Use of 3D glandular network spatial features to generate medical prognosis for cancer patients

Abstract

The present disclosure, in some embodiments, relates to a method. The method includes accessing one or more digitized three-dimensional (3D) pathology images of glandular tissue from a cancer patient. One or more 3D gland models are generated using the one or more digitized 3D pathology images. A plurality of glandular architecture features are extracted from the one or more 3D gland models. The plurality of glandular architecture features are provided to a machine learning model that has been trained to generate a medical prognosis for the cancer patient.

Description

USE OF 3D GLANDULAR NETWORK SPATIAL FEATURES TO GENERATE MEDICAL PROGNOSIS FOR CANCER PATIENTSREFERENCE TO RELATED APPLICATION

[0001] This Application claims the benefit of U.S. Provisional Application No. 63 / 567,969, filed on March 21 , 2024, the contents of which are incorporated by reference in their entirety.FEDERAL FUNDING INFORMATION

[0002] This invention was made with government support under R01CA268207 awarded by the National Institutes of Health / National Cancer Institute. The government has certain rights in the invention.BACKGROUND

[0003] Prostate cancer is an uncontrolled growth of cells in the prostate, a small walnut-shaped gland in males that produces seminal fluid that nourishes and transports sperm. Prostate cancer is one of the most common types of cancer. Prostate cancer typically causes no symptoms in its early stages. While some types of prostate cancer grow slowly and may need minimal or even no treatment, other types are aggressive and can spread quickly. Prostate cancer that's detected when it's still confined to the prostate gland has the best chance for successful treatment.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 an assessment system configured to utilize glandular architecture features extracted from a glandular model formed using one or more digitized 3D pathology images to make a medical prognosis.

[0006] Fig. 2 illustrates some embodiments of a BCR assessment system configured to utilize glandular architecture features extracted from a glandular model to make a medical prognosis.

[0007] Fig. 3 Illustrates an exemplary workflow of a disclosed BCR prognostic system configured to utilize glandular architecture features extracted from a glandular skeleton to make a medical prognosis.

[0008] Fig. 4 illustrates exemplary violin plots illustrating selected feature distributions for non-BCR and BCR patients.

[0009] Fig. 5 illustrates an exemplary receiver operating characteristic (ROC) curve corresponding to a disclosed BCR assessment system.

[0010] Fig. 6 illustrates a flow diagram showing some embodiments of a method of using glandular architecture features extracted from a glandular model to make a medical prognosis.

[0011] Fig. 7 illustrates some additional embodiments of a BCR assessment system configured to utilize glandular architecture features extracted from a glandular model to make a medical prognosis.

[0012] Fig. 8 illustrates some additional embodiments of a BCR assessment system configured to utilize glandular architecture features extracted from a glandular model to make a medical prognosis.

[0013] Fig. 9 illustrates some embodiments of a block diagram of an apparatus configured to generate a medical prognosis using glandular architecture features extracted from a glandular model formed using one or more digitized 3D pathology images.DETAILED DESCRIPTION

[0014] 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 setforth 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.

[0015] Prostate cancer is often detected using screening tests. For example, blood tests may be used to detect levels of prostate-specific antigen (PSA), which indicate unusual growth of prostate tissue. If high levels of PSA are detected, a diagnosis of prostate cancer typically requires a biopsy of the prostate. The biopsy removes tissue from the prostate. The tissue is subsequently embedded in paraffin and sliced into thin sections that are used to form two-dimensional (2D) histology slides. Examination of the 2D histology slides is then performed to diagnose prostate cancer and / or to identify cancer staging.

[0016] Once a patient has been diagnosed with prostate cancer, treatment options for the patient may vary depending on factors such as cancer stage, aggressiveness, overall patient health, and / or the like. One common treatment for early stage prostate cancer is a radical prostatectomy. A radical prostatectomy is a surgical procedure during which an entire prostate gland, some surrounding tissue, and the seminal vesicles may be removed. The goal of a radical prostatectomy is to remove all cancer cells.

[0017] In some cases, after a radical prostatectomy, PSA levels may begin to rise again in a patient’s blood. Biochemical recurrence (BCR) is a condition in which PSA levels in the blood of a prostate cancer patient increase after treatment with surgery and / or radiation (e.g., to above levels exceeding 0.2 ng / mL). Current tools for predicting BCR after a radical prostatectomy often depend on parameters determined by pathologists, such as tumor grade. However, tumor grade is known to vary between reviewers. Furthermore, the examination of a complex three- dimensional (3D) prostate gland using 2D sections may distort a 3D architecture of the gland leading to misdiagnosis of cancer stage. Genomic risk classifiers may also provide useful information relating to BCR, but require substantial tissue, are costly, and are not commonly accessible in many medical centers. Therefore, there is an urgent, unmet clinical need for predictive biomarkers that can guide therapeutic decision-making, minimize ineffective treatments, and enable more assertive therapy in prostate cancer patients showing a high-risk of BCR.

[0018] In some embodiments, the present disclosure relates to a method and apparatus configured to utilize glandular architecture features extracted from a glandular model formed using one or more digitized three-dimensional (3D) pathology images to make a medical prognosis relating to BCR. In some embodiments, the method may be performed by accessing one or more digitized 3D pathology images of prostate tissue from a cancer patient having undergone a radical prostatectomy for treatment of prostate cancer. One or more 3D prostate gland models are generated using the one or more digitized 3D pathology images, and a plurality of glandular architecture features are extracted from the one or more 3D prostate gland models. The plurality of glandular architecture features are provided to a machine learning model that has been trained to generate a medical prognosis regarding a risk of biochemical recurrence (BCR) for the cancer patient. It has been determined that operating a machine learning model on glandular architecture features extracted from a glandular model formed from digitized 3D pathology images results in a highly accurate medical prognosis, which can be utilized by health care professionals to make a more informed decision relating to the treatment of a cancer patient and thereby allow the cancer patient to have an improved quality of life (e.g., to avoid post operative treatment that is not likely to produce positive outcomes).

