Search

Background: Genomic testing such as OncotypeDx remains an important component of the treatmentdecision process for many breast cancer (BC) patients. Evidence from Sparano et al. JCO 2021;39:557-564 demonstrated the importance of combining clinical features such as tumor grade, size and age with the 21-gene recurrence score (i.e., RSClin). Given the challenges associated with reliability of BC grading as a prognostic feature, we sought to develop a broadly accessible AI-digital test (PDxBr) which included a BC AI-grade combined with clinical features (i.e., age, tumor size, stage and lymph node status) to predict recurrence risk in an Oncotype categorized cohort. Methods: We evaluated performance of the PDxBr test along with the AI-grade, clinical feature (i.e., age, size, tumor stage and LN status) and histology grade models in a subgroup analysis from a retrospective longitudinal clinical development validation study utilizing samples from breast cancer (BC) patients in the Mount Sinai Health Care System (NYC, NY) from 2004-2016. Eligible participants were ≥23 years old with infiltrating ductal or mixed ductal and lobular carcinoma of the breast (IDC) and all with an Oncotype RS, and a median 6-year follow-up. All participants had H&E slides or paraffin blocks (for slide generation) from the resected BC specimen. H&E slides were digitized (40X magnification) using a Philips UltraFast Digital slide scanner (Netherlands) and a single whole slide image (WSI) was selected for model development. The AUC/C-index was used to demonstrate performance. Results: 599 patients with Oncotype RS results were interrogated: 57% white, mean age 57, mean tumor size 1.3cm, 100% T1/2 and HR+ve, 55% grade 2, 26% grade 3 and 18% grade 1; 95% pN0 with 36 events (6%). 21 (60%) of events were local-regional recurrences. Of note, there were 55% histologic Grade 2 cases in this population. Combining Oncotype RS with assorted sub-models including histologic grade, clinical features, AI-grade, or PDxBr model in a SVRc analysis demonstrated incremental improvement in the C-index for predicting BC recurrence (Table 1). Conclusionss: Both PDxBr test and AI-grade when combined with Oncotype were superior to Oncotype alone, or Oncotype with grade or clinical features suggesting that the incorporation of an improved BC grade with Oncotype RS enhances overall risk discrimination. PDxBr is the first digital BC test combining automated AI-BC prognostic grade with clinical-pathologic features to predict risk of early-stage BC recurrence. Additional validation studies are underway to confirm these results. AUC comparison of Oncotype alone and then combined with histology grade, clinical data (age, stage, tumor size and LN pos), AI-grade, and the PDxBr model. Citation Format: Gerardo Fernandez, Marcel Prastawa, Richard Scott, Bahram Marami, Nina Shpalensky, Abishek Madduri, Krystal Cascetta, Mary Sawyer, Monica S. Chan, Giovanni Koll, Rebecca E. DeAngel, Alexander Shtabsky, Aaron Feliz, Thomas Hansen, Brandon Veremis, Carlos Cordon-Cardo, Jack Zeineh, Michael Donovan. A novel AI-digital Test with an automated approach for grading and phenotyping breast cancer enriches recurrence score risk prediction in an Oncotype evaluated cohort. [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P3-05-21.

