WO2025199457A1

WO2025199457A1 - Artificial intelligence (ai)-based assessment of immune cells from digitized pathology images of cancer patients

Info

Publication number: WO2025199457A1
Application number: PCT/US2025/020942
Authority: WO
Inventors: Arpit AGGARWAL; Anant Madabhushi; German CORREDOR PRADA; James Lewis
Original assignee: Emory University; Vanderbilt University
Current assignee: Emory University; Vanderbilt University
Priority date: 2024-03-22
Filing date: 2025-03-21
Publication date: 2025-09-25
Anticipated expiration: 2026-09-22

Abstract

In some embodiments, the present disclosure relates to a method. The method includes accessing digitized pathology imaging data from a cancer patient. The digitized pathology imaging data has been generated using immunohistochemical (IHC) stained pathology images and has been segmented to identify a plurality of immune cells. The plurality of immune cells are used to extract a plurality of immune cell features from the digitized pathology imaging data. The plurality of immune cell features are operated upon by a machine learning stage that is trained to generate a medical prognosis relating to the cancer patient.

Description

ARTIFICIAL INTELLIGENCE (AI)-BASED ASSESSMENT OF IMMUNE CELLS FROM DIGITIZED PATHOLOGY IMAGES OF CANCER PATIENTS

REFERENCE TO RELATED APPLICATION

[0001] This Application claims the benefit of U.S. Provisional Application No. 63/568,691 , filed on March 22, 2024, the contents of which are incorporated by reference in their entirety.

FEDERAL FUNDING INFORMATION

[0002] This invention was made with government support under 1 R01 CA220581 - 01 A1 awarded by the National Institutes of Health/National Cancer Institute. The government has certain rights in the invention.

BACKGROUND

[0003] Cancer is an uncontrolled growth of cells in the human body. Cells may become cancerous after mutations accumulate in genes that control cell proliferation. Cancer can start in any part of the human body and over time may spread to other parts of the body. As cancer spreads it may disrupt organ function, ultimately leading to death. In the United States, cancer is the second most common cause of death after heart disease.

BRIEF DESCRIPTION OF THE DRAWINGS

[0004] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate various example operations, apparatus, methods, and other example embodiments of various aspects discussed herein. It will be appreciated that the illustrated element boundaries (e.g., boxes, groups of boxes, or other shapes) in the figures represent one example of the boundaries. One of ordinary skill in the art will appreciate that, in some examples, one element can be designed as multiple elements or that multiple elements can be designed as one element. In some examples, an element shown as an internal component of another element may be implemented as an external component and vice versa. Furthermore, elements may not be drawn to scale. [0005] Fig. 1 illustrates some embodiments of a block diagram of a cancer assessment system configured to utilize immune cell features extracted from digitized pathology imaging data to make a medical prognosis regarding a cancer patient.

[0006] Fig. 2 illustrates some embodiments of a block diagram of a cancer assessment system configured to utilize M2-subtype tumor-associated macrophage (M2-TAM) features extracted from digitized pathology imaging data to make a medical prognosis.

[0007] Fig. 3 illustrates some additional embodiments of a block diagram of a cancer assessment system configured to utilize M2-TAM features extracted from digitized pathology imaging data to generate a medical prognosis.

[0008] Figs. 4A-4B illustrate some exemplary images related to the determination of a M2-TAM feature that is M2-TAM density.

[0009] Figs. 5A-5B illustrate some exemplary images showing M2-TAM densities associated with high-risk and low-risk cancer patients.

[00010] Figs. 6A-6B illustrate some exemplary images related to the determination of TAM features that are M2-TAM spatial characteristics.

[00011] Fig. 7 illustrates some additional embodiments of a block diagram of a cancer assessment system configured to utilize immune cell features extracted from digitized pathology imaging data to make a medical prognosis.

[00012] Fig. 8 illustrates some additional embodiments of a block diagram of a cancer assessment system configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis.

[00013] Fig. 9A illustrates some additional embodiments of a block diagram of a cancer assessment system configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis.

[00014] Figs. 9B-9C illustrate some embodiments of block diagrams of an immunohistochemical (IHC) simulator configured to generate one or more emulated IHC images based on a digitized pathology image.

[00015] Fig. 9D illustrates an exemplary work-flow corresponding to a cancer assessment system using the IHC simulator configured to generate a simulated IHC stained image, an IHC+ segmentation mask, and a nuclei segmentation mask.

[00016] Fig. 10 illustrates a flow diagram showing some embodiments of a method of generating a medical prognosis regarding a cancer patient using immune cell features extracted from digitized pathology imaging data. [00017] Fig. 11 illustrates some additional embodiments of a block diagram of a cancer assessment apparatus configured to utilize M2-TAM features extracted from digitized pathology imaging data to generate a medical prognosis.

[00018] Fig. 12 illustrates a table showing exemplary performance results of a cancer assessment system configured to utilize a M2-TAM density feature extracted from digitized pathology imaging data to generate a medical prognosis.

[00019] Fig. 13 illustrates graphs showing exemplary performance results of a cancer assessment system configured to utilize M2-TAM spatial features extracted from digitized pathology imaging data to generate a medical prognosis.

[00020] Fig. 14 illustrates some embodiments of a block diagram of an apparatus configured to generate a medical prognosis regarding a cancer patient using immune cell features extracted from digitized pathology imaging data.

DETAILED DESCRIPTION

[00021] The description herein is made with reference to the drawings, wherein like reference numerals are generally utilized to refer to like elements throughout, and wherein the various structures are not necessarily drawn to scale. In the following description, for purposes of explanation, numerous specific details are set forth in order to facilitate understanding. It may be evident, however, to one of ordinary skill in the art, that one or more aspects described herein may be practiced with a lesser degree of these specific details. In other instances, known structures and devices are shown in block diagram form to facilitate understanding.

[00022] Squamous cell carcinomas can develop in many locations through the human body, including the skin, lungs, oral cavity, larynx, colon, and cervix. Oral cancer generally refers to cancer occurring in the oral cavity and oropharynx (e.g., between the vermilion border of the lips and the junction of the hard and soft palates or the posterior one third of the tongue). Most oral cancers are squamous cell carcinomas. These cancers start in squamous cells, which are flat, thin cells that form the lining of the mouth and throat. Patients that are in advanced stages of oral cancer will typically exhibit low survival rates despite aggressive treatment.

[00023] Oropharyngeal squamous cell carcinoma (OPSCC) is a common type of oral cancer. Spatial characteristics derived from tumor-infiltrating lymphocytes in HPV- associated oropharyngeal squamous cell carcinoma (OPSCC) may have prognostic significance in the assessment of oral cancer. Macrophages (e.g., CD163+ M2-subtype tumor-associated macrophages (M2-TAMs)) are also a critical part of anti-tumoral immunity. However, because M2-TAMs are difficult to identify on routine pathology images their quantitative and spatial attributes have yet to be studied.

[00024] The present disclosure relates to a method and apparatus configured to generate a medical prognosis using immune cell features (e.g., M2-subtype tumor- associated macrophage (M2-TAM) features) extracted from a digitized pathology image from a cancer patient (e.g., a cancer patient having HPV-associated OPSCC). In some embodiments, the method may include accessing digitized pathology imaging data (e.g., one or more digitized pathology images) from a cancer patient. The digitized pathology imaging data has been generated using immunohistochemical (IHC) stained pathology images and has been segmented to identify a plurality of immune cells (e.g., TAMs). A plurality of immune cell features are extracted from the digitized pathology imaging data. The plurality of immune cell features are provided to a machine learning model that is trained to generate a medical prognosis relating to the cancer patient. Generating the digitized pathology imaging data using IHC stained pathology images enables the plurality of immune cells to be accurately identified and thus used to generate immune cell features that can be operated upon with a machine learning model to form a medical prognosis with a high degree of accuracy.

[00025] Fig. 1 illustrates some embodiments of a block diagram of a cancer assessment system 100 configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis regarding a cancer patient.

[00026] The cancer assessment system 100 comprises a memory 101 configured to store digitized pathology imaging data 102 from a cancer patient. In various embodiments, the cancer patient may have, have had, and/or be suspected of having oral cavity cancer, oropharynx cancer, colon cancer, laryngeal cancer, cervical cancer, ovarian cancer, and/or the like. In some embodiments, the digitized pathology imaging data 102 may comprise a digitized immunohistochemical (IHC) stained image 104 that was generated from an IHC stained pathology slide (e.g., a slide stained with CD163+ stain). In some embodiments, the digitized pathology imaging data 102 may comprise a digitized histologically stained image 106 that was generated from a histologically stained slide (e.g., a Hematoxylin and Eosin (H&E) stained slide).

[00027] A machine learning identification tool 108 is configured to access the digitized pathology imaging data 102. The machine learning identification tool 108 is further configured to segment the digitized pathology imaging data 102 to identify a plurality of immune cells 110 (e.g., M2-subtype tumor-associated macrophage (M2-TAM) features) within a region of interest (e.g., a tumor microarray punch) of the digitized pathology imaging data 102. In some embodiments, the machine learning identification tool 108 may be configured to further segment the digitized pathology imaging data 102 to identify a plurality of nuclei 112 (e.g., non-immune cell nuclei) within the region of interest. In some embodiments, after identification of the plurality of immune cells 110 and the plurality of nuclei 1 12, the machine learning identification tool 108 may be configured to store segmented digitized pathology images in the memory 101 as part of the digitized pathology imaging data 102.

[00028] In some embodiments, the machine learning identification tool 108 may be configured to identify the plurality of immune cells 1 10 directly from the digitized IHC stained image 104. In other embodiments, the machine learning identification tool 108 may be configured to identify the plurality of immune cells 110 from the digitized histologically stained image 106. In such embodiments, a plurality immune cells are identified within digitized IHC stained images and then used to train the machine learning identification tool 108 to identify the plurality immune cells 110 within the digitized histologically stained image 106. By training the machine learning identification tool 108 to identify the plurality of immune cells 110 from the digitized histologically stained image 106, the machine learning identification tool 108 is able to identify both the plurality of immune cells 110 and the plurality of nuclei 112 from the digitized histologically stained image 106. In yet other embodiments, the machine learning identification tool 108 may be configured to identify the plurality of immune cells 1 10 from one or more emulated immunohistochemical (IHC) image 113. In such embodiments, the machine learning identification tool 108 may be trained to generate the one or more emulated IHC images, which mimic an immunohistochemical stained pathology slide, from the digitized histologically stained image 106.