[0019] Fig. 1 illustrates some embodiments of a block diagram of an assessment system 100 configured to utilize glandular architecture features extracted from a glandular model formed using one or more digitized 3D pathology images to make a medical prognosis.

[0020] The assessment system 100 comprises a memory 101. The memory 101 is configured to store digitized pathology imaging data including one or more digitized three-dimensional (3D) pathology images 102. In some embodiments, the one or more digitized 3D pathology images 102 may be images of tissue taken from a cancer patient 104 (e.g., a prostate cancer patient, an endocrine cancer patient, a cervical cancer patient, and / or the like). In some embodiments, the one or more digitized 3D pathology images 102 may include glandular tissue. For example, the one or more digitized 3D pathology images 102 may be images of prostate tissue that has been excised from the cancer patient during a radical prostatectomy and subsequently imaged by a 3D image generator 106.

[0021] A gland modeling tool 108 is configured to access the one or more digitized 3D pathology images 102. The gland modeling tool 108 is further configured to use the one or more digitized 3D pathology images 102 to generate one or more 3D gland models that respectively model a gland to show a topology, geometry, and / or scale of a gland in three-dimensions. For example, the one or more 3D gland models 110 may comprise one or more 3D prostate gland models that respectively model a prostate gland in three-dimensions. The one or more digitized 3D pathology images 102 provide for increased microscopic sampling over 2D images, thereby allowing for more reliable modeling of a gland that can reveal a spatial organization of gland networks and a volumetric shape of each gland. The more reliable modeling of the gland can allow for morphological information (e.g., length, tortuosity, torsion, etc.) to be extracted, thereby reducing variability and subjectivity of human interpretation that is present in 2D images and that can result in an unreliable medical prognosis.

[0022] In some embodiments, the one or more 3D gland models 1 10 may comprise skeleton models (e.g., models formed using a skeleton algorithm) that show how a gland branches and winds along one or more continuous paths extending through an organ. In some embodiments, the one or more 3D gland models 110 may comprise tree-shaped structures with branches that meet at junctions. In some embodiments, the one or more 3D gland models 110 may comprise models of 3D glandular networks including a plurality of glands respectively comprising tree-shaped structures with branches.

[0023] A feature extraction tool 1 12 is configured to extract a plurality of glandular architecture features 114 from the one or more 3D gland models 110. In some embodiments, the plurality of glandular architecture features 1 14 may include features relating to dimensions, structures, and / or contortions of the one or more 3D gland models 110. In some embodiments, the plurality of glandular architecture features 114 may quantify a tortuosity using multiple metrics that capture how glandular paths deviate from straight lines, reflecting geometrical changes of gland architecture. For example, the plurality of glandular architecture features 1 14 may quantify local and global curvature measurements that describe a smoothness of gland networks and branch angle features (e.g., an angle between branches) that characterize a spatial relationships between connecting ducts, both of which becomemore chaotic in poorly differentiated tumors. In some embodiments, the plurality of glandular architecture features 1 14 may include length features that capture the physical scale of single glands and their size variability. The plurality of glandular architecture features 114 provide a characterization of a glandular architecture that aligns with known biological changes during cancer progression. In some embodiments, the plurality of glandular architecture features 1 14 include measures of network topology through branching patterns and branch-point density, which reflect how the hierarchical organization of glands changes during cancer progression.

[0024] The plurality of glandular architecture features 114 are provided to a machine learning model 116 that is configured to generate a medical prognosis 118. In some embodiments, the medical prognosis 118 may relate to survival (e.g., overall survival, disease free survival, etc.), a treatment response, and / or the like of the cancer patient 104. In some embodiments, the medical prognosis 118 may correspond to a determination as to whether or not the cancer patient 104 is likely to experience biochemical recurrence (BCR) free survival and / or a time to recurrence. For example, the medical prognosis 118 may categorize the cancer patient 104 as having a low-risk 120 of BCR or a high-risk 122 of BCR. While adjuvant therapy is effective in reducing metastasis and disease-specific death, it is not universally suitable for prostate cancer patients due to the low overall mortality rate of prostate cancer. Therefore, by providing an accurate risk estimate for BCR, the medical prognosis 118 can assist in identifying cancer patients who may benefit from adjuvant therapy and thereby avoid unnecessary treatment for cancer patients at low-risk of BCR.

[0025] Fig. 2 illustrates some embodiments of a biochemical recurrence (BCR) assessment system 200 configured to utilize glandular architecture features extracted from a glandular model to make a medical prognosis.

[0026] The BCR assessment system 200 comprises a memory 101 . The memory 101 is configured to store digitized pathology imaging data including one or more digitized 3D pathology images 102 obtained from a pathological tissue sample 204 taken from a prostate of a cancer patient 104. The one or more digitized 3D pathology images 102 offer additional information over 2D histology images, thereby allowing for significantly increased microscopic sampling that leads to more reliablesegmentation of objects that are relatively continuous in space {e.g., prostate glands). 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).

[0027] In some embodiments, the one or more digitized 3D pathology images 102 may be generated from the pathological tissue sample 204 e.g., a prostate tissue sample) surgically excised from a cancer patient 104 using a tissue resection tool 202 {e.g., a scalpel, a needle, scissors, a biopsy punch, and / or the like). The pathological tissue sample 204 comprises a 3D tissue sample {e.g., a non-sectioned tissue sample). The pathological tissue sample 204 is provided to a 3D image generator 106 that is configured to digitize the pathological tissue sample 204 to form the one or more digitized 3D pathology images 102. In some embodiments, the 3D image generator 106 may comprise a light-sheet microscopy platform 206 {e.g., an open-top light-sheet microscopy platform). The light-sheet microscopy platform 206 comprises two objective lenses that are configured to separate illumination and collection paths of the microscope. In some embodiments, the two objective lenses are placed at an orientation of 90° with respect to one another to allow for the pathological tissue sample 204 to be selectively illuminated with a thin sheet of light that can be used to generate a plurality of spatially off-set cross-sections of the pathological tissue sample 204. For example, the light-sheet microscopy platform 206 may generate a plurality of vertically adjacent cross-sections of the pathological tissue sample 204. In some embodiments, the one or more digitized 3D pathology images 102 may be configured to span an entirety of the pathological tissue sample 204. In some embodiments, the pathological tissue sample 204 may be stained {e.g., using an H&E stain) prior to imaging.

