhood signature 360 illustrated by Fig. 3 may be determined as follows. The image 310 may be partitioned into bins according to distance from the boundary of the region of interest 302. By way of nonlimiting example, the unit of distance used for such a partition may be 25pm. By way of non-limiting example, four bins are shown in Fig. 3; fewer or additional bins may beconsidered in an embodiment. Bin 3 corresponds to the region that is inside of, and within 25pm of the boundary of, the region of interest 302. Bin 2 corresponds to the region that is inside, and within 25pm of the boundary, of bin 3. Bin 1 corresponds to the region that is inside, and within 25pm of the boundary, of bin 2. Bin 4 corresponds to the region that is outside, and within 25pm of the boundary, of the region of interest 302. Additional bins may be included, e.g., for regions further outside of the region of interest 302. For the example neighborhood signature 360 shown in Fig. 3 for the location 320, the location 320 is a distance 330 from the boundary of the region of interest 302, which falls within bin 3. Note that the second part of the example neighborhood signature 360 may be in any of a variety of forms, such as bin number (e.g., “bin 3”) or distance quantum (e.g., “+50pm”). For the latter example form, the sign may indicate whether the respective location is inside or outside of the region of interest 302.

[0034] Thus, the example neighborhood signature 360 as shown in Fig. 3 may be represented as, by way of non-limiting example, ((1 , 2, 2, 1 ), bin2).

[0035] Note that the neighborhood signature 360 shown and described in reference to Fig. 3 is a non-limiting example. Other neighborhood signatures according to various embodiments may include or utilize other parameters. For example, instead of a tally of cells according to phenotype, a neighborhood signature may include a tally of cells according to any of a variety of different properties, such as necrotic / living, cell marker expression, etc. As another example, a neighborhood signature may include a distance to a boundary other than a boundary of a region of interest 302, such as a regression region boundary (e.g., a boundary of a region of residual fibrotic tissue remaining subsequent to a tumor regression). As another example, a neighborhood signature may include a distance to an anatomic structure,such as, by way of non-limiting example, a blood vessel. According to some embodiments, multiple distances corresponding to multiple different boundaries may be included in a neighborhood signature. According to some embodiments, a neighborhood signature may lack a distance parameter and only include a tally of cells according to phenotype, where multiple phenotypes are represented. Other types and arrangements of neighborhood signatures are also possible.

[0036] Fig. 4 is a flow diagram of method 400 of determining respective neighborhood signature discriminants for a plurality of neighborhood signatures in an image of a tissue sample, according to various embodiments. At least some of the actions of the method 400 must be performed by a computer due to the impracticality of performing its actions relative to the very large number of cells in a typical wholeslide image. The method 400 may be used to obtain neighborhood signature discriminants, which may be used in a method of measuring a presence of a target cell phenotype in a patient tissue sample, e.g., as shown and described herein in reference to Fig. 5.

[0037] At 402, the method 400 includes obtaining a training tissue sample image. The training tissue sample may be a tissue sample that includes a region of interest, e.g., a tumor. The training tissue sample image may be obtained from an individual that is independent of any patient for which the neighborhood signature discriminants obtained using the method 400 are used. The training tissue sample image may be a whole-slide image or a subsampled whole-slide image, by way of nonlimiting examples.

[0038] At 404, the method 400 includes segmenting and ascribing phenotypes to the cells in the training tissue sample image. Note that the actions of 408 must be performed using a computer due to the impracticality of performing them manually.For example, the actions of 404 may be performed using, for example, computer- assisted phenotyping, e.g., using the AstroPath platform. According to some embodiments, the training tissue sample image may be obtained at 402 presegmented and / or with cellular phenotypes already ascribed; therefore, some or all of the actions of 404 are optional for some embodiments.

[0039] At 406, the method 400 includes determining a respective neighborhood signature for each cell in the training tissue sample image. The neighborhood signatures may be determined as shown and described herein in reference to Fig. 3. Note that multiple different cells may have the same neighborhood signature. For example, more than one cell may have the same tally of cells according to phenotype in their proximity and be in the same boundary-of-interest distance bin. Thus, the actions of 406 result in a set of neighborhood signatures for the cells in the training tissue sample image, where each neighborhood signature corresponds to at least one, and possibly multiple, cells in the training tissue sample image. Note that the actions of 406 must be performed by a computer due to the impracticality of performing them manually.

[0040] At 408, the method 400 includes, for each neighborhood signature from 406, counting signal and background cells. The signal cells are the cells that have a target phenotype, and the background cells are the cells that have a phenotype other than the target phenotype. Thus, for a given neighborhood signature from 406, the actions of 408 include counting the number of signal cells in the training tissue sample image with the given neighborhood signature and counting the number of background cells with the given neighborhood signature. Note that the actions of 408 must be performed by a computer due to the impracticality of performing them manually. Thecounts described herein may be non-transiently stored in a data structure in association with identification of the associated neighborhood signatures.