[00029] A feature extraction tool 1 14 is configured to extract a plurality of immune cell features 116 from the digitized pathology imaging data 102 using the plurality of immune cells 110 and/or the plurality of nuclei 112. In some embodiments, the plurality of immune cell features 116 may include one or more of a macrophage density (e.g., a M2-subtype tumor-associated macrophage (M2-TAM) index including a ratio of M2- TAMs to nuclei within a region of interest comprising a tissue microarray (TMA) punch), macrophage spatial features, features relating to types of collagen, tumor budding features, and/or the like.

[00030] The plurality of immune cell features 1 16 are provided to a machine learning stage 122 comprising one or more machine learning models configured to generate a medical prognosis 124 relating to the cancer patient. In some embodiments, the machine learning stage 122 has been trained to generate the medical prognosis 124 using test data and/or validation data. In some embodiments, the medical prognosis 124 may comprise a classification of the cancer patient as being a low-risk 126 or a high-risk 128 (e.g., of oral cavity cancer, oropharynx cancer, colon cancer, laryngeal cancer, cervical cancer, ovarian cancer, bladder cancer, prostate cancer, different types of breast cancer, etc.). The low-risk 126 classification may be associated with significantly better overall survival than the high-risk 128 classification. The medical prognosis 124 generated by the machine learning stage 122 can be utilized by health care professionals to make a more informed decision relating to a treatment of the cancer patient, thereby allowing for the cancer patient to have an improved quality of life and/or a lower risk of death. For example, the medical prognosis may allow for healthcare professionals to make decisions that would more accurately decide if the cancer patient would benefit from post operative therapy, efc.

[00031] It has been appreciated that cellular components of the immune system influence the progression of cancer. Therefore, the use of the plurality of immune cell features 116 by the machine learning stage 122 allows for the machine learning stage 122 to accurately generate the medical prognosis 124 relating to the cancer patient. Furthermore, while immunohistochemical staining can be expensive and may not be widely accessible in some geographic areas (e.g. middle to low income countries), the disclosed cancer assessment system 100 is able to use digitized histologically stained slides to extract the plurality of immune cells features 116 e.g., M2-TAM features) and generate the medical prognosis 124 thereby providing for a widely accessible and cost- effective method of accurately identifying cancer patients that are at a high-risk of death. [00032] Fig. 2 illustrates some embodiments of a block diagram of a cancer assessment system 200 configured to utilize M2-subtype tumor-associated macrophages (M2-TAM) features extracted from digitized pathology imaging data to generate a medical prognosis.

[00033] The cancer assessment system 200 comprises a memory 101 configured to store digitized pathology imaging data 102 including one or more digitized pathology images from one or more cancer patients. In some embodiments, the one or more cancer patients may comprise one or more oral cancer patients (e.g., having head and neck carcinoma, oral cavity squamous cell carcinoma (OCSCC), human papillomavirus (HPV) induced oropharyngeal squamous cell carcinoma (OPSCC), and/or the like). In some embodiments, the digitized pathology imaging data 102 comprises a digitized histologically stained image 106 (e.g., a digitized H&E stained digitized biopsy slide) obtained from a pathological tissue sample taken from a cancer patient. In other embodiments, the digitized pathology imaging data 102 may comprise a digitized IHC stained image 104 generated from an immunohistochemical stained pathology slide. For example, the digitized IHC stained image 104 may be generated from a slide that has been immunohistochemically stained using monoclonal antibodies directed against CD163+ (e.g., CD163+ monoclonal antibodies, novocastra antibodies, clone 10D6 antibodies).

[00034] A machine learning identification tool 108 is configured to access the digitized pathology imaging data 102. The machine learning identification tool 108 is configured to identify a plurality of immune cells 110 from within the digitized pathology imaging data 102. In some embodiments, the plurality of immune cells 110 may comprise and/or be a plurality of M2-subtype tumor-associated macrophages (M2-TAMs) 202. In some embodiments, the machine learning identification tool 108 may be configured to further identify a plurality of nuclei 112 (e.g., non M2-TAM nuclei) from the digitized pathology imaging data 102.

[00035] In some embodiments, the machine learning identification tool 108 may be configured to identify the plurality of M2-TAMs 202 from the digitized IHC stained image 104 and the plurality of nuclei 112 from the digitized histologically stained image 106. In other embodiments, the machine learning identification tool 108 may be trained using a plurality of M2-TAMs identified from digitized IHC stained images to also identify the plurality of M2-TAMs in digitized histologically stained images. By training the machine learning identification tool 108 to identify the plurality of M2-TAMs using the digitized histologically stained image, both the plurality of M2-TAMs 202 and the plurality of nuclei 112 may be identified from the digitized histologically stained image 106. In yet other embodiments, the machine learning identification tool 108 may be trained to generate one or more emulated IHC images 113 from the digitized histologically stained image 106 and to identify the plurality of M2-TAMs 202 using the one or more emulated IHC images. [00036] A feature extraction tool 1 14 is configured to extract a plurality of immune cell features 116 using the plurality of immune cells 110 and/or the plurality of nuclei 112. In some embodiments, the plurality of immune cell features 116 may comprise and/or be a plurality of M2-TAM features 204 extracted using the plurality of M2-TAMs 202. In some additional embodiments, the feature extraction tool 114 may be configured to further use the plurality of nuclei 112 to extract the plurality of M2-TAM features 204. In some embodiments, the plurality of M2-TAM features 204 may include one or more of a M2-TAM density 206 (e.g., M2-TAM index) and M2-TAM spatial features 208.

[00037] The plurality of M2-TAM features 204 are provided to a machine learning stage 122 that comprises one or more machine learning models that have been trained to generate a medical prognosis 124 relating to the cancer patient. In some embodiments, the medical prognosis 124 may comprise a classification of the cancer patient as being a low-risk 126 or a high-risk 128. The low-risk 126 classification is associated with significantly better overall survival than the high-risk 128 classification. It has been appreciated that cellular components of the tumor microenvironment play a role in the progression of cancer (e.g., oral cancer). Therefore, the use of the plurality of M2-TAM features 204 by the machine learning stage 122 allows for the machine learning stage 122 to accurately generate the medical prognosis 124.

[00038] Fig. 3 illustrates some additional embodiments of a block diagram of a cancer assessment system 300 configured to utilize M2-TAM features extracted from digitized pathology imaging data to generate a medical prognosis regarding a cancer patient. [00039] The cancer assessment system 300 comprises a memory 101 configured to store digitized pathology imaging data 102 comprising one or more digitized pathology images from a cancer patient 302. In some embodiments, the cancer patient 302 may have and/or be suspected of having HPV-associated OPSCC. In some embodiments, the one or more digitized pathology images may comprise one or more tissue microarray (TMA) slides, one or more WSIs, patches of a WSI, segmented images, and/or the like. In some embodiments, the digitized pathology imaging data 102 may comprise a digitized IHC stained image 104 that was generated from an immunohistochemical stained pathology slide, a digitized histologically stained image 106 that was generated from a histologically stained slide (e.g., Hematoxylin and eosin (H&E) stained slides), and/or one or more emulated IHC images that were generated from a digitized histologically stained image to mimic an immunohistochemical stained image. In some embodiments, the memory 101 may comprise electronic memory (e.g., solid state memory, SRAM (static random-access memory), DRAM (dynamic random-access memory), and/or the like).

[00040] In some embodiments, the digitized pathology imaging data 102 may be generated by an image generation stage 303 that is configured to digitize a stained slide generated from a tissue sample taken from the cancer patient 302. In some embodiments, the tissue sample may be from an oral cavity of the cancer patient 302 (e.g., from a soft palate, tonsil, tongue, etc.). In some embodiments, the image generation stage 303 may comprise a tissue resection tool 304 (e.g., a scalpel, a needle, scissors, a punch biopsy, and/or the like) that is used to surgically excise tissue from the cancer patient 302. In some embodiments, the tissue is provided to a tissue sectioning tool 306, which is configured to slice the tissue into thin slices that are placed on one or more transparent pathology slides (e.g., one or more glass slides). In some embodiments, the tissue on the one or more transparent slides is provided to an immunostainer 308 configured to perform immunohistochemical (IHC) staining on the one or more pathology slides. In some embodiments, the immunostainer 308 is configured to immunohistochemically stain the one or more pathology slides using a CD163+ immunohistochemical stain. The one or more tissue slides are subsequently converted to the digitized IHC stained image 104 by a slide digitization tool 312 (e.g., comprising an CMOS image sensor, a CCD camera, and/or the like). In other embodiments, the tissue on the one or more transparent slides is provided to a histological Stainer 310 configured to generate one or more stained tissue slides (e.g., H&E stained slides). The one or more stained tissue slides are subsequently converted to the digitized histologically stained image 106 by the slide digitization tool 312.

[00041] In some embodiments, a patch generator 314 is configured to access the digitized pathology imaging data 102. The patch generator 314 may comprise an additional machine learning model (e.g., deep learning model) that is configured to identify tumor regions on the digitized pathology imaging data 102 (e.g., whole slide image (WSI)) and to separate the tumor regions into a plurality of non-overlapping patches 316 (e.g., tiles). The plurality of non-overlapping patches 316 may subsequently be provided to one or more downstream machine learning models to mitigate computation intensity. In some embodiments, the plurality of non-overlapping patches 316 may have a size of approximately 50 pixels x 50 pixels, 64 pixels x 64 pixels, or other similar values. In some embodiments, non-overlapping patches 316 that are extracted from the digitized pathology imaging data 102 and that contain less than approximately 50% of viable tissue may be discarded. In some embodiments, the non-overlapping patches 316 may be saved in the memory 101 as part of the digitized pathology imaging data 102.