[0028] A gland modeling tool 108 is configured to access the one or more digitized 3D pathology images 102 and to use the one or more digitized 3D pathology images 102 to generate one or more 3D prostate gland models 110p. In some embodiments, the gland modeling tool 108 may comprise one or more of an immunofluorescence generator 208, a mask generator 212, and a skeleton algorithm 216.

[0029] The immunofluorescence generator 208 is configured to convert the one or more digitized 3D pathology images 102 into an immunofluorescence dataset 210.In some embodiments, the immunofluorescence dataset 210 may mimic 3D immunofluorescence images of cytokeratin 8 (CK8). For example, the immunofluorescence generator 208 may be configured to synthetically convert 3D H&E images to mimic 3D immunofluorescence images of CK8. Because immunolabeling is expensive and time-consuming (e.g., due to the slow diffusion of antibodies in thick tissues), the immunofluorescence generator 208 may provide for a quick and relatively low cost method for generating the immunofluorescence dataset 210. In some embodiments, the immunofluorescence generator 208 may comprise a general adversarial network (GAN). In some embodiments, the immunofluorescence generator 208 may comprise an image-sequence translation 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).

[0030] In some embodiments, the mask generator 212 is configured to operate upon the immunofluorescence dataset 210 to identify one or more prostate glands within the immunofluorescence dataset 210. The mask generator 212 may be configured to use the immunofluorescence dataset 210 to generate a glandular structure mask 214 that identifies the one or more prostate glands. In other embodiments, the mask generator 212 may comprise a deep learning model configured to use a deep learning algorithm (e.g., having a UNet architecture) to segment glandular networks of the one or more digitized 3D pathology images 102 and generate the glandular structure mask 214. By using the deep learning algorithm to generate the one or more 3D prostate gland models 110p, the use of immunofluorescence imaging data can be avoided (e.g., so that the gland modeling tool 108 does not include the immunofluorescence generator 208). In some embodiments, the glandular structure mask 214 may comprise a binary mask having a value of “1 ” in image units (e.g., pixels, voxels, etc.) identified as being within the one or more prostate glands and having a value of “0” in image units outside of the one or more prostate glands. In some embodiments, the mask generator 212 may comprise a machine learning model configured to run a computer vision algorithm implemented as computer code run on one or more processors (e.g., a CPU, a microcontroller, a GPU, and / or the like).

[0031] The skeleton algorithm 216 is configured to operate upon the glandular structure mask 214 to generate the one or more 3D prostate gland models 11 Op. The one or more 3D prostate gland models 11 Op model a three-dimensional structure of the prostate glands. In some embodiments, the one or more 3D prostate gland models 1 1 Op may comprise models of 3D glandular networks respectively including a plurality of glands. In some embodiments, the one or more 3D prostate gland models 110p may comprise tree-shaped structures with branches that meet at junctions. In some embodiments, the skeleton algorithm 216 may be implemented as computer code run on one or more processors (e.g., a CPU, a microcontroller, a GPU, and / or the like).

[0032] In some embodiments, the skeleton algorithm 216 may comprise a Treestructure Extraction Algorithm delivering Skeletons that are Accurate and Robust (TEASER). It has been appreciated that using a TEASER algorithm having a relatively low granularity (e.g., that uses a relatively large cube size) will provide for improved prediction of recurrence over a TEASER algorithm having a higher granularity. For example, a TEASER algorithm that uses a cube size of 5 units (e.g., voxels) will have a higher granularity, but a lower accuracy, than a TEASER algorithm that uses a cube size of 180 units.

[0033] In other embodiments, the skeleton algorithm 216 may form one or more 3D prostate gland models using 2D glandular slices from a 3D volume. In some such embodiments, a center of mass of each lumen contour may be considered as a point in a skeleton approximation. This approximation offers the advantage of leveraging 2D lumen knowledge to ascertain a center of mass, providing an easily interpretable and computationally inexpensive solution.

[0034] A feature extraction tool 112 is configured to extract a plurality of glandular architecture features 114 from the one or more 3D prostate gland models 110p. The feature extraction tool 112 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 glandular architecture features 114 may include handcrafted features. In some embodiments, the plurality of glandular architecture features 114 may include one or more glandular spatial features 218 and one or more glandular curvature features 220. The one or more glandular spatial features 218 describe spatial attributes of the one or more 3D prostate gland models 110p. Insome embodiments, the one or more glandular spatial features 218 may comprise features that describe a length of a prostate gland and / or branch, a diameter of a prostate gland and / or branch, a branch to branch distance, a branch to junction distance, a junction-to-junction distance, and / or the like. In some embodiments, the one or more glandular curvature features 220 may include a curvature and a tortuosity of a prostate gland.

[0035] In some embodiments, the one or more glandular spatial features 218 and / or the one or more glandular curvature features 220 may include one or more statistical measures (e.g., mean, median, skew, kurtosis, standard deviation, and / or the like). For example, in some embodiments the one or more glandular spatial features 218 may be a junction-to-junction mean distance, a junction-to-junction histogram, a mean distance of branches of a prostate gland, and / or the like. In some embodiments, the one or more glandular curvature features 220 may be a mean value of curvature, a mean value of tortuosity, and / or the like.