[0041] At 410, the method 400 includes computing a respective neighborhood signature discriminant for each of the neighborhood signatures from 406. For a given neighborhood signature / VS, the respective neighborhood signature discriminant D( / VS) may be computed as follows, by way of non-limiting example, DQVS) =represents the count of signal cells with neighborhood signature NS, NbackgroundNS) represents the count of background cells with neighborhood signature NS, and c is a parameter that may be selected by a user. Note that the non-limiting example D( / VS) as presented herein is a number between 0 and 1 , however, other discriminant calculations may yield other values. In general, a neighborhood signature discriminant for a neighborhood signature of a particular cell may represent a ratio of a probability that the particular cell is a signal cell to the probability that the particular cell is a background cell. Note that the actions of 410 must be performed by a computer due to the impracticality of performing them manually. Thus, the neighborhood signature discriminants described herein may be non-transiently stored in a data structure in association with identifications of the associated neighborhood signatures (or the associated neighborhood signatures themselves).

[0042] Fig. 5 is a flow diagram of a method 500 of measuring a presence of a target cell phenotype in a patient tissue sample, according to various embodiments. At least some of the actions of the method 500 must be performed by a computer due to the impracticality of performing its actions relative to the very large number of cells in a typical whole-slide image. The target cell phenotype may be a particular T-cell phenotype, for example.

[0043] At 502, the method 500 includes obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample. The patient tissue sample may be a tissue sample that includes a region of interest, e.g., a tumor, a portion of the slide (such as a single high power field), a region annotated by a pathologist, etc. The patient tissue sample may be a biopsy or another type of tissue resection from a patient. The patient tissue sample image may be a whole-slide image, for example. The actions of 502 may include obtaining and / or scanning a patient tissue sample, according to various embodiments.

[0044] At 504, the method 500 includes segmenting and ascribing phenotypes to the cells in the training tissue sample image. Note that the actions of 504 must be performed using a computer due to the impracticality of performing them manually. For example, the actions of 504 may be performed using, for example, the AstroPath platform or another image analysis suite. According to some embodiments, the patient tissue sample image may be obtained at 502 pre-segmented and / or with cellular phenotypes already ascribed; therefore, some or all of the actions of 504 are optional for some embodiments.

[0045] At 506, the method 500 includes selecting a plurality of locations in the patient tissue sample image. The plurality of locations may be selected according to a regular pattern, e.g., a grid, may be selected randomly according to any of a variety of statistical distributions (e.g., Gaussian, Poisson, uniform, etc.), or may be selected to correspond to some or all cells in the patient tissue sample image. According to some embodiments, the plurality of locations are selected according to a uniform distribution with approximately the same density as the cells represented in the patient tissue sample image.

[0046] At 508, the method 500 includes determining respective neighborhood signatures for the plurality of locations in the patient tissue sample image from 506. Each respective neighborhood may be determined as shown and described herein in reference to Fig. 3. By way of non-limiting example, each respective neighborhood signature may be based on at least a tally of cells in a respective region proximate a respective location of the plurality of locations. The actions of 506 result in obtaining a plurality of neighborhood signatures, which may be non-transiently stored in electronic memory.

[0047] At 510, the method 500 includes assigning respective neighborhood signature discriminants to at least some of the plurality of neighborhood signatures. The actions of 508 may utilize an output from the method 400 as shown and described herein in reference to Fig. 4. For example, the actions of 508 may utilize a non- transiently electronically stored data structure that associates neighborhood signature discriminants with identifications of the associated neighborhood signatures (or the associated neighborhood signatures themselves). Thus, the actions of 508 may assign respective neighborhood signature discriminants determined using a training tissue sample image to neighborhood signatures determined for the patient tissue sample image, where the training tissue sample is from an individual different from the patient. The assignment may be non-transiently stored electronically in a data structure that associates respective neighborhood signature discriminants to at least some of the neighborhood signatures.

[0048] Note that the actions of 508 may occasionally not assign a neighborhood signature discriminant to one or more locations of the plurality of locations, e.g., for a location that has a neighborhood signature that does not appear in the data structure that associates neighborhood signature discriminants with identifications of theassociated neighborhood signatures. This situation may occur when the training tissue sample image does not have identical neighborhood signatures to those from the patient tissue sample image; nevertheless, perfect identity is not required according to various embodiments.

[0049] At 512, the method 500 includes calculating a proportion of the plurality of locations as described presently. The actions of 512 utilize a threshold, which may be preselected or may be selected at 512. The actions of 512 may include determining a number of the locations selected at 506 that correspond to a neighborhood signature discriminant that is above the threshold, and determining a number of locations of the locations selected at 506 that correspond to a neighborhood signature discriminant that is below the threshold. The proportion may be calculated as a ratio (e.g., fraction) of these two numbers.

[0050] At 514, the method 500 includes providing the proportion as a proxy for a density of cells of the target phenotype in the patient tissue sample. The proportion may be provided by displaying on a monitor, by storage in non-transitory electronic storage, or by providing to a system, e.g., a system that performs the actions of 516.