[00042] A machine learning identification tool 108 is configured to access the digitized pathology imaging data 102 (e.g., the plurality of non-overlapping patches 316). The machine learning identification tool 108 is further configured to identify M2-subtype tumor-associated macrophages (M2-TAMs) 202 (e.g., CD163-positive TAMs) and/or nuclei 112 within the digitized pathology imaging data 102. In some embodiments, the M2-TAMs 202 may comprise stromal M2-TAMs and/or epithelial M2-TAMs. In some embodiments, the nuclei 112 may comprise stromal non M2-TAMs and/or epithelial non M2-TAMS.

[00043] In some embodiments, the machine learning identification tool 108 is configured to generate one or more binary masks that comprise the M2-TAMs 202 and one or more binary masks that comprise the nuclei 112. In some such embodiments, the one or more binary masks comprise images having a value of “1” in image units (e.g., pixels, voxels, etc.) identified as being within the M2-TAMs 202 and/or the nuclei 112 and having a value of “0” in image units outside of the M2-TAMs 202 and/or the nuclei 112. In some embodiments, the machine learning identification tool 108 may comprise a machine learning model implemented as computer code run on one or more processors (e.g., a central processing unit (CPU) including one or more transistor devices configured to operate computer code to achieve a result, a microcontroller, a graphics processing unit (GPU), and/or the like). In some embodiments, the machine learning model may comprise or be a HoverNet model, which is the state-of-the-art model for nuclei segmentation in H&E-stained slides.

[00044] A feature extraction tool 1 14 is configured to extract a plurality of M2-TAM features 204 using the M2-TAMs 202 and/or nuclei 112. In some embodiments, the feature extraction tool 114 may be implemented as computer code run on one or more processors (e.g., a CPU, a microcontroller, a GPU, and/or the like). In some embodiments, the plurality of M2-TAM features 204 may comprise a M2-TAM density 206. In some such embodiments, the M2-TAM density 206 may be determined by dividing a number of the M2-TAMs 202 identified by the machine learning identification tool 108 by a total number of the nuclei 112 identified by the machine learning identification tool 108 (e.g., TAM density = — ^numbei °f ^TAMs — ) total number of nuclei [00045] In some embodiments, the plurality of M2-TAM features 204 may include M2- TAM spatial features 208 relating to spatial positions and/or arrangements of the M2- TAMs 202. For example, the plurality of M2-TAM spatial features 208 may include spatial interactions of M2-TAMs, spatial architectures of epithelial M2-TAMs, spatial architectures of epithelial non M2-TAMs, spatial architectures of stromal M2-TAMs, spatial architectures of M2-TAMs, spatial architectures of TILs, and/or the like. In some embodiments, the M2-TAM spatial features 208 may be extracted by generating cell cluster graphs of the M2-TAMs 202 (e.g., of CD163-postiive TAMs). In some such embodiments, the M2-TAM spatial features 208 comprise features that characterize local spatial architecture by constructing sub-graphs on the nuclear nodes. In some embodiments, the M2-TAM spatial features 208 may comprise statistical measures (e.g., a mean, median, skewness, standard deviation, efc.) of spatial measures.

[00046] The plurality of M2-TAM features 204 are provided to a machine learning stage 122 that has been trained to generate a medical prognosis 124 regarding the cancer patient 302. In various embodiments, the medical prognosis 124 may include a risk of death, an overall survival, a risk of disease, and/or the like. In some embodiments, the machine learning stage 122 may be configured to use the plurality of M2-TAM features 204 to perform survival analysis to generate a risk score 318. The machine learning stage 122 is further configured to compare the risk score 318 to a threshold 320 to identify the cancer patient 302 as a low-risk 126 (e.g., a cancer patient having a low-risk of disease and/or death) or a high-risk 128 (e.g., a cancer patient having a high-risk of disease and/or death). In some embodiments, the machine learning stage 122 may comprise and/or be a HoverNet model. In some embodiments, the machine learning stage 122 may be implemented as computer code run on one or more processors (e.g., a CPU, a microcontroller, a GPU, and/or the like).

[00047] In some embodiments, the machine learning stage 122 may be configured to generate the medical prognosis 124 to be associated with disease free survival (DFS) using the M2-TAM density 206 and/or the M2-TAM spatial features 208. For example, in some embodiments the machine learning stage 122 may be configured to generate a medical prognosis 124 that is indicative of poor disease free survival for cancer patients having a high M2-TAM density 206 and to generate a medical prognosis 124 that is indicative of good disease free survival for cancer patients having a low M2-TAM density 206. [00048] Figs. 4A-4B illustrate some exemplary series of images, 400 and 410, related to the determination of a M2-TAM feature that is a M2-TAM density.

[00049] Fig. 4A illustrates a series of images 400 associated with the segmentation and identification of nuclei and M2-subtype tumor-associated macrophages. Image 402 illustrates a patch of tissue (e.g., tissue from an oral cavity of a cancer patient). Image 404 shows a binary mask illustrating segmented nuclei (e.g., non-TAM nuclei) generated from image 402. The segmented nuclei are shown in white and non-nuclei are shown in black. Image 406 shows a binary mask illustrating segmented M2-TAMs generated from image 402. The segmented M2-TAMs are shown in white and nonnuclei are shown in black. Image 408 shows the image 402 annotated to show nuclei in blue and segmented M2-TAM in yellow. A M2-TAM density may be determined by dividing a first number of the segmented M2-TAM by a second number of the segmented nuclei.

[00050] Fig. 4B illustrates a series of additional images 410 associated with the segmentation and identification of nuclei and M2-subtype tumor-associated macrophages. Image 412 illustrates a patch of tissue (e.g., tissue from an oral cavity of a cancer patient). Image 414 shows a binary mask illustrating segmented nuclei (e.g., non M2-TAM nuclei) generated from image 412. The segmented nuclei are shown in white and non-nuclei are shown in black. Image 416 shows a binary mask illustrating segmented M2-TAMs generated from image 412. The segmented M2-TAMs are shown in white and non-nuclei are shown in black. Image 418 shows the image 412 annotated to show nuclei in blue and segmented M2-TAM in yellow. A M2-TAM density may be determined by dividing a first number of the segmented M2-TAM by a second number of the segmented nuclei.

[00051] Figs. 5A-5B illustrate some exemplary images, 500 and 506, showing M2- TAM densities associated with high-risk and low-risk cancer patients.

[00052] Fig. 5A illustrates a slide 502 and a corresponding segmented image 504 corresponding to a cancer patient that is categorized by the disclosed cancer assessment system as being at a high-risk. The segmented image 504 shows M2-TAM in white. Fig. 5B illustrates a slide 508 and a corresponding segmented image 510 corresponding to a cancer patient that is categorized by the disclosed cancer assessment system as being at a low-risk. The segmented image 510 shows M2-TAM in white. Comparison of the segmented image 504 of Fig. 5A and the segmented image 510 of Fig. 5B, shows that the M2-TAM density for high-risk cancer patients is significantly higher than that of low-risk cancer patients, therefore showing that a medical prognosis that is indicative of poor disease free survival for cancer patients may have a high M2-TAM density and a medical prognosis that is indicative of good disease free survival for cancer patients may have a low M2-TAM density.

[00053] Figs. 6A-6B illustrate some exemplary images related to the determination of M2-TAM features that are M2-TAM spatial characteristics.

[00054] Fig. 6A illustrates images 600 showing formation of cell graphs. The images 600 include a segmented image 602 showing M2-TAMs 604 outlined in blue. In some embodiments, the M2-TAMs 604 may comprise stromal M2-TAMs and epithelium M2- TAMs. The images 600 also include a segmented image 606 showing a cell cluster 608 comprising interconnected M2-TAMs.

[00055] Fig. 6B illustrates an image 612 of a part 610 of the cell cluster 608 of Fig. 6A increased in size. As shown in image 612, graphs are constructed using M2-TAM centroids 614 as nodes and edges 616 are disposed between the M2-TAM centroids 614. Using the M2-TAM centroids 614 and edges 616, a spatial arrangement between the M2-TAM centroids 614 may be characterized by a plurality of features. In some embodiments, a decaying function of Euclidean distance may be employed to limit the edges 616 connecting M2-TAM centroids 614 located proximally to each other. In some such embodiments, edges 616 are added to a graph based on a function’s threshold.

[00056] A plurality of M2-TAM spatial features are extracted from the graphs. The plurality of M2-TAM spatial features comprise mathematical descriptors of the graphs. For example, in some embodiments, using the plurality of M2-TAM spatial features may comprise and/or be one or more of spatial interactions of M2-TAMs, spatial architectures of epithelial M2-TAMs, spatial architectures of epithelial non M2-TAMs, spatial architectures of stromal M2-TAMs, spatial architectures of stromal non M2-TAMs, spatial architectures of M2-TAMs, TILs and non M2-TAMs/TILs, M2-TAM and collagen architectures, and/or the like. In some embodiments, the plurality of M2-TAM spatial features may comprise statistical measures of the M2-TAM spatial features.