[0036] The plurality of glandular architecture features 114 are provided to a machine learning model 116 that is configured to generate a medical prognosis 118 relating to BCR. In some embodiments, the medical prognosis 118 may correspond to a determination as to whether or not the cancer patient 104 is likely to experience BCR free survival. For example, the medical prognosis 118 may categorize a cancer patient as having a low-risk 120 of BCR or a high-risk 122 of BCR. In some embodiments, the machine learning model 1 16 may comprise a K-nearest neighbor classifier, a Naive Bayes classifier, a random forest classifier, and / or the like. In some embodiments, the machine learning model 116 may be implemented as computer code run on one or more processors (e.g., a CPU, a microcontroller, a GPU, and / or the like). The medical prognosis 118 concerning BCR can be utilized by health care professionals to make a more informed decision relating to the treatment of the cancer patient 104, thereby allowing for the cancer patient 104 to have an improved quality of life.

[0037] In some embodiments, the medical prognosis 118 may be determined to have a direct correlation between longer gland length and BCR free survival. For example, the medical prognosis 118 may be determined to indicate that cancer patients having longer glands are more likely to experience BCR free survival (e.g., be classified as low-risk 120) than cancer patients having shorter glands. In someembodiments, the medical prognosis 118 may be determined to have an indirect correlation between greater gland curvature and BCR free survival. For example, the medical prognosis 118 may be determined to indicate that cancer patients having more twisted glands are less likely to experience BCR free survival (e.g., be classified as high-risk 122) than cancer patients having less twisted glands.

[0038] Fig. 3 illustrates an exemplary workflow of a disclosed BCR assessment system that is configured to utilize glandular architecture features extracted from a glandular model to make a medical prognosis.

[0039] The workflow includes an image acquisition stage 300, an image segmentation stage 312, and a skeletonization stage 318. During the image acquisition stage 300, a tissue sample is obtained from prostate tissue extracted during a radical prostatectomy 302. In various embodiments, the tissue sample may be obtained according to different methods. For example, in some embodiments the tissue sample may be obtained by way of a core needle biopsy 304a while in other embodiments the tissue sample may be obtained by way of a punch biopsy 304b. The tissue sample is stained and optically cleared 306. In some embodiments, the tissue sample may be stained for 48 hours in 70% ethanol (pH 4) using a 1 :200 dilution of Eosin-Y and 1 :500 dilution of To-PRO-3 Iodide at room temperature with gentle agitation. Following dehydration in 100% ethanol (twice for 2 hours), the tissue sample may be optically cleared in ethyl cinnamate for 8 hours. Image acquisition may be performed using an open top light sample (OTLS) microscopy 308 to generate a 3D image of a core needle biopsy 310a or a 3D image of a punch biopsy 310b. The OTLS microscope may image each tissue sample using two laser wavelengths (e.g., 488nm and 638nm) with ethyl cinnamate as the immersion medium.

[0040] During the image segmentation stage 312, the 3D images 310a-310b may be segmented to identify a gland, lumen, and stroma. In some embodiments, the 3D image of the core needle biopsy 310a may be converted into a synthetic CK8 (cytokeratin 8) immunofluorescence dataset 314a using an image-sequence translation model comprising a deep learning image translation algorithm configured to translate analogue H&E-stained images into synthetic CK8 IHC (immunohistochemistry) stains. A thresholding algorithm is then applied to the synthetic CK8 immunofluorescence dataset 314a to generate a segmented coreneedle biopsy 316a by first segmenting the glandular epithelium and then segmenting the glandular lumen area by filing epithelial regions. In other embodiments, the 3D image of the punch biopsy 310b may be segmented to generate a segmented punch biopsy 316b by using a nnllNet 314b that extracts information directly from an input dataset and determines how to optimize various hyperparameters and identify the glands, lumen, and stroma.

[0041] During the skeletonization stage 318, the lumen identified in the segmented biopsies 316 are isolated and taken as a reference for the glandular morphology. A skeletonization algorithm is used to characterize the glandular morphology to generate highly complex forms within a 3D space. The skeletonization algorithm may be applied to each individual lumen to generate skeleton models (e.g., skeletons) that characterize gland morphology. The skeletons may represent a volume shape of each gland, as shown in 320, or branches of different lengths 322 according to a gland’s shape, curvature, and tortuosity.

[0042] Fig. 4 illustrates exemplary violin plots 400-408 illustrating selected feature distributions for non-BCR patients (e.g., patients categorized as low-risk for BCR) and BCR patients (e.g., patients categorized as high-risk for BCR). The violin plots 400-408 include three violin plots 400-404 corresponding to glandular spatial features related to branch distances and junction-to-junction distances of one or more prostate glands. The violin plots 400-408 further include two violin plots 406- 408 corresponding to glandular curvature features related to the curvature and tortuosity of branches of one or more prostate glands.

[0043] As shown in the violin plots 400-408, differences in glandular architecture features between the BCR patients (e.g., patients categorized as high-risk for BCR) and the non-BCR patients (e.g., patients categorized as low-risk for BCR) were found to be statistically significant. This suggests that the use of glandular architecture features extracted from skeleton models can be used to predict BCR in patients (e.g., to differentiate between non-BCR / low-risk patients and BCR / high-risk patients).

[0044] Fig. 5 illustrates an exemplary receiver operating characteristic (ROC) curve 500 corresponding to a disclosed BCR assessment system.

[0045] The ROC curve 500 shows a true positive rate along a y-axis and a false positive rate along an x-axis. The average area under the ROC curve (AUC) was0.73 with a standard deviation of 0.04 using five gland architecture features including a junction-to-junction mean distance, a junction-to-junction histogram, a mean distance of all branches of a prostate gland, a mean value of curvature, and a mean value of tortuosity. The relatively high AUC of the ROC curve 500 further suggests that the use of glandular architecture features extracted from skeleton models can be used to accurately predict BCR in cancer patients.

[0046] Therefore, the disclosed BCR assessment system improves a computer’s ability to analyze medical images in a manner that accurately identifies cancer patients that are likely to experience BCR. The improved ability to accurately identify cancer patients that are likely to experience BCR can improve treatment of the cancer patients.

[0047] Fig. 6 illustrates a flow diagram showing some embodiments of a method 600 of using glandular architecture features extracted from a glandular model to make a medical prognosis.

[0048] While the disclosed method 600 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.