[0051] At 516, the method 500 includes predicting, based on the proportion, a patient response to a therapy. Because the proportion is a proxy for a density of cells, for embodiments for which the target phenotype is an immune cell phenotype, e.g., a particular T-cell phenotype, the proportion may be directly indicative of a patient response to immunotherapy that utilizes such immune cells. Alternately, or in addition, the proportion may be indicative of a patient’s response to chemotherapy. Accordingly, the actions of 516 may include comparing the proportion to a value that is known to be indicative of a patient response to a therapy. Such comparison may include determining whether the proportion is greater than or less than the value.

[0052] At 518, the method 500 includes treating the patient with the therapy based on the predicting. For example, the actions of 518 may include treating the patient with an immunotherapy, such as neoadjuvant anti-PD-1 therapy, and / or a chemotherapy, such as docetaxel and / or platinum doublet, based on the results of 516. Specific examples are presented below in reference to particular use cases.

[0053] Note that, according to various embodiments, the actions of 502, 504, 506, 508, 510, and 512 of method 500 may be used for a variety of purposes, not limited to predicting a patient’s response to immunotherapy or chemotherapy. According to some embodiments, the actions of 502, 504, 506, 508, 510, and 512 may be performed on a patient tissue specimen image, e.g., a biopsy, and may be followed by determining a pathology characterization of the patient tissue sample based on the proportion. Example pathology characterizations include sub-classification of tumor type or disease process due to spatial arrangements of cells, an assessment of the patient’s response to a therapy after treatment has been initiated, and / or monitoring the patient after a treatment has been initiated.

[0054] The following presents a non-limiting first example use case of an embodiment. The first example use case concerns determining a patient response to a particular non-small cell lung carcinoma (NSCLC) immunotherapy that utilizes CD8+FoxP3+ cells. One of the historic barriers to detecting and studying CD8+FoxP3+ cells is that they are extremely rare. CD8+FoxP3+ cells represent approximately 0.4 and 0.1 % of peripheral blood T-cells in humans and mice, respectively. Within the tumor microenvironment, prevalence rates of 0.2-5.2% of all CD8+ cells have been reported in melanoma, NSCLC, and cervical carcinoma. As described in reference to the first example use case, an embodiment facilitated theidentification of such statistically rare events. An overview of the first example use case is provided presently, followed by details of the first example use case.

[0055] The recent approval of neoadjuvant PD-1 -based therapy for NSCLC has highlighted the high unmet need for biomarkers to predict clinical efficacy. A 6-plex immunofluorescence assay was applied to paired pre-treatment biopsies and on- treatment resections from patients receiving neoadjuvant anti-PD-1 therapy. Wholeslide analysis using AstroPath allowed for data collection on over 31 million cells across over 48,000 high powered fields. The presence of CD8+FoxP3+ cells was associated with a subsequent major pathologic response (MPR) (area under curve; AUC=0.79) and improved recurrence-free survival (RFS, 83 vs. 21 months). These findings were replicated in a separate cohort of patients with advanced NSCLC, where high densities of CD8+FoxP3+ cells associated with improved five-year overall survival (OS), (p<0.05). The contact neighbors of CD8+FoxP3+ cells demonstrated an immunologically active niche with comparable predictive power to CD8+FoxP3+ numbers themselves (AUC=0.80 for MPR and RFS 83 vs. 21 months for neoadjuvant- treated; OS, p<0.05 for advanced disease). In particular, the use of the method 500 in this context captured the complexity of the tumor microenvironment and outperformed existing biomarkers. A detailed description of the first example use case follows.

[0056] Anti-PD-1 / L1 immune checkpoint blockade (ICB) is now the standard first-line therapy for patients with many advanced cancers, including non-small cell lung cancer (NSCLC). Neoadjuvant and perioperative anti-PD-1 / L1 blockade in patients with earlier stage, resectable disease has also been shown to improve event- free survival. Neoadjuvant / Perioperative ICB leverages the pre-operative period to activate the immune response to tumor neoantigens, generating a circulating memoryimmune response that persists following tumor resection. The first phase 1 / 2 study of neoadjuvant anti-PD1 therapy showed a major pathologic response (MPR, defined as <10% residual viable tumor, RVT) in the surgical resection specimens from 45% of patients. Subsequent studies have confirmed efficacy in this setting, including positive phase 3 trials for patients treated with neoadjuvant chemo-immunotherapy resulting in FDA approval.

[0057] For patients with advanced NSCLC, PD-L1 immunohistochemistry has been shown to enrich for response to anti-PD-1 / L1 . By contrast, there are currently no clinically approved biomarkers for neoadjuvant ICB, and there have been conflicting reports regarding the predictive value of PD-L1 immunohistochemistry (IHC) expression in this setting. Multiplex immunofluorescence (mIF) allows visualization of several immune markers simultaneously while preserving tissue architecture and has been shown in meta-analyses to outperform PD-L1 IHC and other biomarker strategies. Using mIF, a rare population of CD8+FoxP3+ cells highly associated with response and improved long-term survival following anti-PD-1 therapy in advanced melanoma has been identified. CD8+FoxP3+ cells are most commonly considered a CD8+ variant of a classical CD4+FoxP3+ regulatory T-cell that suppress effector T- cell responses. However, a small number of studies have suggested that these cells have a productive role in antitumor immune responses.