[00057] In some embodiments, the plurality of M2-TAM spatial features may comprise and/or be one or more of a mean density of clusters of stromal non M2-TAM nuclei, a minimum area of clusters of stromal M2-TAM nuclei, a mean density of clusters of stromal M2-TAM nuclei, a minimum density of clusters of epithelium non M2-TAM nuclei, a skewness of intersected areas of clusters of epithelium M2-TAM nuclei and epithelium non M2-TAM nuclei, a standard deviation of ratios of intersected areas of clusters of epithelium M2-TAM nuclei and epithelium non M2-TAM nuclei to an area of clusters of epithelium non M2-TAM nuclei, a mean of ratios of intersected areas of clusters of stromal M2-TAM nuclei and epithelium non M2-TAM nuclei to an area of clusters of stromal M2-TAM nuclei, a skewness of ratios of intersected areas of clusters of stromal M2-TAM nuclei and epithelium non M2-TAM nuclei to an area of epithelium non M2-TAM nuclei, a standard deviation of ratios of intersected areas of clusters of stromal M2-TAM nuclei and epithelium non M2-TAM nuclei to an average area of areas of clusters of stromal M2-TAM nuclei and epithelium non M2-TAM nuclei, a mean of intersected areas of clusters of stromal non M2-TAM nuclei and epithelium non M2-TAM nuclei, a median of a ratio of intersected areas of clusters of stromal non M2-TAM nuclei and epithelium non M2-TAM nuclei to an area of epithelium non M2-TAM nuclei, a standard deviation of ratios of intersected areas of clusters of stromal non M2-TAM nuclei and epithelium M2- TAM nuclei to an area of stromal non M2-TAM nuclei, a median of ratios of intersected areas of clusters of stromal non M2-TAM nuclei and stromal M2-TAM nuclei to an area of stromal M2-TAM nuclei, a mean percentage of clusters of epithelium M2-TAM surrounding stromal non M2-TAM nuclei neighborhoods, a median percentage of clusters of epithelium M2-TAM surrounding stromal non M2-TAM nuclei neighborhoods, a skewness of percentage of clusters of epithelium M2-TAM surrounding stromal non M2- TAM nuclei neighborhoods, a minimum percentage of clusters of stromal M2-TAM nuclei surrounding stromal M2-TAM nuclei neighborhoods, a mean percentage of clusters of epithelium M2-TAM nuclei surrounding epithelium M2-TAM nuclei neighborhoods, a standard deviation of a percentage of clusters of epithelium non M2-TAM nuclei surrounding epithelium M2-TAM nuclei neighborhoods, a maximum percentage of clusters of epithelium M2-TAM nuclei surrounding epithelium M2-TAM nuclei neighborhoods, a mean percentage of clusters of stromal non M2-TAM nuclei surrounding epithelium non M2-TAM nuclei neighborhoods, a minimum percentage of clusters of epithelium non M2-TAM nuclei surrounding epithelium non M2-TAM nuclei neighborhoods, a mean of a grouping factor of epithelium M2-TAM nuclei.

[00058] Fig. 7 illustrates a block diagram showing some additional embodiments of a cancer assessment system 700 configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis regarding a cancer patient.

[00059] The cancer assessment system 700 comprises a memory 101 configured to store digitized pathology imaging data 102 from a cancer patient. In some embodiments, the digitized pathology imaging data 102 may comprise a digitized IHC stained image 104 generated from an immunohistochemical stained pathology slide and a digitized histologically stained image 106 generated from a histologically stained slide (e.g., a Hematoxylin and eosin (H&E) stained slide).

[00060] In some embodiments, the digitized IHC stained image 104 and the digitized histologically stained image 106 may be co-registered using a co-registration tool 702 to generate co-registered imaging data 102c. The co-registered imaging data 102c includes a co-registered digitized IHC stained image 104c and a co-registered digitized histologically stained image 106c. The co-registered digitized IHC stained image 104c is a co-registered image generated from the digitized IHC stained image 104 and the co-registered digitized histologically stained image 106c is a co-registered image generated from the digitized histologically stained image 106.

[00061] A machine learning identification tool 108 is configured to utilize the coregistered digitized IHC stained image 104c to identify a plurality of IHC immune cells 110a (e.g., M2-TAMs). In some embodiments, the machine learning identification tool 108 may be configured to identify M2-TAMs within the co-registered digitized IHC stained image 104c due to a brown staining of the IHC stained cores. The machine learning identification tool 108 comprises a machine learning model 704 that is configured to use the plurality of IHC immune cells 110a and the co-registered digitized histologically stained image 106c to train the machine learning model 704 to identify immune cells 110 within a histologically stained slide. By identifying immune cells 110 within a histologically stained slide, the machine learning identification tool 108 is able to identify the plurality of immune cells 110 from cheaper and more accessible histologically stained slides (e.g., H&E stained slides).

[00062] For example, memory 101 may be configured to further store an additional digitized histologically stained image 706. The additional digitized histologically stained image 706 may be generated from a histologically stained tissue sample taken from an additional cancer patient. The machine learning identification tool 108 is configured to operate upon the additional digitized histologically stained image 706 with the machine learning model 704 to identify the plurality of immune cells 110 (e.g., macrophages) and a plurality of nuclei 112 (e.g., non-immune cell nuclei). Therefore, the machine learning identification tool may identify both the plurality of immune cells 110 and the plurality of nuclei 112 from a digitized histologically stained image. [00063] A feature extraction tool 1 14 is configured to extract a plurality of immune cell features 116 using the plurality of immune cells 110 and/or the plurality of nuclei 112. In some embodiments, the plurality of immune cell features 116 may include one or more of a macrophage density (e.g., a CD163+ M2-subtype tumor-associated macrophage density), macrophage spatial features, neutrophil features, features relating to types of collagen, tumor budding features, and/or the like.

[00064] The plurality of immune cell features 1 16 are provided to a machine learning stage 122 configured to generate a medical prognosis 124 relating to the additional cancer patient. In some embodiments, the medical prognosis 124 may comprise a classification of the additional cancer patient as being a low-risk 126 or a high-risk 128 for a type of cancer.

[00065] Fig. 8 illustrates a block diagram showing some additional embodiments of a cancer assessment system 800 configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis regarding a cancer patient.

[00066] The cancer assessment system 800 comprises a memory 101 configured to store digitized pathology imaging data 102 from a cancer patient. In some embodiments, the digitized pathology imaging data 102 may comprise a digitized histologically stained image 106 generated from a histologically stained slide (e.g., an H&E stained slide).

[00067] A machine learning identification tool 108 is configured to access the digitized pathology imaging data 102. The machine learning identification tool 108 comprises an IHC simulator 802, which is configured to operate an IHC simulation algorithm on the digitized histologically stained image 106 to transform the digitized histologically stained image 106 into one or more emulated IHC images 113. The machine learning identification tool 108 is configured to identify a plurality of immune cells 110 (e.g., M2- TAMs) and/or a plurality of nuclei 112 from the one or more emulated IHC images 1 13. In some embodiments, the IHC simulator 802 is configured to generate the one or more emulated IHC images 113 using one or more general adversarial networks that have been trained to convert histologically stained images to emulated IHC images.

[00068] A feature extraction tool 1 14 is configured to extract a plurality of immune cell features 116 using the plurality of immune cells 110 and/or the plurality of nuclei 112. In some embodiments, the plurality of immune cell features 116 may include one or more of a macrophage density (e.g., a CD163+ M2-subtype tumor-associated macrophage density), macrophage spatial features, features relating to types of collagen, tumor budding features, and/or the like.

[00069] The plurality of immune cell features 1 16 are provided to a machine learning stage 122 configured to generate a medical prognosis 124 relating to the cancer patient. In some embodiments, the medical prognosis 124 may comprise a classification of the cancer patient as being a low-risk 126 or a high-risk 128 for cancer. [00070] Fig. 9A illustrates a block diagram of some additional embodiments of a cancer assessment system 900 configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis.

[00071] The cancer assessment system 900 comprises a machine learning identification tool 108 that is configured to access digitized pathology imaging data 102. The machine learning identification tool 108 comprises an IHC simulator 802 configured to generate mappings between a histologically stained image and an emulated IHC stained image. In some embodiments, the IHC simulator 802 comprises a Pix2Pix Generative Adversarial Network (Pix2Pix) based calibration framework configured to represent mappings between a histologically stained image and an emulated IHC stained image using one or more generators 904 and one or more discriminators 906. [00072] In some embodiments, the IHC simulator 802 may comprise a general adversarial network 902 (e.g., a Pix2Pix) trained on co-registered digitized IHC stained images 104c and co-registered digitized histologically stained images 106c to perform an image translation between a digitized histologically stained image and one or more emulated IHC images. In some embodiments the general adversarial network 902 may comprise a generator 904 configured to perform an image translation on an additional digitized histologically stained image 706 to generate one or more emulated IHC images 113 including a simulated IHC stained image 907. In some additional embodiments, the general adversarial network 902 may comprise a plurality of generators 904 configured to perform an image translation between the additional digitized histologically stained image 706 and one or more emulated IHC images 113 including a simulated IHC stained image 907, an IHC+ segmentation mask 908, and/or a nuclei segmentation mask 910. A feature extraction tool 1 14 may be configured to use the simulated IHC stained image 907, the IHC+ segmentation mask 908, and/or the nuclei segmentation mask 910 to generate different ones of the plurality of TAM features 204. [00073] For example, Fig. 9B illustrates some embodiments of a block diagram 912 of an IHC simulator 802 comprising a generator 904 configured to generate a simulated IHC stained image 907 based on a digitized histologically stained image.

[00074] The IHC simulator 802 of Fig. 9B comprises one or more generators 904 that are configured to receive a co-registered digitized histologically stained image 106c and to perform an image translation of the co-registered digitized histologically stained image 106c to generate the simulated IHC stained image 907. The IHC simulator 802 further comprises a discriminator 906 that attempts to determine if the simulated IHC stained image 907 generated by the generator 904 is distinguishable from a coregistered digitized IHC stained image 104c. The IHC simulator 802 utilizes one or more loss functions 914 to ensure accuracy between the simulated IHC stained image 907 and the co-registered digitized IHC stained image 104c.

[00075] Fig. 9C illustrates a block diagram 916 of some additional embodiments of an IHC simulator 802 comprising a plurality of generators 904a-904c configured to generate a simulated IHC stained image 907, an IHC+ segmentation mask 908, and a nuclei segmentation mask 910.

[00076] The IHC simulator 802 of Fig. 9C comprises a first generator 904a, a second generator 904b, and a third generator 904c respectively configured to access the digitized imaging data within the memory.

[00077] In some embodiments, the first generator 904a is configured to receive a coregistered digitized histologically stained image 106c and to perform image translation of the co-registered digitized histologically stained image 106c to generate the simulated IHC stained image 907. A first discriminator 906a is configured to determine if the simulated IHC stained image 907 generated by the first generator 904a is distinguishable from ground truth images 918. In some embodiments, the first generator 904a may comprise and/or be a ResNet-9 generator. In some embodiments, the ResNet-9 generator comprises a convolution layer, followed by a batch normalization layer, a rectified linear unit (ReLU) activation function, two down sampling layers, nine residual blocks, two up sampling layers, and a final convolutional layer followed by a tanh activation function. The residual blocks may respectively comprise two convolutional layers with a same number of output channels. The convolutional layers within a residual block are respectively followed by a batch normalization layer and a ReLU activation function. [00078] In some embodiments, the first discriminator 906a utilizes one or more first loss functions 914a to determine a difference between the ground truth images 918 and the simulated IHC stained image 907. In some embodiments, the one or more first loss functions 914a include:

The notation E represents an expectation value, Di corresponds to the first discriminator 906a, Gi corresponds to the first generator 904a, 01 corresponds to the simulated IHC stained image 907, and /corresponds to a ground truth image 918. [00079] The second generator 904b is configured to receive the co-registered digitized histologically stained image 106c and to generate the IHC+ segmentation mask 908. A second discriminator 906b is configured to determine if the IHC+ segmentation mask 908 generated by the second generator 904b is distinguishable from the ground truth images 918. The third generator 904c is configured to receive the co-registered digitized histologically stained image 106c and to generate the nuclei segmentation mask 910. A third discriminator 906c is configured to determine if the nuclei segmentation mask 910 generated by the third generator 904c is distinguishable from the ground truth images 918.