[0049] At act 602, one or more digitized 3D pathology images of glandular tissue from one or more cancer patients are accessed. In some embodiments, the one or more cancer patients may include cancer patients that have undergone a radical prostatectomy as treatment for prostate cancer.

[0050] At act 604, one or more 3D gland models are generated using the one or more digitized 3D pathology images. In some embodiments, the one or more 3D gland models may be generated according to acts 606-610.

[0051] At act 606, an immunofluorescence dataset may be generated from the one or more digitized 3D pathology images. In some embodiments, an imagesequence translation model may be used to generate the immunofluorescence dataset from the one or more digitized 3D pathology images.

[0052] At act 608, a glandular structure mask is formed from the immunofluorescence dataset. In some embodiments, a computer vision algorithm may be used to form the glandular structure mask from the immunofluorescence dataset. In other embodiments (not shown), the glandular structure mask may be formed by operating a deep learning algorithm upon the one or more digitized 3D pathology images.

[0053] At act 610, one or more 3D prostate gland models are formed from the glandular structure mask. In some embodiments, a skeleton algorithm may be used to form the one or more 3D prostate gland models from the glandular structure mask.

[0054] At act 612, a plurality of glandular architecture features are extracted from the one or more 3D gland models. In some embodiments, the plurality of glandular architecture features may be extracted according to acts 614-616.

[0055] At act 614, a plurality of glandular spatial features are extracted from the one or more 3D prostate gland models.

[0056] At act 616, a plurality of glandular curvature features are extracted from the one or more 3D prostate gland models.

[0057] At act 618, a machine learning model is operated on the plurality of glandular architecture features to generate a medical prognosis. In some embodiments, the medical prognosis may relate to biochemical recurrence (BCR). In other embodiments, the medical prognosis may relate to survival, a treatment response, and / or the like.

[0058] At act 620, 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.

[0059] Therefore, the disclosed method 600 utilizes glandular architecture features extracted from one or more 3D models to generate a medical prognosis of a cancer patient.

[0060] 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 ifexecuted 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.

[0061] Fig. 7 illustrates some additional embodiments of a BCR assessment system 700 configured to utilize glandular features extracted from a glandular model to make a medical prognosis.

[0062] The BCR assessment system 700 comprises a memory 101 configured to store an imaging data including a plurality of digitized 3D pathology images 102 of prostate tissue from a cancer patient 104 and one or more 3D prostate gland models 110p corresponding to the plurality of digitized 3D pathology images 102. A feature extraction tool 1 12 is configured to extract a plurality of glandular architecture features 114 from the one or more 3D prostate gland models 110p. A machine learning model 116 is configured to use the plurality of glandular architecture features 114 to generate a medical prognosis 118.

[0063] In some embodiments, the memory 101 may be further configured to store a cancer stage 702 (e.g., a T-stage describing a size of a main tumor) relating to the cancer patient 104. The machine learning model 116 may be configured to receive the cancer stage 702 as part of an input vector. It has been appreciated that the machine learning model 116 may demonstrate an improved prognostic value for generating the medical prognosis 118 for BCR when combined with the cancer stage 702 independent of other traditional clinical factors.

[0064] In some other embodiments, the memory 101 may be further configured to store demographic information 704 of the cancer patient 104 (e.g., demographic information identifying the cancer patient as white, black, Asian, etc.). In some embodiments, the machine learning model 1 16 may be configured to receive the demographic information 704 as part of an input vector. It has been appreciated that different architectural patterns of cancer progression may be present between racial groups. Therefore, the machine learning model 116 may demonstrate an improved prognostic value for generating the medical prognosis 118 for BCR when combinedwith the demographic information 704. In some embodiments, the machine learning model 116 may comprise a plurality of different population specific models {e.g., a first model for black patients, a second model for white patients, a third model for Asian patients, etc.) that may be selectively used depending on the demographic information 704.

[0065] Fig. 8 illustrates some additional embodiments of a BCR assessment system 800 configured to utilize glandular features extracted from a glandular model to make a medical prognosis.

[0066] The BCR assessment system 800 comprises a memory 101 configured to store digitized pathology imaging data including one or more digitized 3D pathology images 102 of tissue {e.g., prostate tissue) from cancer patients 104. In some embodiments, the cancer patients 104 have undergone radical prostatectomies. In various embodiments, the one or more digitized 3D pathology images 102 may be from a 3D image generator 106 and / or from an on-line database 802 and / or archive containing digitized 3D pathology images from cancer patients generated at different sites {e.g., different hospitals, research laboratories, and / or the like).

[0067] The one or more digitized 3D pathology images 102 may be disposed within a training set 102t and a validation set 102v. The training set 102t comprises digitized 3D pathology images from a first plurality of cancer patients. The validation set 102v comprises digitized 3D pathology images from a second plurality of cancer patients. The training set 102t may be used to train one or more downstream machine learning models {e.g., mask generator 212, skeleton algorithm 216, machine learning model 116, etc.). For example, the training set 102t may be used to determine weighting values that are applied to different input variables provided to the one or more machine learning models. The validation set 102v may be used to validate the results of the machine learning models trained by the training set 102t.

[0068] A gland modeling tool 108 is configured to generate one or more 3D prostate gland models 110p using the one or more digitized 3D pathology images 102. A feature extraction tool 112 is configured to extract a plurality of glandular architecture features 114 from the one or more 3D prostate gland models 110p. A machine learning model 116 is configured to generate a medical prognosis 1 18 using the plurality of glandular architecture features 1 14.

[0069] In some embodiments, the machine learning model 116 may include a feature selection element 804 configured to select a set of most prognostic glandular architecture features to generate the medical prognosis 118. For example, the feature extraction tool 112 may extract a first number of glandular architecture features from the one or more 3D prostate gland models 110p. From the first number of glandular architecture features, the feature selection element 804 may select a smaller second number of the glandular architecture features that are most prognostic (e.g., that have a most significant impact in determining a likelihood of BCR). In some embodiments, the second number of glandular architecture features may be used to train and validate the machine learning model 116. In some embodiments, the selected features may characterize a torsion, a curvature, a branch length, and a branch density of a prostate gland. In some embodiments, the selected features may comprise one or more of junction distances, a length of the principal branch, and statistical measurements of gland torsion.