[0058] The AstroPath platform for biomarker discovery has been optimized for mapping whole slides stained with mIF to support high throughput generation of spatially resolved single-cell data. The first example use case leveraged the AstroPath platform to explore the association of pre-treatment CD8+FoxP3+ cells with clinical outcomes in patients with NSCLC receiving anti-PD-1 therapy in both the neoadjuvant and advanced disease settings. Spatial analyses according to the method 500 shownand described in reference to Fig. 5 was performed to identify a CD8 T cell-enriched niche surrounding CD8+FoxP3+ cells in the tumor microenvironment (TME). The method was used to detect the CD8+FoxP3+ cell niche, including in the absence of an actual CD8+FoxP3+ cell. The CD8+FoxP3+ niche was more abundant in the pretreatment TME than CD8+FoxP3+ cells themselves and was equally effective in predicting response to anti-PD-1 , with reduced statistical error.

[0059] A quantitatively validated six-marker multiplex mIF assay (PD-1 , PD-L1 , CD8, FoxP3, CD163 and pan-CK) was performed on pre-treatment biopsies (n=25) from patients who received neoadjuvant anti-PD-1 -based therapy prior to surgical resection, i.e. “Cohort 1.” Using the AstroPath platform, spatially-resolved single-cell data from 1 ,659,155 cells was collected from 4,150 HPFs. The association of the pretreatment cell densities for each lineage (CD163+, CD8+, CD8+FoxP3+, CD8(-)FoxP3+, and pan-CK+) with on-treatment MPR was assessed by calculating the areas under the curve (AUC) for the receiver operator characteristic (ROC) curve. Among all lineages evaluated, the highest AUC value (0.79) was observed for the CD8+FoxP3+ cell population. CD8+FoxP3+ cells were rare overall (3.4% of all CD8+ T-cells and 0.11 % of all cells within the pre-treatment TME) and were significantly enriched in the pre-treatment biopsies of tumors that exhibited a subsequent major pathologic responses to neoadjuvant anti-PD-1 -based therapy. Refined CD8+FoxP3+ cellular subsets based on expression of PD-1 and PD-L1 further strengthened the association with MPR, e.g., CD8+FoxP3+PD-1 + (AUC 0.83) and CD8+FoxP3+PD- L1 (-) (AUC 0.81 ). The median recurrence-free survival (RFS) among patients with CD8+FoxP3+ cells was also improved (>0 / mm2vs. 0 / mm2; 83 vs. 21 months. Those with CD8+FoxP3+ cells have shown a durable improvement in OS (89% vs. 50% 2- year OS, respectively; p=0.05). 1

[0060] Pre-treatment biopsies from two cohorts of patients with advanced NSCLC were assessed to extend the understanding of CD8+FoxP3+ cells across tumor stage and treatment settings. Patients in Cohort 2 were enrolled in a clinical trial of anti-PD-1 for advanced, pre-treated, non-squamous NSCLC. This analysis of n=14 patients included 6,774,394 cells collected from 5,394 HPFs. Patients with high CD8+FoxP3+ cell densities had significantly improved overall survival (OS) (100% vs. 20% 2-year OS, respectively; p=0.046).

[0061] To benchmark CD8+FoxP3+ cell density performance against the current clinical standard, the association with survival outcomes was compared to that of tumor cell PD-L1 using a 50% threshold. In both early-stage (neoadjuvant, Cohort 1 ) and advanced NSCLC (Cohort 2), CD8+FoxP3+ cell assessment improved stratification of patient overall survival (p=0.01 and p=0.04, respectively) compared to PD-L1 alone, suggesting non-redundant contributions.

[0062] The AstroPath tumor-immune maps were used to characterize CD8+FoxP3+ cell contact neighbors, i.e., the CD8+FoxP3+ niche. Specifically, the target phenotype was CD8+FoxP3+ cells as employed in the method 500. The relative proportion of each of the contact neighbors was determined to characterize the CD8+FoxP3+ cell niche and was compared to the entire pre-treatment TME. Across all three pre-treatment cohorts, CD8+FoxP3+ cells were found to be consistently embedded in a T-cell-enriched immune niche. The proportion of CD8+ T-cells was significantly greater in the CD8+FoxP3+ cell niche relative to the background TME (30.7 % vs. 6.4%, p=0.03; 51.9% vs. 12.2%, p<0.001 ; and 32.8% vs 6.7%, p<0.001 in Cohorts 1 , 2, and 3, respectively). In advanced NSCLC (Cohorts 2-3), the niche was also enriched in FoxP3+CD8(-) cells, previously shown to represent CD4+ regulatory T-cells. Although rare overall, neighboring CD8+FoxP3+ cell pairs were consistentlyidentified). The proportion of tumor cells in the CD8+FoxP3+ cell niche was markedly reduced relative to their frequency in the background TME, while no consistent pattern was identified for CD163+ macrophages. Collectively, these findings highlight the localization of CD8+FoxP3+ cells to a specialized and predictable cellular niche characterized by abundant cytotoxic T-cells and a relative paucity of tumor cells.