Background: Clinical practice guidelines emphasize the critical importance of grading and stage in breast cancer treatment. Although histologic grade is subjective, non-quantitative, skill-dependent, and oftentimes inaccurate it remains an independent prognostic feature and therefore plays a direct role in patient management including neoadjuvant therapy vs surgery, and interpretation of genomic studies. We developed an AI-based platform which combines digital H&E features with select clinical variables to assess risk of breast cancer recurrence and evaluated ability to predict Oncotype low risk RS categorization. Methods: Retrospective study to identify a subset of Mt. Sinai, NY invasive ductal breast cancer (IDC) patients from 2010-2016 with H&E stained slides, clinical features and OncotypeDx recurrence scores (RS). Recurrence endpoint(s): local- regional, distant-recurrence free and overall survival. Digital images generated with Philips scanning system; reviewed by two pathologist for tumor content and quality prior to image analysis and feature extraction. Support vector machine learning models were used for initial feature performance and final models generated. Positive predictive value (PPV), sensitivity (S) and likelihood ratios (LR) were used for performance. Results: 391 patients: mean age 57 years, 100% Stage I/II, 59% Grade 2, and 6% LN+ve 0-3; 97% IDC, 100% ER+ve, 94% PR+ve, 0% Her2 amplified; median follow-up 61 months; 323 (83%) low risk (25). There were 23 events (6%) and 13 (56%) were locoregional recurrence. PreciseDx model with age, and PR levels combined with imaging features reflective of mitotic activity and nuclear characteristics (clinical grade not selected) correctly identified LR RS categorization: PPV of 91%, [95CI 0.87-0.94], Se 92% [0.86-0.95] and positive / negative likelihood ratio : 6.8, and 0.5, respectively. Conclusions: Application of an AI-digital breast cancer risk assessment platform using only the H&E image and limited clinical data successfully classified low risk RS patients with high accuracy. Future models will extend outcome to 10 years and evaluate treatment selection and duration Citation Format: Michael Joseph Donovan, Nina Shpalensky, Marcel Prastawa, SM Abishek, Richard Scott, Mary Sawyer, Jack Zeineh, Gerardo Fernandez. An AI-digital breast cancer risk discrimination platform (PreciseDx) using a representative H&E image and selected clinical variables accurately categorized patients with oncotype Dx low risk recurrence scores (RS) [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS6-09.

Background: The 2019 NCCN guidelines and the College of American Pathology (CAP) endorse consistent, unambiguous comprehensive pathology reporting for invasive breast cancer. Challenges surrounding inter-pathologist variability and the lack of quantitative, standardized approaches to histologic grade are significant and critical to patient management. We developed an automated multi-network machine learning platform for histologic grading and examined performance with clinical outcome. Methods: Using the Cancer Genome Atlas (TCGA) breast cancer (BCA) image dataset and clinical data as a training cohort, we evaluated 420 conventional H&E images (one slide per patient). An artificial intelligent (AI) / deep-learning based workflow is used to normalize staining differences, identify regions of tumor vs. normal and characterize individual cellular attributes of breast cancer grading systems including overall gland structure, nuclear morphology, and mitotic figures. Individual features were correlated with overall survival using the concordance index (c-index). Support vector models and Kaplan-Meier incidence curves were used to further understand feature importance. Results: Using the AI platform, 88 image features representing BCA gland morphology and cellular-nuclear attributes were generated from 420 patients with 36 events including metastasis and or death with a median overall survival (OS) of 5-years. Three clinical variables were available including AJCC stage, grade and age at diagnosis. Image features were prioritized on C-index 0.5 which reflects increased or reduced risk of poor outcome, respectively. A training model consisting of image features + clinical data produced a C-index of 0.85, Hazards ratio 11, p Conclusions: Our innovative lab-based AI / deep-learning platform produced accurate BCA risk models to predict metastasis and OS through automated H&E BCA grading. Future analyses include a multi-site validation study to confirm these initial training results. Citation Format: Marcel Prastawa, Abishek Sainath Madduri, Brandon Veremis, Alexander Shtabsky, Bahram Marami, Jack Zeineh, Michael Joseph Donovan, Gerardo Fernandez. The application of machine learning techniques to standardize breast cancer grading and develop multivariate risk outcome models [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P3-08-11.