[0070] In some embodiments, the machine learning model 116 may comprise a K-nearest neighbor classifier and the feature selection element 804 may comprise a Fisher selector configured to generate a Fisher score (e.g., a Fisher’s discriminant ratio) for different glandular architecture features. Glandular architecture features having the highest Fisher scores are selected as being the most prognostic. In other embodiments, the machine learning model 116 may comprise a Cox regression model and the feature selection element 804 may comprise a LASSO operator.

[0071] Fig. 9 illustrates some embodiments of a block diagram of an apparatus 900 configured to generate a medical prognosis using glandular features extracted from a glandular model formed using one or more digitized 3D pathology images.

[0072] The apparatus 900 comprises an assessment apparatus 902. The assessment apparatus 902 is coupled to a 3D image generator 106, which is configured to generate one or more digitized 3D pathology images 102 of tissue samples (e.g., prostate tissue samples) collected from a cancer patient 104 that has or that has had cancer (e.g., prostate cancer). In some embodiments, the 3D image generator 106 may comprise an open top light sample (OTLS) microscope.

[0073] The assessment apparatus 902 comprises a processor 906 and a memory 904. The processor 906 can, in various embodiments, comprise circuitry such as,but not limited to, one or more single-core or multi-core processors. The processor 906 can include any combination of general-purpose processors and dedicated processors (e.g., graphics processors, application processors, etc.). The processor(s) 906 can be coupled with and / or can comprise memory (e.g., memory 904) or storage and can be configured to execute instructions stored in the memory 904 or storage to enable various apparatus, applications, or operating systems to perform operations and / or methods discussed herein.

[0074] The memory 904 can be configured to store the one or more digitized 3D pathology images 102. The one or more digitized 3D pathology images 102 may comprise digitized biopsy images having a plurality of pixels, each pixel having an associated intensity. In some additional embodiments, the one or more digitized 3D pathology images 102 may be stored in the memory 904 as one or more training sets for training a classifier and / or one or more test sets (e.g., validation sets).

[0075] The assessment apparatus 902 also comprises an input / output (I / O) interface 908 (e.g., associated with one or more I / O devices), a display 910, and an interface 912 that connects the processor 906, the memory 904, and the I / O interface 908. The I / O interface 912 can be configured to transfer data between the memory 904, the processor 906, and external devices, for example, the 3D image generator 106.

[0076] In some embodiments, the assessment apparatus 902 may further comprise one or more circuits 914 that include one or more of an immunofluorescence generator circuit 916, a mask generator circuit 918, a model generator circuit 920, a feature extraction circuit 922, and a machine learning circuit 924. In some embodiments, the one or more circuits 914 may operate according to computer code (e.g., machine learning algorithms) stored in the memory 904.

[0077] In some embodiments, the immunofluorescence generator circuit 916 is configured to convert the one or more digitized 3D pathology images 102 into an immunofluorescence dataset 210. The mask generator circuit 918 may be configured to operate upon the immunofluorescence dataset 210 to generate a glandular structure mask 214 that identifies one or more prostate glands within the immunofluorescence dataset 210. In other embodiments, the mask generator circuit 918 may be configured to run a deep learning algorithm on the one or more digitized 3D pathology images 102 to generate the glandular structure mask 214. In someembodiments, the model generator circuit 920 is configured to operate upon the glandular structure mask 214 to generate the one or more 3D gland models 110 (e.g. 3D prostate gland skeleton models). In some embodiments, the feature extraction circuit 922 is configured to extract a plurality of glandular architecture features 114 from the one or more 3D gland models 110. In some embodiments, the machine learning circuit 924 is configured to utilize the plurality of glandular architecture features 114 to generate a medical prognosis 118 within the cancer patient 104. In some embodiments, the display 910 is configured to output or display the medical prognosis 118 generated by the assessment apparatus 902.Example use case 1 :

[0078] Introduction: Prostate cancer is the most common cancer in American men and is mainly diagnosed from two-dimensional histology sections. Three- dimensional (3D) prostate histopathology is a valuable strategy to enhance the understanding of the disease by increasing microscopic sampling of specimens. This strategy enhances the examination of specimens by facilitating the analysis of the volumetric shape of cells and glands, as well as the pathway structures through entire biopsies and other morphological features. These aspects hold the potential to be linked with biochemical recurrence (BCR) outcomes. This study highlights the potential of features derived from the 3D structure of glands in samples of patients with prostate cancer for identifying patients who are at higher risk of BCR.

[0079] Methods: Radical prostatectomy specimens were collected from 50 patients along with 5-year postoperative BCR follow-up from the University of Washington. A total of 118 ex vivo whole-biopsy 3D pathology images were obtained from RP by an open-top light-sheet microscopy platform. Images were converted into a synthetic CK8 immunofluorescence dataset using an image-sequence translation model. From these 3D images, a glandular structure mask was generated by a computer-vision algorithm. The 3D structure of prostate glands was modeled through a Tree-structure Extraction Algorithm delivering Skeletons that are Accurate and Robust (TEASER), which provides a simplification of the length of the gland, its shape, and its pathway. 91 features were then extracted from the glandular skeleton architecture. A 3-fold cross-validation approach was used in which patients were divided into training (D1 ) and test (D2) sets of 33 and 17 patients, respectively.Fisher’s score, used to select features based on their correlation to the outcome, was applied to D1 to identify the top 5 features most correlated with BCR. We then utilized these features to train a K-Nearest Neighbors classifier on D1 , aimed at predicting whether a patient would experience BCR. The model’s performance was evaluated on D2 and averaged across all folds.