[0063] The method 500 was used to identify the immediate arrangement of cells surrounding CD8+FoxP3+ cells and their location relative to the tumor-stromal boundary. The lineage distribution of cells within 5-10pm around a CD8+FoxP3+ cell were evaluated in conjunction with the distance from the tumor stromal boundary. The lower limit of 5pm was chosen so that the cell at the center of the arrangement was not itself counted. Each zone was also classified based on its signed distance from the tumor-stromal boundary, using 25pm bins extending both inward to the central tumor and outward into the peritumoral zone. Optimal threshold for identification of CD8+FoxP3+-like neighborhoods were established on a separate training set (n=6) of NSCLC specimens stained with the same mIF panel (inclusive of 4292360 total cells and 17814 CD8+Foxp3+ cells) prior to analysis of the pre-treatment biopsies from the early-stage and advanced cohorts.

[0064] Given an observation, denoted x, and two hypotheses, denoted A and B, that could have produced the observation, the best possible metric to distinguish between A and B is the ratio of the probabilities of x occurring under those hypotheses: d(x) = pj4(x) / pB(x). A different formulation, D(x) =1 / ( 1 + —!—) = pj4(x) / (pj4(x) + pB(x)), may be used in place of c / (x), because it is bounded between 0 and 1 . An observation that can only be produced by hypothesis A will have D=1 , while one that can only be produced by hypothesis B will have D=0. This property is applied to each cell in the patient tissue sample image: pAisproportional to the fraction of CD8+FoxP3+ cells with the same neighborhood signature and pAis proportional to the fraction of other cells with the same neighborhood signature. From their ratio, a neighborhood signature discriminant may be determined for any cell or for any point in the tissue sample image. This discriminant quantifies how similar the arrangement is to one that typically surrounds a central CD8+FoxP3+ cell, regardless of the actual lineage of the central cell or whether there is even a central cell at all. Larger values of this discriminant mean that the neighborhood looks more like a CD8+FoxP3+ neighborhood. D=1 means that the neighborhood only appears around CD8+FoxP3+ cells in the training tissue sample image, D=0 means that it only appears around other types of cells, and D=0.5 means that it is equally likely to appear around a CD8+FoxP3+ cell as around another type.

[0065] The discriminant threshold, t, which can be set to any value between 0 and 1 according to the first example use case, the number of points with discriminant (D)>t in each sample is counted. The points sampled are randomly generated at an average density of 8000 per HPF, approximately 5900 points / mm2, approximately the same as the average cell density in the TME. In this way, the density of points satisfying a particular condition, such as D>t, can be directly compared with cell densities. Lower thresholds include more points in the analysis, which lends greater statistical power. Using t=0.45 gives both an excellent prediction of response to treatment and a sufficient density of neighborhoods to be reliably detectable. This threshold was used for analyzing the pre-treatment specimens from neoadjuvant and advanced disease cohorts.

[0066] For example, there were 169 cells in the training tissue sample image, including three CD8+FoxP3+ cells, outside the tumor but within 25 pm of its boundary, with exactly two neighbors: one CD8+FoxP3+ and one CD8+FoxP3(-). Theneighborhood signature discriminant for this neighborhood signature may thus be calculated as D=(3 / 17814) / ((3 / 17814)+(166 / 4274546))=0.81 . This same configuration appeared 29 times around the randomly chosen points in the analysis tissue sample images, and so D=0.81 was assigned to those points. Because this is above the threshold t=0.45, those 29 points are identified as CD8+FoxP3+ niches.

[0067] Thus, not only are CD8+FoxP3+ cells a biomarker of a productive antitumor immune response, but their niche also constitutes a biomarker signal associated with response to neoadjuvant immunotherapy. This niche was tested in the first example use case for predicting MPR. The density of the niche was associated with pathologic response (AUC=0.80 for MPR) and RFS in patients with early-stage NSCLC receiving neoadjuvant therapy. Validating this was the niche density associated with OS in patients with advanced pre-treated NSCLC receiving anti-PD-1 (Cohort 2).

[0068] While the calculated AUC values were similar for both biomarkers (0.79 vs. 0.80 for cells vs. niche), the associated statistical error was reduced by 400% for the niche relative to the CD8+FoxP3+ cells (95% Cl (0.76-0.81 ) vs. (0.66, 0.86), respectively, Figure 3G-H). This is particularly advantageous in lung cancer and other tumor types where pre-treatment biopsies tend to be very small and the failure to identify a rare cell that is present could lead to a false negative result.