[00080] In some embodiments, the second generator 904b and/or the third generator 904c may comprise a U-Net generator. The U-Net generator may include a U-Net with skip connections between each layer / and layer n - /; where n is the total number of layers. Each skip connection involved concatenation of all channels at layer / with layer n - i.

[00081] In some embodiments, the second generator 904b and the third generator 904c are configured to respectively utilize one or more second loss functions 914b e.g. LSGAN loss function) to determine a difference between the IHC+ segmentation mask 908 and the ground truth images 918 and one or more third loss functions 914c (e.g., LSGAN loss function) to determine a difference between the nuclei segmentation mask 910 and the ground truth images 918. In some embodiments, the one or more second loss functions 914b and the one or more third loss functions 914c may comprise:

L_s (fi₂, G_3I D_2I D₃, ) = L_t [(G₂ -3> ^2-3)] + -^ ^x ^Ll (^2-3) '

The notation E represents an expectation value, D_x corresponds to a discriminator, G_x corresponds to a generator, o_x corresponds to the IHC+ segmentation mask 908 or the nuclei segmentation mask 910, and / corresponds to a ground truth image 918.

[00082] In some embodiments, a final loss function may be used to optimize the image translation and segmentation tasks simultaneously. The final loss function may be defined as follows where wi and 1/1/2 represent the predefined weights for the image translation and segmentation tasks, respectively.

[00083] In some embodiments, the first discriminator 906a, the second discriminator 906b, and the third discriminator 906c may comprise PatchGAN discriminators, which are able to address the capture of intricate details present at high frequencies in the image. This discriminator penalizes structural aspects at a patch level and categorizes each N x N patch within an image as real or fake.

[00084] The disclosed virtual staining enabled by the IHC simulator 802 is able to generate high quality emulated IHC images (e.g., stained with a CD163+ stain) from histology slides (e.g., H&E stained slides) according to multiple metrics. For example, disclosed virtual staining is able to achieve a structural similarity index (SSIM) of 0.72, Peak Signal-to-Noise Ratio (PSNR) of 21 .9, and a Frechet Inception Distance (FID) of 58.7. This is significantly better than other state-of-the-art frameworks and/or methods. Additionally, a TAM index derived from the disclosed virtual staining achieved a significant association with survival outcomes (e.g., the disclosed virtual staining achieved a Hazard Ratio = 3.96 [1 .03-8.87] with a p-value = 0.006), whereas TAM indexes from state-of-the-art frameworks and/or methods were not able to achieve a significant association with survival outcomes (e.g., a state-of-the-art framework achieved a Hazard Ratio = 1 .58 [0.71 -3.51 ] with a p-value = 0.26). [00085] Therefore, the disclosed assessment system is able to perform virtual staining that can be used to generate an accurate medical prognosis from medical images, thereby improving a computer’s ability to accurately identify cancer patients as being high-risk or low-risk from medical images. The improved ability to accurately identify cancer patients as being high-risk or low-risk can improve treatment of the cancer patients.

[00086] Fig. 9D illustrates an exemplary work-flow 920 corresponding to a disclosed cancer assessment system using an IHC simulator configured to generate a simulated IHC stained image, an IHC+ segmentation mask, and a nuclei segmentation mask. [00087] As shown in the work-flow 920, a patient population 922 is identified. For respective members 924 of the patient population 922, a tissue microarray 926 comprising a plurality of tissue cores is generated. The plurality of tissue cores may include H&E stained cores 928 and CD 163+ IHC stained cores 930. The H&E stained cores 928 and CD 163+ IHC stained cores 930 may be co-registered.

[00088] Regions of interest 932a-932b within the plurality of tissue cores is identified. The region of interest 932b corresponding to an H&E stained core is provided to an IHC simulator 802 comprising one or more general adversarial networks. For example, the region of interest 932b corresponding to the H&E stained core may be provided to one or more generators 904. The one or more generators 904 are configured to operate upon the region of interest 932b corresponding to the H&E stained core to generate an IHC stained image 907, an IHC+ segmentation mask 908, and a nuclei segmentation mask 910. One or more discriminators 904a-904c are configured to compare the IHC stained image 907, the IHC+ segmentation mask 908, and the nuclei segmentation mask 910 to a region of interest 932a corresponding to an IHC stained core so as to minimize associated loss functions. By minimizing the associated loss functions over a plurality of cores, the one or more general adversarial networks may be trained to mimic the IHC stained image 907, the IHC+ segmentation mask 908, and the nuclei segmentation mask 910.

[00089] Once the one or more general adversarial networks have been trained, the cancer assessment system may be operated upon additional patients. For example, a surgical sample 934 may be taken from an additional cancer patient. The surgical sample 934 may be stained with an H&E stained to generate an H&E whole slide image (WSI) 936. The H&E WSI 936 is subsequently provided to the IHC simulator 802 to generate a virtual CD163+ IHC WSI. The cancer assessment system may extract a plurality of immune cell features from the virtual CD163+ IHC WSI. The plurality of immune cell features can subsequently be used to perform survival analysis 938 to generate a medical prognosis relating to the additional cancer patient.

[00090] Fig. 10 illustrates a flow diagram showing some embodiments of a method 1000 of generating a medical prognosis regarding a cancer patient’s survival using immune cell features extracted from digitized pathology imaging data.

[00091] While the disclosed method 1000 is illustrated and described herein as a series of acts or events, it will be appreciated that the illustrated ordering of such acts or events are not to be interpreted in a limiting sense. For example, some acts may occur in different orders and/or concurrently with other acts or events apart from those illustrated and/or described herein. In addition, not all illustrated acts may be required to implement one or more aspects or embodiments of the description herein. Further, one or more of the acts depicted herein may be carried out in one or more separate acts and/or phases.

[00092] At act 1002, digitized pathology imaging data from a cancer patient is accessed. In some embodiments, the cancer patient may have and/or be suspected to have human papillomavirus (HPV) induced oropharyngeal squamous cell carcinoma. [00093] At act 1004, the digitized pathology imaging data is segmented to identify nuclei (e.g., non M2-TAM nuclei) and/or immune cells (e.g., tumor associated macrophages (M2-TAMs) etc.). In some embodiments, the digitized pathology imaging data may be segmented to identify nuclei and/or immune cells according to one or more of acts 1006-1010.

[00094] At act 1006, an IHC simulation algorithm is trained using a digitized IHC pathology image and a digitized histologically stained image. The IHC simulation algorithm is configured to generate one or more emulated IHC images from a digitized histologically stained image (e.g., a digitized image generated from an H&E stained slide).

[00095] At act 1008, the IHC simulation algorithm is operated upon an additional digitized histologically stained image to generate one or more emulated IHC images. [00096] At act 1010, the immune cells (e.g., M2-TAM nuclei) are identified within the one or more emulated IHC images.

[00097] At act 1012, the segmented digitized pathology images may be stored in electronic memory, in some embodiments. [00098] At act 1014, a plurality of immune cell features (e.g., M2-TAM features) are extracted using the nuclei and/or immune cells. In some embodiments, the plurality of M2-TAM features may comprise one or more of a M2-TAM density (e.g., a M2-TAM index including a ratio of M2-TAM to nuclei within a region of interest) and M2-TAM spatial features. In some embodiments, the plurality of M2-TAM features may be generated according to one or more of acts 1016-1018.

[00099] At act 1016, a M2-TAM density is determined using the nuclei and the M2- TAMs. The M2-TAM density (e.g., M2-TAM index) may comprise a ratio of the M2- TAMs and the nuclei within a region of interest (e.g., a TMA punch).

[000100] At act 1018, M2-TAM spatial features may be determined using cell graphs generated from M2-TAMs.

[000101] At act 1020, a machine learning model is operated onto the plurality of immune cell features to generate a medical prognosis regarding the cancer patient. The medical prognosis may relate to a survival outcome of the cancer patient.

[000102] At act 1022, a treatment may be provided to the cancer patient based upon the medical prognosis, in some embodiments. For example, based upon the medical prognosis it may be determined that postoperative treatment may be beneficial to a cancer patient and post operative treatment may be applied to the cancer patient.

[000103] Therefore, the disclosed method 1000 utilizes immune cell features extracted from immune cell regions and/or nuclei within tissue of a cancer patient to make a medical prognosis regarding a survival of the cancer patient.

[000104] It will be appreciated that the disclosed methods and/or block diagrams may be implemented as computer-executable instructions, in some embodiments. Thus, in one example, a computer-readable storage device (e.g., a non-transitory computer- readable medium) may store computer executable instructions that if executed by a machine (e.g., computer, processor) cause the machine to perform the disclosed methods and/or block diagrams. While executable instructions associated with the disclosed methods and/or block diagrams are described as being stored on a computer- readable storage device, it is to be appreciated that executable instructions associated with other example disclosed methods and/or block diagrams described or claimed herein may also be stored on a computer-readable storage device.

[000105] Fig. 11 illustrates some additional embodiments of a block diagram of a cancer assessment apparatus 1100 configured to utilize immune cell features extracted from digitized pathology imaging data to generate a medical prognosis regarding a cancer patient.