[0080] Results: The classifier was able to accurately differentiate between BCR+ and BCR- patients, achieving an average ROCAUC of 0.73 (±0.04) in D2. Significant differences in 3D skeletal features were observed between BCR+ and BCR- groups (p<0.05). Among the five selected features, three were related to the length and diameter of the gland, with the longest glands observed in BCR- patients. The remaining two features pertained to gland curvature and tortuosity, revealing more twisted glands in BCR+ patients.

[0081] Conclusion: Features derived from a skeleton model of the 3D glandular structure of biopsies showed promise in identifying patients with prostate cancer who are at a higher risk of BCR. Additional, multi-site independent validation of these findings is warranted.Example use case 2:

[0082] Introduction: The architectural characteristics of prostate cancer (PC) form the basis for assignment of prognostic grade groups that inform treatment decisions. Traditional 2D cross-sections provide valuable information but may not capture the full complexity of glandular morphology. Open-Top Light-Sheet (OTLS) microscopy, a non-destructive, slide-free imaging technique, allows for detailed examination of tissues in 3D. Using 3D OTLS images, we captured intricate glandular structures and demonstrated that novel 3D features can predict biochemical recurrence (BCR) in post-radical prostatectomy (RP) patients. In this study, we evaluate the generalizability and prognostic power of this feature set for estimating time to BCR.

[0083] Methods: A total of 148 patients who underwent prostatectomy were included in this study. A total of 82 needle biopsies from 34 cases from D1 and 114 punches from 114 individual patients from D2 were nondestructively imaged at micron-scale resolution in 3D using OTLS microscopy. An Al-based automatic segmentation model was applied to obtain glandular and lumen structures (Fig.1 ).The 3D structure of prostate glands was modeled using the Tree-structure Extraction Algorithm delivering Skeletons that are Accurate and Robust (TEASER) algorithm, followed by the extraction of 111 features capturing branch skeleton length, curvature, tortuosity, and angles between branches. We developed two samplespecific models to predict BCR and estimate time to recurrence using 3-fold cross- validation. We selected the most-relevant features for each case using Fisher's score. Cox models, Kaplan-Meier curves and log-rank tests were used for survival analysis.

[0084] Results: Our models effectively predicted the risk of BCR within 5 years after prostatectomy. The hazard ratios were 5.38 (95% confidence interval [Cl]: 1.64- 17.6, p=0.015) for D1 and 2.24 (95% Cl: 1 .19-4.22, p=0.023) for D2. The 3D glandular architecture-based classifier achieved an average area under the curve (AUC) of 0.73 (±0.04) for the D1 cohort and 0.69 (±0.04) for the D2 cohort. Our models are prognostic with different discriminative features: D2 emphasized angles between branches with 5 features whereas D1 prioritized skeleton length and curvature with 10 features.

[0085] Conclusion: Al-based glandular features on 3D histopathology of prostate cancer demonstrate prognostic value for predicting BCR in a multi-cohort setting. Further validation with larger datasets is warranted to identify a generalizable set of prognostic glandular features.

[0086] Therefore, the present disclosure relates to a method and apparatus configured to utilize glandular architecture features extracted from a glandular model formed using one or more digitized three-dimensional (3D) pathology images to make a medical prognosis.

[0087] In some embodiments, the present disclosure relates to a method including accessing one or more digitized three-dimensional (3D) pathology images of glandular tissue from a cancer patient; generating one or more 3D gland models using the one or more digitized 3D pathology images; extracting a plurality of glandular architecture features from the one or more 3D gland models; and providing the plurality of glandular architecture features to a machine learning model that has been trained to generate a medical prognosis for the cancer patient. In some embodiments, the plurality of glandular architecture features include a length, adiameter, a curvature, and a tortuosity of a prostate gland. In some embodiments, the plurality of glandular architecture features include one or more of a junction-to- junction mean distance, a junction-to-junction histogram, a mean distance of branches of a prostate gland, a mean value of curvature, and a mean value of tortuosity. In some embodiments, the one or more 3D gland models are skeleton models. In some embodiments, the method further includes generating the one or more digitized 3D pathology images using an open-top light-sheet microscopy platform. In some embodiments, the method further includes operating a deep learning algorithm on the one or more digitized 3D pathology images to form a glandular structure mask; and forming the one or more 3D gland models from the glandular structure mask. In some embodiments, the method further includes using a skeleton algorithm to form the one or more 3D gland models from the glandular structure mask. In some embodiments, the skeleton algorithm is a Tree-structure Extraction Algorithm delivering Skeletons that are Accurate and Robust (TEASER). In some embodiments, the method further includes using an image-sequence translation model to generate an immunofluorescence dataset from the one or more digitized 3D pathology images; using a computer vision algorithm to form a glandular structure mask from the immunofluorescence dataset; and using a skeleton algorithm to form the one or more 3D gland models from the glandular structure mask. In some embodiments, the glandular tissue includes prostate tissue obtained from the cancer patient during a radical prostatectomy.

[0088] 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 one or more digitized three-dimensional (3D) pathology images from a cancer patient having undergone a radical prostatectomy for treatment of prostate cancer; generating one or more 3D skeleton models of one or more prostate glands using the one or more digitized 3D pathology images; extracting a plurality of glandular architecture features from the one or more 3D skeleton models, the plurality of glandular architecture features including features relating to dimensions and a curvature of the one or more 3D skeleton models; and providing the plurality of glandular architecture features to a machine learning model configured to generate a medical prognosis regarding a risk of biochemical recurrence (BCR) for the cancer patient. In someembodiments, the plurality of glandular architecture features include features relating to a length, a diameter, the curvature, and a tortuosity of the one or more 3D skeleton models. In some embodiments, the operations further include operating a deep learning algorithm on the one or more digitized 3D pathology images to form a glandular structure mask; and forming the one or more 3D skeleton models from the glandular structure mask. In some embodiments, the operations further include extracting a first plurality of glandular architecture features from the one or more 3D skeleton models; and selecting a second plurality of glandular architecture features from the first plurality of glandular architecture features, the second plurality of glandular architecture features being less than the first plurality of glandular architecture features, the second plurality of glandular architecture features being features that are highly prognostic of BCR. In some embodiments, the second plurality of glandular architecture features are identified using a Fisher’s score. In some embodiments, the medical prognosis is determined to have a direct correlation between longer gland length and BCR free survival. In some embodiments, the medical prognosis is determined to have an indirect correlation between greater gland curvature and BCR free survival.