[0069] The statistical error on the survival predictions was similarly reduced. The 2-year overall survival rate for patients who had CD8+FoxP3+ cells in their biopsy and those who did not were 89% and 57%, respectively, with 95% confidence intervals of [80%, 90%] and [25%, 78%]. The 2-year OS for patients with high and low densities of niches were 83% and 69%, respectively, with much smaller 95% confidence intervals of [82%, 83%] and [69%, 73%].

[0070] Further details regarding the methodology employed in the first example use case follows.

[0071] Cohort 1 was derived from the first-in-human clinical trial of neoadjuvant anti-PD-1 -based therapy in patients with resectable non-small cell lung cancer. Newly diagnosed patients with resectable, stage IA - IV NSCLC were treated with anti-PD-1 based therapy prior to definitive surgical resection. Patients received two to three doses (3mg / kg) of either nivolumab or pembrolizumab alone (n=19) or in combination with one dose (1 mg / kg) of ipilimumab (n=6). Therapy was administered over 8 weeks followed by surgical resection performed within 10 days of the last dose. Recurrence- free survival was calculated using the date of enrollment to the date of clinical recurrence or last follow-up in patients who did not recur.

[0072] Archival formalin-fixed paraffin-embedded (FFPE) pre-treatment tumor tissues (n=42) from n=29 unique patients were obtained from surgical pathology archives. The final cohort consisted of n=25 specimens that met inclusion criteria from n=25 unique patients. Percent residual viable tumor (% RVT) quantified using the immune related pathologic response criteria (irPRC) was reported in 10% increments. Patients whose on-treatment tumors contained <10% RVT were classified as major pathologic responders (MPR).

[0073] Cohort 2 consisted of patients with advanced, pre-treated non- squamous NSCLC who received anti-PD-1 therapy. Of the n=21 pre-treatment tumor tissues available, n=14 met inclusion criteria for cohort 2.

[0074] For the discovery cohort, a six-plex mIF panel (PD-1 , PD-L1 , CD8, FoxP3, CD163, pan-cytokeratin) was quantitatively validated against chromogenic IHC in NSCLC specimens. Briefly, four-micron thick FFPE slides underwent sequential staining cycles. Signals were visualized with Opal fluorophores (AkoyaBiosciences). To normalize staining intensity across batches, each staining batch included a tissue microarray (TMA) control slide. A similar mIF panel, that also contained PD-L1 , CD8, FoxP3, and pan-cytokeratin was applied to specimens that formed the validation cohort.

[0075] A Vectra Imaging System (Akoya Biosciences, MA) was used to acquire whole slide survey scans at 10x (1 .Opm / pixel) magnification (.QPTIFF file). Twenty percent overlapping high powered field (HPF) image tiles at 20x (0.5pm / pixel) magnification were acquired for the entire tissue area using multispectral scanning (,im3 files). Annotations were performed on the survey scans using HALO (Indica Labs, NM) by a board-certified pathologist to delineate the tumor boundary and exclude necrosis and staining artifacts. Cell segmentation maps and single-cell image analysis data were generated using inForm v2.4.8 (Akoya Biosciences), and visually verified. Individual HPF tiles and associated single-cell data were assembled into a seamless whole slide image with an absolute coordinate system at sub-micron resolution. Densities of all cell lineages were calculated for cells within the tumor annotation plus a 250pm peritumoral zone. One specimen was a lymph node metastasis and only the intratumoral area was analyzed unless otherwise stated.

[0076] In summary, the first example use case demonstrates that the presence of CD8+FoxP3+ cells and their niches in the TME both associate positively with MPR and RFS following neoadjuvant immunotherapy in resectable NSCLC and with OS in patients with advanced disease.

[0077] The following presents a non-limiting second example use case of an embodiment. The second example use case concerns determining a patient response to chemotherapy treatment of non-small cell lung carcinoma (NSCLC). The second example use case was similar to the first example use case, but was based on adifferent cohort. In particular, the cohort of the second example use case included patients from the control arm of CheckMate 057 (with tissue available for study) who received only second-line chemotherapy (docetaxel) after progression on first-line platinum doublet chemotherapy. For the patients in this cohort who received second- line chemotherapy, increased niche density was associated with improved OS (median 22.8 vs. 9.5, p=0.042) and PFS (AUC 0.67).

[0078] Note that embodiments are not limited to neighborhood signatures that each include a tally of cells in a region. For example, according to some embodiments, a neighborhood signature may include a tally of pixels according to one or more markers. According to such embodiments, a marker for a given pixel may include a phenotype of a cell, where the given pixel forms part of the image of the cell. According to some pixel-based embodiments, segmentation of cells may not be included. According to some pixel-based embodiments, there may be a large number of neighborhood signatures. The large number of neighborhood signature types may be accounted for using any of a variety of techniques, such as, by way of non-limiting example, binning. By way of non-limiting example, a given tally may be binned according to bins such as 0-9, 10-19, 20-29, etc., pixels. According to some pixelbased examples, the method of determining neighborhood signature discriminants as shown and described in reference to Fig. 4 may be implemented as follows: 408 may be implemented by counting pixels instead of cells, and 410 may be implemented based on signal and noise pixels rather than signal and noise cells. Other embodiments are possible within the scope of the invention.