[000106] The cancer assessment apparatus 1100 comprises a memory 101 configured to store digitized pathology imaging data 102 including a plurality of digitized pathology images from cancer patients having and/or suspected of having cancer (e.g., HPV induced OPSCC). In various embodiments, the plurality of digitized pathology images may be obtained by an image generation stage 303 and/or from an on-line database 1102 and/or archive containing digitized pathology images from cancer patients generated at different sites (e.g., different hospitals, research laboratories, and/or the like). Prior to including digitized pathology images within the digitized pathology imaging data, the digitized pathology images may be subjected to a pre-processing stage 1104. The pre-processing stage 1104 may be configured to normalize image characteristics (e.g., color, brightness, contrast) so as to mitigate batch effects (e.g., differences between images obtained from different sites).

[000107] In some embodiments, the digitized pathology imaging data 102 may include a training set 102t and a validation set 102v. The training set 102t comprises digitized pathology images from a first plurality of cancer patients. The validation set 102v comprises digitized pathology images from a second plurality of cancer patients. In some embodiments, both the training set 102t and the validation set 102v may comprise both images generated from immunohistochemical stained pathology slides and images generated from histologically stained slides (e.g., H&E stained slides). In some embodiments (not shown), the memory 101 may also be configured to store ground truth segmentation data (e.g., segmentation results provided by an expert human pathologist).

[000108] In some embodiments, the training set 102t may be used to train a downstream machine learning identification tool 108 to perform segmentations that identify immune cells 110 (e.g., M2-TAMs 202) and/or nuclei 112 (e.g., from digitized images generated from histologically stained slides). The training set 102t may also be used to train a downstream machine learning stage 122 to generate a medical prognosis 124 relating to a survival of a cancer patient 302.

[000109] The validation set 102v may be used to validate the results of the machine learning identification tool 108 to perform segmentations that identify immune cells 110 (e.g., M2-TAMs 202) and/or nuclei 1 12. The validation set 102v may also be used to validate the results of the machine learning stage 122 and the medical prognosis 124. [000110] In some embodiments, machine learning stage 122 may include a feature selection element 1106 configured to select a set of most prognostic immune cell features (e.g., M2-TAM features) to generate the medical prognosis 124. For example, the features extraction tool 114 may extract a first number of M2-TAM features and then the feature selection element 1106 may select a smaller second number of the M2-TAM features (from the first number of M2-TAM features) that are most prognostic (e.g., that have a most significant impact in determining survival). In some embodiments, the second number of M2-TAM features may be used to train and validate the machine learning stage 122. In some embodiments, the machine learning stage 122 may comprise a Cox regression model (e.g., a Cox proportional hazards model). In some embodiments, the Cox regression model may comprise a LASSO (least absolute shrinkage and selection operator) algorithm (e.g., a LASSO Cox regression model) that is configured to operate as the feature selection element 1106. [000111] In some embodiments, the machine learning stage 122 may be configured to generate a risk score 318 related to survival of a patient. In some embodiments, a median risk score (e.g., a median of a plurality of risk scores) obtained by the machine learning stage 122 for the training set 102t may be used for risk stratification in the validation set 102v. For example, in some embodiments, the machine learning stage 122 may operate upon a first plurality of M2-TAM features extracted from digitized images within the training set 102t to determine a plurality of risk scores and to subsequently determine a median risk score. The median risk score may be subsequently set as a threshold 320 that is configured to distinguish between cancer patients classified as low-risk 126 and cancer patients classified as high-risk 128. The machine learning stage 122 is then operated upon a second plurality of M2-TAM features extracted from one or more digitized images within the validation set 102v to determine risk scores 318 associated with one or more cancer patients. The risk scores 318 are compared to the threshold 320 (e.g., the median risk score) to classify the one or more cancer patients as low-risk or high-risk.

[000112] Fig. 12 illustrates a table 1200 showing exemplary performance results of a disclosed cancer assessment system configured to utilize a TAM density feature extracted from digitized pathology imaging data to generate a medical prognosis regarding an oral cancer patient.

[000113] Table 1200 illustrates exemplary hazard ratios (HR) associated with medical prognosis of disease free survival (DFS) generated using M2-TAM density features extracted from digitized pathology images within a first data set 1202 and a second data set 1204 comprising images from cancer patients having HPV-associated oropharyngeal squamous cell carcinoma (OPSCC). Within the first data set 1202, a hazard ratio of 3.28 was achieved with a 95% confidence interval and a p-value of 0.02. In some embodiments, the hazard ratios achieved by the first data set varied between 1.16 and 9.32. Within the second data set 1204, a hazard ratio of 4.89 was achieved with a 95% confidence interval and with a p-value of 0.02. In some embodiments, the hazard ratios achieved by the second data set varied between 1 .5 and 16. The relatively high value of the hazard ratios indicates that the M2-TAM density features have a high prognostic value and that they therefore improve a computer’s ability to accurately identify cancer patients as being high-risk or low-risk from medical images.

[000114] Fig. 13 illustrates graphs, 1300 and 1306, showing exemplary performance results of a disclosed cancer assessment system configured to utilize M2-TAM spatial features extracted from digitized pathology imaging data to generate a medical prognosis regarding an oral cancer patient.

[000115] Graph 1300 shows lines indicative of a survival probability (y-axis) as a function of time (x-axis) for high-risk and low-risk patents generated using M2-TAM spatial features extracted from images within a first data set comprising images from cancer patients having HPV-associated OPSCC. As shown in graph 1300, a first line is indicative of low-risk cancer patients 1302 and a second line is indicative of high-risk cancer patients 1304. The low-risk cancer patients 1302 have a higher survival probability over time than the high-risk cancer patients 1304. In some embodiments, the TAM spatial features within the first data set achieved a hazard ratio of 2.73 with a 95% confidence interval and a p-value of 0.0415. In some embodiments, the hazard ratios achieved by the second data set varied between 1.1 and 8.9.

[000116] Graph 1306 shows lines indicative of a survival probability (y-axis) as a function of time (x-axis) for high-risk and low-risk patents generated using M2-TAM spatial features extracted from images within a second data set comprising images from cancer patients having HPV-associated OPSCC. As shown in graph 1306, a first line is indicative of low-risk cancer patients 1308 and a second line is indicative of high- risk cancer patients 1310. The low-risk cancer patients 1308 have a higher survival probability over time than the high-risk cancer patients 1310. In some embodiments, the M2-TAM spatial features achieved a hazard ratio of 8.54 with a 95% confidence interval and a p-value of 0.0195. In some embodiments, the hazard ratios achieved by the second data set varied between 1 .3 and 56. The relatively high value of the hazard ratios indicates that the M2-TAM spatial features have a high prognostic value and that they therefore improve a computer's ability to accurately identify cancer patients as being high-risk or low-risk from medical images.

[000117] Fig. 14 illustrates some embodiments of a block diagram of an apparatus 1400 configured to generate a medical prognosis regarding a cancer patient using immune cell features extracted from digitized pathology imaging data.

[000118] The apparatus 1400 comprises a cancer assessment apparatus 1402. The cancer assessment apparatus 1402 is coupled to an image generation stage 303, which is configured to generate a digitized pathology image of tissue samples collected from a cancer patient 302 that has and/or that is suspected of having cancer e.g., HPV-associated OPSCC).

[000119] The cancer assessment apparatus 1402 comprises a processor 1406 and a memory 1404. The processor 1406 can, in various embodiments, comprise circuitry such as, but not limited to, one or more single-core or multi-core processors. The processor 1406 can include any combination of general-purpose processors and dedicated processors (e.g., graphics processors, application processors, etc.). The processor(s) 1406 can be coupled with and/or can comprise memory (e.g., memory 1404) or storage and can be configured to execute instructions stored in the memory 1404 or storage to enable various apparatus, applications, or operating systems to perform operations and/or methods discussed herein.

[000120] The memory 1404 can be configured to store digitized pathology imaging data comprising digitized pathology images. The digitized pathology images may comprise digitized surgical specimen images having a plurality of pixels, each pixel having an associated intensity. In some additional embodiments, the digitized pathology images may be stored in the memory 1404 as one or more training sets of digitized images for training a classifier and/or one or more test sets (e.g., validation sets) of digitized images.

[000121] The cancer assessment apparatus 1402 also comprises an input/output (I/O) interface 1408 (e.g., associated with one or more I/O devices), a display 1410, and an interface 1412 that connects the processor 1406, the memory 1404, and the I/O interface 1408. The I/O interface 1412 can be configured to transfer data between the memory 1404, the processor 1406, and external devices, for example, the image generation stage 303. [000122] In some embodiments, the cancer assessment apparatus 1402 may further comprise one or more circuits 1414 that include one or more of a machine learning identification circuit 1416, a feature extraction circuit 1418, and a machine learning circuit 1420. In some embodiments, the one or more circuits 1414 may operate according to machine learning algorithms stored in the memory 1404.

[000123] In some embodiments, the machine learning identification circuit 1416 is configured to segment the plurality of digitized pathology images to generate segmented imaging data 1417 that identify immune cells and non-immune cell nuclei (e.g., M2-TAM and/or non M2-TAM nuclei) within the digitized pathology images. In some embodiments, the machine learning identification circuit 1416 may be configured to run an artificial intelligence IHC simulation algorithm that generates one or more emulated IHC images from a digitized histologically stained image. In such embodiments, the machine learning identification circuit 1416 may be configured to extract the immune cells from the one or more emulated IHC images and the non- immune cell nuclei from the digitized histologically stained image, thereby eliminating the need for IHC stained samples to extract features related to immune cells. The feature extraction circuit 1418 is configured to extract a plurality of immune cell features 116 from the immune cells and non-immune cell nuclei. The machine learning circuit 1420 is configured to utilize the plurality of immune cell features 116 to generate a medical prognosis 124 regarding a survival of the cancer patient 302. In some embodiments, the display 1410 is configured to output or display the medical prognosis 124 generated by the cancer assessment apparatus 1402.

1^st Example use case:

[000124] Background: There is prognostic significance of spatial characteristics derived from tumor-infiltrating lymphocytes in HPV-associated oropharyngeal squamous cell carcinoma (OPSCC). While macrophages (tumor-associated or M2- TAMs) are a critical part of anti-tumoral immunity, they are difficult to identify on H&E slides, and their quantitative and spatial attributes have yet to be studied through image analysis. This study introduces the development and validation of two artificial intelligence (Al) based M2-TAM classifiers using CD163 immunostains and H&E slides for prognosis in HPV-associated OPSCC patients.