[0089] In yet other embodiments, the present disclosure relates to an apparatus including a memory configured to store one or more digitized three-dimensional (3D) pathology images of prostate tissue from a cancer patient having undergone a radical prostatectomy; one or more processors configured to generate one or more 3D prostate gland models using the one or more digitized 3D pathology images and generate a plurality of glandular architecture features from the one or more 3D prostate gland models; and a machine learning model configured to utilize the plurality of glandular architecture features to generate a medical prognosis regarding a risk of biochemical recurrence (BCR) for the cancer patient. In some embodiments, the plurality of glandular architecture features includes a junction-to- junction mean distance, a junction-to-junction histogram, a mean distance of branches of a prostate gland, a mean value of curvature, and a mean value of tortuosity. In some embodiments, the machine learning model is a K-nearest neighbor classifier.

[0090] Examples herein can include subject matter such as an apparatus, including a digital whole slide scanner, a CT system, an MRI system, a personalizedmedicine 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.

[0091] 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.

[0092] “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.

[0093] “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 multiplelogical 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.

[0094] 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.

[0095] 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.

[0096] 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).

[0097] 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 one or more digitized three-dimensional (3D) pathology images of glandular tissue from a cancer patient; generating one or more 3D gland models using the one or more digitized 3D pathology images; extracting a plurality of glandular architecture features from the one or more 3D gland models; and providing the plurality of glandular architecture features to a machine learning model that has been trained to generate a medical prognosis for the cancer patient.

2. The method of claim 1 , wherein the plurality of glandular architecture features include a length, a diameter, a curvature, and a tortuosity of a prostate gland.

3. The method of claim 1 , wherein the plurality of glandular architecture features include one or more of a junction-to-junction mean distance, a junction-to- junction histogram, a mean distance of branches of a prostate gland, a mean value of curvature, and a mean value of tortuosity.

4. The method of claim 1 , wherein the one or more 3D gland models are skeleton models.

5. The method of claim 1 , further comprising: generating the one or more digitized 3D pathology images using an open-top light-sheet microscopy platform.

6. The method of claim 1 , further comprising: operating a deep learning algorithm on the one or more digitized 3D pathology images to form a glandular structure mask; and forming the one or more 3D gland models from the glandular structure mask.

7. The method of claim 6, further comprising: using a skeleton algorithm to form the one or more 3D gland models from the glandular structure mask.

8. The method of claim 7, wherein the skeleton algorithm is a Treestructure Extraction Algorithm delivering Skeletons that are Accurate and Robust (TEASER).

9. The method of claim 1 , further comprising: using an image-sequence translation model to generate an immunofluorescence dataset from the one or more digitized 3D pathology images; using a computer vision algorithm to form a glandular structure mask from the immunofluorescence dataset; and using a skeleton algorithm to form the one or more 3D gland models from the glandular structure mask.

10. The method of claim 1 , wherein the glandular tissue comprises prostate tissue obtained from the cancer patient during a radical prostatectomy.

11. A non-transitory computer-readable medium storing computerexecutable instructions that, when executed, cause a processor to perform operations, comprising: accessing one or more digitized three-dimensional (3D) pathology images from a cancer patient having undergone a radical prostatectomy for treatment of prostate cancer; generating one or more 3D skeleton models of one or more prostate glands using the one or more digitized 3D pathology images; extracting a plurality of glandular architecture features from the one or more 3D skeleton models, wherein the plurality of glandular architecture features include features relating to dimensions and a curvature of the one or more 3D skeleton models; andproviding the plurality of glandular architecture features to a machine learning model configured to generate a medical prognosis regarding a risk of biochemical recurrence (BCR) for the cancer patient.

12. The non-transitory computer-readable medium of claim 11 , wherein the plurality of glandular architecture features include features relating to a length, a diameter, the curvature, and a tortuosity of the one or more 3D skeleton models.

13. The non-transitory computer-readable medium of claim 11 , wherein the operations further comprise: operating a deep learning algorithm on the one or more digitized 3D pathology images to form a glandular structure mask; and forming the one or more 3D skeleton models from the glandular structure mask.

14. The non-transitory computer-readable medium of claim 11 , wherein the operations further comprise: extracting a first plurality of glandular architecture features from the one or more 3D skeleton models; and selecting a second plurality of glandular architecture features from the first plurality of glandular architecture features, the second plurality of glandular architecture features being less than the first plurality of glandular architecture features, wherein the second plurality of glandular architecture features are features that are highly prognostic of BCR.

15. The non-transitory computer-readable medium of claim 14, wherein the second plurality of glandular architecture features are identified using a Fisher’s score.

16. The non-transitory computer-readable medium of claim 11 , wherein the medical prognosis is determined to have a direct correlation between longer gland length and BCR free survival.

17. The non-transitory computer-readable medium of claim 11 , wherein the medical prognosis is determined to have an indirect correlation between greater gland curvature and BCR free survival.

18. An apparatus, comprising: a memory configured to store one or more digitized three-dimensional (3D) pathology images of prostate tissue from a cancer patient having undergone a radical prostatectomy; one or more processors configured to: generate one or more 3D prostate gland models using the one or more digitized 3D pathology images; generate a plurality of glandular architecture features from the one or more 3D prostate gland models; and a machine learning model configured to utilize the plurality of glandular architecture features to generate a medical prognosis regarding a risk of biochemical recurrence (BCR) for the cancer patient.

19. The apparatus of claim 18, wherein the plurality of glandular architecture features include a junction-to-junction mean distance, a junction-to- junction histogram, a mean distance of branches of a prostate gland, a mean value of curvature, and a mean value of tortuosity.

20. The apparatus of claim 19, wherein the machine learning model is a K- nearest neighbor classifier.

Images