[0079] Certain examples can be performed using a computer program or set of programs. The computer programs can exist in a variety of forms both active and inactive. For example, the computer programs can exist as software program(s)comprised of program instructions in source code, object code, executable code or other formats; firmware program(s), or hardware description language (HDL) files. Any of the above can be embodied on a transitory or non-transitory computer readable medium, which include storage devices and signals, in compressed or uncompressed form. Exemplary computer readable storage devices include conventional computer system RAM (random access memory), ROM (read-only memory), EPROM (erasable, programmable ROM), EEPROM (electrically erasable, programmable ROM), flash memory, and magnetic or optical disks or tapes.

[0080] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented using computer readable program instructions that are executed by an electronic processor.

[0081] These computer readable program instructions may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the electronic processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises anarticle of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0082] In embodiments, the computer readable program instructions may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, statesetting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the C programming language or similar programming languages. The computer readable program instructions may execute entirely on a user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.

[0083] As used herein, the terms “A or B” and “A and / or B” are intended to encompass A, B, or {A and B}. Further, the terms “A, B, or C” and “A, B, and / or C” are intended to encompass single items, pairs of items, or all items, that is, all of: A, B, C, {A and B}, {A and C}, {B and C}, and {A and B and C}. The term “or” as used herein means “and / or.”

[0084] As used herein, language such as “at least one of X, Y, and Z,” “at least one of X, Y, or Z,” “at least one or more of X, Y, and Z,” “at least one or more of X, Y, or Z,” “at least one or more of X, Y, and / or Z,” or “at least one of X, Y, and / or Z,” is intended to be inclusive of both a single item (e.g., just X, or just Y, or just Z) and multiple items (e.g., {X and Y}, {X and Z}, {Y and Z}, or {X, Y, and Z}). The phrase “at least one of” and similar phrases are not intended to convey a requirement that each possible item must be present, although each possible item may be present.

[0085] The techniques presented and claimed herein are referenced and applied to material objects and concrete examples of a practical nature that demonstrably improve the present technical field and, as such, are not abstract, intangible or purely theoretical. Further, if any claims appended to the end of this specification contain one or more elements designated as “means for [perform]ing [a function]...” or “step for [performing [a function]...”, it is intended that such elements are to be interpreted under 35 U.S.C. § 112(f). However, for any claims containing elements designated in any other manner, it is intended that such elements are not to be interpreted under 35 U.S.C. § 112(f).

[0086] While the invention has been described with reference to the exemplary examples thereof, those skilled in the art will be able to make various modifications to the described examples without departing from the true spirit and scope. The terms and descriptions used herein are set forth by way of illustration only and are not meant as limitations. In particular, although the method has been described by examples, the steps of the method can be performed in a different order than illustrated or simultaneously. Those skilled in the art will recognize that these and other variations are possible within the spirit and scope as defined in the following claims and their equivalents.

Claims

What is claimed is:1 . A method of measuring a presence of cells of a target phenotype in a patient tissue sample, the method comprising: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample; determining respective neighborhood signatures for a plurality of locations in the patient tissue sample image, from which a plurality of neighborhood signatures are obtained; assigning respective discriminants to at least some of the plurality of neighborhood signatures; calculating a proportion of the plurality of locations, wherein the proportion is based on a number of locations that correspond to a neighborhood signature discriminant that is above a threshold and a number of locations that correspond to a neighborhood signature discriminant that is below the threshold; and providing the proportion as a proxy for a density of cells of the target phenotype in the patient tissue sample.

2. The method of claim 1 , further comprising predicting, based on the proportion, a patient response to a therapy.

3. The method of claim 2, further comprising treating the patient with the therapy based on the predicting.

4. The method of claim 3, wherein the therapy comprises at least one of an immunotherapy or a chemotherapy.

5. The method of claim 4, wherein the target phenotype is a T-cell phenotype.

6. The method of claim 1 , wherein a respective neighborhood signature is based on at least a tally of cells in a respective region proximate to a location.

7. The method of claim 6, wherein a respective neighborhood signature comprises a respective cell phenotype tally for a respective region proximate to a location and a respective distance to a boundary of interest.

8. The method of claim 7, wherein each respective cell phenotype tally comprises a number of cells of the target phenotype and a number of cells for at least one other phenotype.

9. The method of claim 7, wherein the boundary of interest comprises a tumor boundary.

10. The method of claim 1 , wherein each respective region comprises an area between two concentric circles.11 . The method of claim 1 , further comprising determining each respective discriminant based on a training tissue sample image.

12. The method of claim 1 , further comprising determining a pathology characterization of the patient tissue sample based on the proportion.