[000125] Methods: H&E-stained slides from tissue microarray (TMA) slides of OPSCC patients were gathered from two institutions (Vanderbilt University for modeling (N=102), Emory for validation (N=50)). We digitally scanned the slides and immunostained them for CD163 (Novocastra; monoclonal; clone 10D6) on a Leica Bond III immunostainer. We then digitally scanned these slides and co-registered them. For the first classifier, we quantified M2-TAM density by calculating the ratio of M2- TAMs to the number of nuclei present in the TMA punches. To identify M2-TAMs, we leveraged the IHC-stained TMA cores to recognize their characteristic brown staining. For the second classifier, we derived spatial characteristics through the creation of cell cluster graphs of the M2-TAMs. These spatial attributes were employed as features for the development of a Cox regression model (CRM+LASSO), which was trained on the modeling cohort. The CRM+LASSO model identified 23 features for predicting the risk of disease in both cohorts. For these two prognostic classifiers, the mean risk score determined from the modeling cohort was used to stratify patients in each of the cohorts into low and high-risk categories.

[000126] Results: High M2-TAM density was found to be associated with poorer disease free survival for the modeling cohort (HR= 3.28, 95% Cl= 1 .1 -9.32, p= 0.02) and for the validation cohort (HR= 7.04, 95% Cl=1 .1 -22.5, p= 0.03). Similarly, spatial M2-TAM features were also found to be associated with poorer disease free survival in both cohorts (modeling, HR= 2.73, 95% Cl=1.1 -8.9, p= 0.04 and validation, HR= 8.5, 95% Cl= 1.3-56, p= 0.02).

[000127] Conclusion: This pilot study shows the prognostic value of M2-TAM density and spatial attributes of M2-TAMs. Additional independent multi-site and prospective validation of these Al derived M2-TAM features, specifically in whole slides images, is warranted.

2^nd Example use case:

[000128] Background: CD163+ M2-subtype tumor-associated macrophages (M2- TAMs) play a significant role in predicting outcomes in various cancers, including their ability to suppress anti-tumoral immune responses and promote tumor progression. In HPV-associated oropharyngeal squamous cell carcinoma (OPSCC), M2-TAMs have been linked to poorer prognosis. Traditionally, M2-TAMs are identified through CD163+ immunohistochemical (IHC) staining. However, this process is labor-intensive and resource-dependent. We introduce a generative artificial intelligence framework designed to predict the presence of M2-TAMs directly from Hematoxylin and Eosin (H&E) images. [000129] Design: H&E-stained images from tissue microarrays (TMA) of HPV- associated OPSCC patients were obtained from two institutions: Vanderbilt University for training (T 1 , N=102) and Emory University for testing (T2, N=50). The images were digitally scanned and immunostained for CD163 (Novocastra; monoclonal; clone 10D6) using a Leica Bond III immunostainer, then co-registered after scanning. To evaluate the efficacy of the disclosed artificial intelligence framework, we compared the generated CD163+ IHC images to ground truth images using metrics such as the Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR). Additionally, we derived a M2-TAM index using the generated CD163+ IHC images for use in survival analysis, defined as the ratio of M2-TAMs to the total number of nuclei. The median M2-TAM index from T1 was used to stratify patients into low- and high-risk categories in both T1 and T2.

[000130] Results: The disclosed artificial intelligence framework significantly improved the quality of the generated CD163+ IHC images in T2 (SSIM = 0.72, PSNR = 20.8) compared to the state-of-the art framework Pyramid Pix2Pix (SSIM = 0.67, PSNR = 19.8). Additionally, the M2-TAM index derived from the disclosed artificial intelligence framework was the only metric significantly associated with survival outcomes in T2 (p = 0.006, Hazard Ratio = 3.96 [1 .03-8.87]), whereas the M2-TAM index from state-of-the-art framework Pyramid Pix2Pix was not significantly associated with survival outcomes (p = 0.26, Hazard Ratio = 1.58 [0.71-3.51]).

[000131] Conclusion: Our results demonstrate that the disclosed artificial intelligence framework could serve as a computational surrogate for generating CD163+ IHC images from H&E images of HPV+ OPSCC and potentially simulate M2- TAM-related biomarkers from routine H&E images. Additional independent multi-site and prospective validation of these Al-derived M2-TAM features, specifically in whole slides images, is warranted.

[000132] Therefore, the present disclosure relates to a method and apparatus configured to generate a medical prognosis using immune cell features e.g., M2- subtype tumor-associated macrophage (M2-TAM) features) extracted from a digitized pathology image from a cancer patient.

[000133] In some embodiments, the present disclosure relates to a method including accessing digitized pathology imaging data from a cancer patient, the digitized pathology imaging data having been generated using immunohistochemical (IHC) stained pathology images and having been segmented to identify a plurality of immune cells; using the plurality of immune cells to extract a plurality of immune cell features from the digitized pathology imaging data; and operating upon the plurality of immune cell features with a machine learning stage that is trained to generate a medical prognosis relating to the cancer patient. In some embodiments, the method further includes identifying a plurality of nuclei within the digitized pathology imaging data, the plurality of immune cells and the plurality of nuclei being used to extract the plurality of immune cell features from the digitized pathology imaging data. In some embodiments, the method further includes immunostaining a pathology slide using a CD163+ immunohistochemical stain; digitizing the pathology slide after immunostaining the pathology slide to generate a digitized IHC stained image; identifying the plurality of immune cells within the digitized IHC stained image; and training a machine learning model to map the plurality of immune cells onto a digitized histologically stained image. In some embodiments, the plurality of immune cells include a plurality of tumor associated macrophages (M2-TAMs); and the plurality of immune cell features include one or more of a M2-TAM density and M2-TAM spatial features. In some embodiments, the method further includes identifying the plurality of M2-TAMs within a region of interest within the digitized pathology imaging data; identifying a plurality of nuclei within the region of interest; and the M2-TAM density being a ratio of a first number of the plurality of M2-TAMs to a second number of the plurality of nuclei. In some embodiments, the method further includes generating a plurality of cell cluster graphs using the plurality of M2-TAMs; and extracting the M2-TAM spatial features from the plurality of cell cluster graphs. In some embodiments, the method further includes immunostaining a pathology slide using a CD163+ immunohistochemical stain; digitizing the pathology slide after immunostaining the pathology slide to generate a digitized IHC stained image; training one or more general adversarial networks on a digitized histologically stained image and the digitized IHC stained image; generating one or more emulated IHC images by operating the one or more general adversarial networks on an additional digitized histologically stained image; and identifying the plurality of immune cells from the one or more emulated IHC images. In some embodiments, the one or more emulated IHC images mimic a CD163+ stained. In some embodiments, the cancer patient has and/or is suspected of having human papilloma virus (HPV) associated with oropharyngeal squamous cell carcinoma (OPSCC). In some embodiments, the method further includes operating the machine learning stage upon a first plurality of immune cell features extracted from digitized images within a training set to determine a median risk score; setting the median risk score as a threshold; operating the machine learning stage upon a second plurality of immune cell features extracted from one or more digitized images within a validation set to determine risk scores associated with one or more cancer patients; and comparing the risk scores to the median risk score to classify the cancer patient as low-risk or high-risk.

[000134] In other embodiments, the present disclosure relates to a non-transitory computer-readable medium storing computer-executable instructions that, when executed, cause a processor to perform operations, including accessing digitized pathology imaging data from a cancer patient, the digitized pathology imaging data including one or more one or more emulated immunohistochemical (IHC) images that have been generated by operating an IHC simulation algorithm upon a digitized histologically stained image; identifying a plurality of M2-subtype tumor-associated macrophages (M2-TAMs) within the one or more emulated IHC images; extracting a plurality of M2-TAM features using the plurality of M2-TAMs, the plurality of M2-TAM features including one or more of a M2-TAM density and M2-TAM spatial features; and providing the plurality of M2-TAM features to a machine learning stage that is trained to generate a medical prognosis relating to the cancer patient. In some embodiments, the one or more emulated IHC images are generated by operating one or more general adversarial networks on the digitized histologically stained image. In some embodiments, the M2-TAM density is a ratio of M2-TAM and total nuclei within a region of interest. In some embodiments, the one or more emulated IHC images include a simulated IHC stained image, an IHC+ segmentation mask, and a nuclei segmentation mask generated by operating a plurality of general adversarial networks on the digitized histologically stained image. In some embodiments, the cancer patient has human papilloma virus (HPV) induced oropharyngeal squamous cell carcinoma. In some embodiments, the machine learning stage is configured to generate the medical prognosis to indicate a poor disease free survival (DFS) when the plurality of M2-TAM features are indicative of a high M2-TAM density.

[000135] In yet other embodiments, the present disclosure relates to an apparatus including a memory configured to store digitized pathology imaging data from a cancer patient, the digitized pathology imaging data having been segmented to identify a plurality of M2-subtype tumor-associated macrophages (M2-TAMs) and a plurality of nuclei; a feature extraction tool configured to access the digitized pathology imaging data and to extract a plurality of M2-TAM features using one or more of the plurality of M2-TAMs and the plurality of nuclei, the plurality of M2-TAM features including one or more of a M2-TAM density and M2-TAM spatial features; and a machine learning stage configured to use the plurality of M2-TAM features to generate a medical prognosis relating to the cancer patient. In some embodiments, the machine learning stage is configured to generate a risk score using the plurality of M2-TAM features and to compare the risk score to a threshold to determine the medical prognosis. In some embodiments, the apparatus further includes a machine learning identification tool having an IHC simulator including one or more general adversarial networks, the IHC simulator being configured to operate the one or more general adversarial networks on a histologically stained image to generate one or more emulated IHC images; and the machine learning identification tool being further configured to identify the plurality of M2-TAMs from the one or more emulated IHC images. In some embodiments, the apparatus further includes a machine learning identification tool having an IHC simulator including a plurality of general adversarial networks, the IHC simulator being configured to operate the plurality of general adversarial networks on a histologically stained image to generate a simulated IHC stained image, an IHC+ segmentation mask, and a nuclei segmentation mask.