13. A system for measuring a presence of cells of a target phenotype in a patient tissue sample, the system comprising: a non-transitory computer readable medium comprising instructions; and at least one electronic processor that executes the instructions to perform operations comprising: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample; determining respective neighborhood signatures for a plurality of locations in the patient tissue sample image, from which a plurality of neighborhood signatures are obtained; assigning respective discriminants to at least some of the plurality of neighborhood signatures; calculating a proportion of the plurality of locations, wherein the proportion is based on a number of locations that correspond to a neighborhood signature discriminant that is above a threshold and a number of locations that correspond to a neighborhood signature discriminant that is below the threshold; and providing the proportion as a proxy for a density of cells of the target phenotype in the patient tissue sample.

14. The system of claim 13, wherein the operations further comprise predicting, based on the proportion, a patient response to a therapy.

15. The system of claim 14, wherein the patient is treated with the therapy based on the predicting.

16. The system of claim 15, wherein the therapy comprises at least one of an immunotherapy or a chemotherapy.

17. The system of claim 16, wherein the target phenotype is a T-cell phenotype.

18. The system of claim 13, wherein a respective neighborhood signature is based on at least a tally of cells in a respective region proximate to a location.

19. The system of claim 18, wherein a respective neighborhood signature comprises a respective cell phenotype tally for a respective region proximate to a location and a respective distance to a boundary of interest.

20. The system of claim 19, wherein each respective cell phenotype tally comprises a number of cells of the target phenotype and a number of cells for at least one other phenotype.21 . The system of claim 19, wherein the boundary of interest comprises a tumor boundary.

22. The system of claim 13, wherein each respective region comprises an area between two concentric circles.

23. The system of claim 13, wherein the operations further comprise determining each respective discriminant based on a training tissue sample image.

24. The system of claim 13, wherein the operations further comprise determining a pathology characterization of the patient tissue sample based on the proportion.

25. A method of predicting a response to immunotherapy, the method comprising: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample stained to identify: tumor cells, cyto-toxic T cells, macrophages, and regulatory T cells; identifying automatically, in the patient tissue sample image, incidences of tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells; determining, based on spatial arrangements of the tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells, a score that correlates with a likelihood of response to and immunotherapy; and providing an indication of the score.

26. The method of claim 25, wherein the immunotherapy comprises an anti-PD1 therapy.

27. The method of 26, further comprising treating the patient with the anti- PD1 therapy based on the score.

28. The method of 25, wherein the patient tissue sample is stained with at least one of a fluorogenic stain or a chromogenic stain.

29. The method of 25, wherein the patient tissue sample is stained based on antibodies for CD8, CD163, Foxp3, cytokeratin, melanocytes, or a cancer cell type.

30. The method of 25, wherein the spatial arrangements are based on a presence of identified cells within annular regions.31 . The method of 25, wherein the spatial arrangements are based on a presence of identified cells within geometric regions.

32. The method of 25, wherein the patient tissue sample is stained based on RNA associated with at least one of: tumor cells, cyto-toxic T cells, macrophages, or regulatory T cells.

33. The method of 25, further comprising segmenting the patient tissue sample image.

34. The method of 25, wherein the identifying automatically uses machine learning.

35. The method of 25, further comprising parameterizing and indexing spatial data regarding the patient tissue sample image in a database.

36. A system for predicting a response to immunotherapy, the system comprising: a non-transitory computer readable medium comprising instructions; and at least one electronic processor that executes the instructions to perform operations comprising: obtaining a patient tissue sample image depicting a two-dimensional slice of a patient tissue sample stained to identify: tumor cells, cyto-toxic T cells, macrophages, and regulatory T cells; identifying automatically, in the patient tissue sample image, incidences of tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells; determining, based on spatial arrangements of the tumor cells, cyto-toxic T cells, macrophages, regulatory T cells, and other cells, a score that correlates with a likelihood of response to and immunotherapy; and providing an indication of the score.

37. The system of claim 36, wherein the immunotherapy comprises an anti-PD1 therapy.

38. The system of 37, wherein the patient is treated with the anti-PD1 therapy based on the score.

39. The system of 36, wherein the patient tissue sample is stained with at least one of a fluorogenic stain or a chromogenic stain.

40. The system of 36, wherein the patient tissue sample is stained based on antibodies for CD8, CD163, Foxp3, cytokeratin, melanocytes, or a cancer cell type.41 . The system of 36, wherein the spatial arrangements are based on a presence of identified cells within annular regions.

42. The system of 36, wherein the spatial arrangements are based on a presence of identified cells within geometric regions.

43. The system of 36, wherein the patient tissue sample is stained based on RNA associated with at least one of: tumor cells, cyto-toxic T cells, macrophages, or regulatory T cells.

44. The system of 36, wherein the operations further comprise segmenting the patient tissue sample image.

45. The system of 36, wherein the identifying automatically uses machine learning.

46. The system of 36, wherein the operations further comprise parameterizing and indexing spatial data regarding the patient tissue sample image in a database.

Images