[000136] It will be appreciated that the disclosed methods and/or block diagrams may be implemented as computer-executable instructions, in some embodiments. Thus, in one example, a computer-readable storage device (e.g.. a non-transitory computer- readable medium) may store computer executable instructions that if executed by a machine (e.g., computer, processor) cause the machine to perform the disclosed methods and/or block diagrams. While executable instructions associated with the disclosed methods and/or block diagrams are described as being stored on a computer-readable storage device, it is to be appreciated that executable instructions associated with other example disclosed methods and/or block diagrams described or claimed herein may also be stored on a computer-readable storage device.

[000137] Examples herein can include subject matter such as an apparatus, including a digital whole slide scanner, a personalized medicine system, a CADx system, a processor, a system, circuitry, a method, means for performing acts, steps, or blocks of the method, at least one machine-readable medium including executable instructions that, when performed by a machine (e.g., a processor with memory, an application- specific integrated circuit (ASIC), a field programmable gate array (FPGA), or the like) cause the machine to perform acts of the method or of an apparatus or system, according to embodiments and examples described.

[000138] References to “one embodiment”, “an embodiment”, “one example”, and “an example” indicate that the embodiment(s) or example(s) so described may include a particular feature, structure, characteristic, property, element, or limitation, but that not every embodiment or example necessarily includes that particular feature, structure, characteristic, property, element or limitation. Furthermore, repeated use of the phrase “in one embodiment” does not necessarily refer to the same embodiment, though it may.

[000139] “Computer-readable storage device”, as used herein, refers to a device that stores instructions or data. “Computer-readable storage device” does not refer to propagated signals. A computer-readable storage device may take forms, including, but not limited to, non-volatile media, and volatile media. Non-volatile media may include, for example, optical disks, magnetic disks, tapes, and other media. Volatile media may include, for example, semiconductor memories, dynamic memory, and other media. Common forms of a computer-readable storage device may include, but are not limited to, a floppy disk, a flexible disk, a hard disk, a magnetic tape, other magnetic medium, an application specific integrated circuit (ASIC), a compact disk (CD), other optical medium, a random access memory (RAM), a read only memory (ROM), a memory chip or card, a memory stick, and other media from which a computer, a processor or other electronic device can read.

[000140] “Circuit”, as used herein, includes but is not limited to hardware, firmware, software in execution on a machine, or combinations of each to perform a function(s) or an action(s), or to cause a function or action from another logic, method, or system. A circuit may include a software controlled microprocessor, a discrete logic (e.g., ASIC), an analog circuit, a digital circuit, a programmed logic device, a memory device containing instructions, and other physical devices. A circuit may include one or more gates, combinations of gates, or other circuit components. Where multiple logical circuits are described, it may be possible to incorporate the multiple logical circuits into one physical circuit. Similarly, where a single logical circuit is described, it may be possible to distribute that single logical circuit between multiple physical circuits.

[000141] To the extent that the term “includes” or “including” is employed in the detailed description or the claims, it is intended to be inclusive in a manner similar to the term “comprising” as that term is interpreted when employed as a transitional word in a claim.

[000142] Throughout this specification and the claims that follow, unless the context requires otherwise, the words 'comprise' and 'include' and variations such as 'comprising' and 'including' will be understood to be terms of inclusion and not exclusion. For example, when such terms are used to refer to a stated integer or group of integers, such terms do not imply the exclusion of any other integer or group of integers.

[000143] To the extent that the term “or” is employed in the detailed description or claims (e.g., A or B) it is intended to mean “A or B or both”. When the applicants intend to indicate “only A or B but not both” then the term “only A or B but not both” will be employed. Thus, use of the term “or” herein is the inclusive, and not the exclusive use. See, Bryan A. Garner, A Dictionary of Modern Legal Usage 624 (2d. Ed. 1995).

[000144] While example systems, methods, and other embodiments have been illustrated by describing examples, and while the examples have been described in considerable detail, it is not the intention of the applicants to restrict or in any way limit the scope of the appended claims to such detail. It is, of course, not possible to describe every conceivable combination of components or methodologies for purposes of describing the systems, methods, and other embodiments described herein. Therefore, the invention is not limited to the specific details, the representative apparatus, and illustrative examples shown and described. Thus, this application is intended to embrace alterations, modifications, and variations that fall within the scope of the appended claims.

Claims

What is claimed is:

1 . A method, comprising: accessing digitized pathology imaging data from a cancer patient, wherein the digitized pathology imaging data has been generated using immunohistochemical (IHC) stained pathology images and has been segmented to identify a plurality of immune cells; using the plurality of immune cells to extract a plurality of immune cell features from the digitized pathology imaging data; and operating upon the plurality of immune cell features with a machine learning stage that is trained to generate a medical prognosis relating to the cancer patient.

2. The method of claim 1 , further comprising: identifying a plurality of nuclei within the digitized pathology imaging data, wherein the plurality of immune cells and the plurality of nuclei are used to extract the plurality of immune cell features from the digitized pathology imaging data.

3. The method of claim 1 , further comprising: immunostaining a pathology slide using a CD163+ immunohistochemical stain; digitizing the pathology slide after immunostaining the pathology slide to generate a digitized IHC stained image; identifying the plurality of immune cells within the digitized IHC stained image; and training a machine learning model to map the plurality of immune cells onto a digitized histologically stained image.

4. The method of claim 1 , wherein the plurality of immune cells comprise a plurality of tumor associated macrophages (M2-TAMs); and wherein the plurality of immune cell features comprise one or more of M2-TAM density and M2-TAM spatial features.

5. The method of claim 4, further comprising: identifying the plurality of M2-TAMs within a region of interest within the digitized pathology imaging data; identifying a plurality of nuclei within the region of interest; and wherein the M2-TAM density is a ratio of a first number of the plurality of M2-TAMs to a second number of the plurality of nuclei.

6. The method of claim 4, further comprising: generating a plurality of cell cluster graphs using the plurality of M2-TAMs; and extracting the M2-TAM spatial features from the plurality of cell cluster graphs.

7. The method of claim 1 , further comprising: immunostaining a pathology slide using a CD163+ immunohistochemical stain; digitizing the pathology slide after immunostaining the pathology slide to generate a digitized IHC stained image; training one or more general adversarial networks on a digitized histologically stained image and the digitized IHC stained image; generating one or more emulated IHC images by operating the one or more general adversarial networks on an additional digitized histologically stained image; and identifying the plurality of immune cells from the one or more emulated IHC images.

8. The method of claim 7, wherein the one or more emulated IHC images mimic a CD163+ stained.

9. The method of claim 1 , wherein the cancer patient has and/or is suspected of having human papilloma virus (HPV) associated with oropharyngeal squamous cell carcinoma (OPSCC).

10. The method of claim 1 , further comprising: operating the machine learning stage upon a first plurality of immune cell features extracted from digitized images within a training set to determine a median risk score; setting the median risk score as a threshold; operating the machine learning stage upon a second plurality of immune cell features extracted from one or more digitized images within a validation set to determine risk scores associated with one or more cancer patients; and comparing the risk scores to the median risk score to classify the cancer patient as low-risk or high-risk.

1 1. A non-transitory computer-readable medium storing computer-executable instructions that, when executed, cause a processor to perform operations, comprising: accessing digitized pathology imaging data from a cancer patient, wherein the digitized pathology imaging data comprises one or more one or more emulated immunohistochemical (IHC) images that have been generated by operating an IHC simulation algorithm upon a digitized histologically stained image; identifying a plurality of M2-subtype tumor-associated macrophages (M2-TAMs) within the one or more emulated IHC images; extracting a plurality of M2-TAM features using the plurality of M2-TAMs, wherein the plurality of M2-TAM features comprise one or more of a M2-TAM density and M2- TAM spatial features; and providing the plurality of M2-TAM features to a machine learning stage that is trained to generate a medical prognosis relating to the cancer patient.

12. The non-transitory computer-readable medium of claim 11 , wherein the one or more emulated IHC images are generated by operating one or more general adversarial networks on the digitized histologically stained image.

13. The non-transitory computer-readable medium of claim 1 1 , wherein the M2-TAM density is a ratio of M2-TAMs and total nuclei within a region of interest.

14. The non-transitory computer-readable medium of claim 11 , wherein the one or more emulated IHC images comprise a simulated IHC stained image, an IHC+ segmentation mask, and a nuclei segmentation mask generated by operating a plurality of general adversarial networks on the digitized histologically stained image.

15. The non-transitory computer-readable medium of claim 1 1 , wherein the cancer patient has human papilloma virus (HPV) induced oropharyngeal squamous cell carcinoma.

16. The non-transitory computer-readable medium of claim 1 1 , wherein the machine learning stage is configured to generate the medical prognosis to indicate a poor disease free survival (DFS) when the plurality of M2-TAM features are indicative of a high M2-TAM density.

17. An apparatus, comprising: a memory configured to store digitized pathology imaging data from a cancer patient, wherein the digitized pathology imaging data has been segmented to identify a plurality of M2-subtype tumor-associated macrophages (M2-TAMs) and a plurality of nuclei; a feature extraction tool configured to access the digitized pathology imaging data and to extract a plurality of M2-TAM features using one or more of the plurality of M2- TAMs and the plurality of nuclei, wherein the plurality of M2-TAM features comprise one or more of a M2-TAM density and M2-TAM spatial features; and a machine learning stage configured to use the plurality of M2-TAM features to generate a medical prognosis relating to the cancer patient.

18. The apparatus of claim 17, wherein the machine learning stage is configured to generate a risk score using the plurality of M2-TAM features and to compare the risk score to a threshold to determine the medical prognosis.

19. The apparatus of claim 17, further comprising: a machine learning identification tool comprising an IHC simulator including one or more general adversarial networks, wherein the IHC simulator is configured to operate the one or more general adversarial networks on a histologically stained image to generate one or more emulated IHC images; and wherein the machine learning identification tool is further configured to identify the plurality of M2-TAMs from the one or more emulated IHC images.

20. The apparatus of claim 17, further comprising: a machine learning identification tool comprising an IHC simulator including a plurality of general adversarial networks, wherein the IHC simulator is configured to operate the plurality of general adversarial networks on a histologically stained image to generate a simulated IHC stained image, an IHC+ segmentation mask, and a nuclei segmentation mask.