The diagnosis of Parkinson’s disease (PD) is challenging at all stages due to variable symptomatology, comorbidities, and mimicking conditions. Postmortem assessment remains the gold standard for a definitive diagnosis. While it is well recognized that PD manifests pathologically in the central nervous system with aggregation of α-synuclein as Lewy bodies and neurites, similar Lewy-type synucleinopathy (LTS) is additionally found in the peripheral nervous system that may be useful as an antemortem biomarker. We have previously found that detection of LTS in submandibular gland (SMG) biopsies is sensitive and specific for advanced PD; however, the sensitivity is suboptimal especially for early-stage disease. Further, visual microscopic assessment of biopsies by a neuropathologist to identify LTS is impractical for large-scale adoption. Here, we trained and validated a convolutional neural network (CNN) for detection of LTS on 283 digital whole slide images (WSI) from 95 unique SMG biopsies. A total of 8,450 LTS and 35,066 background objects were annotated following an inter-rater reliability study with Fleiss Kappa = 0.72. We used transfer learning to train a CNN model to classify image patches (151 × 151 pixels at 20× magnification) with and without the presence of LTS objects. The trained CNN model showed the following performance on image patches: sensitivity: 0.99, specificity: 0.99, precision: 0.81, accuracy: 0.99, and F-1 score: 0.89. We further tested the trained network on 1230 naïve WSI from the same cohort of research subjects comprising 42 PD patients and 14 controls. Logistic regression models trained on features engineered from the CNN predictions on the WSI resulted in sensitivity: 0.71, specificity: 0.65, precision: 0.86, accuracy: 0.69, and F-1 score: 0.76 in predicting clinical PD status, and 0.64 accuracy in predicting PD stage, outperforming expert neuropathologist LTS density scoring in terms of sensitivity but not specificity. These findings demonstrate the practical utility of a CNN detector in screening for LTS, which can translate into a computational tool to facilitate the antemortem tissue-based diagnosis of PD in clinical settings.

Breast cancer is the most common invasive cancer in women, affecting more than 10% of women worldwide. Microscopic analysis of a biopsy remains one of the most important methods to diagnose the type of breast cancer. This requires specialized analysis by pathologists, in a task that i) is highly time- and cost-consuming and ii) often leads to nonconsensual results. The relevance and potential of automatic classification algorithms using hematoxylin-eosin stained histopathological images has already been demonstrated, but the reported results are still sub-optimal for clinical use. With the goal of advancing the state-of-the-art in automatic classification, the Grand Challenge on BreAst Cancer Histology images (BACH) was organized in conjunction with the 15th International Conference on Image Analysis and Recognition (ICIAR 2018). A large annotated dataset, composed of both microscopy and whole-slide images, was specifically compiled and made publicly available for the BACH challenge. Following a positive response from the scientific community, a total of 64 submissions, out of 677 registrations, effectively entered the competition. From the submitted algorithms it was possible to push forward the state-of-the-art in terms of accuracy (87%) in automatic classification of breast cancer with histopathological images. Convolutional neuronal networks were the most successful methodology in the BACH challenge. Detailed analysis of the collective results allowed the identification of remaining challenges in the field and recommendations for future developments. The BACH dataset remains publically available as to promote further improvements to the field of automatic classification in digital pathology., Accepted for publication at Medical Image Analysis (Elsevier). Publication licensed under the Creative Commons CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/

Surgical pathology and the practice of general medicine are often described as having legitimate components of art as well as science. That perception is perhaps less relevant in the current information age with the adoption of genomic medicine, in large part due to the growth in disease-specific causal knowledge. As a surgical pathologist, gestalt and intuition are words you may not use to describe how you arrive at a specific patient's diagnosis, but it is certainly understood that with experience, in particular with a deep immersion in a subspecialty, comes an understanding of histopathology that cannot easily be transmitted or formalized. This is not in any way to delegitimize pathology or relegate it to the unscientific, rather to illustrate a strength of the human brain. For well over 100 years using a microscope, pathologists have been applying this strength to group and separate disease entities into diagnoses and their subtypes. The brain seems to be good at pattern recognition, filtering noise, and dealing with variability, and from a histopathology perspective, this has complemented the medical need quite well for over a century. Although the concept of tumor differentiation and its relationship to clinical outcome dates back to the early days of histopathology until fairly recently, limited treatment options diminished the necessity for more precise and standardized grading. The following chapter explores how the growing fields of computational pathology, machine learning, and deep learning are coming together to address the need for more precise and standardized grading. Starting with tissue preparation and image acquisition, the discussion roughly follows the steps involved in developing an image analysis–based multivariate grading system with attention drawn in each subsection to inherent obstacles, published state-of-the-art techniques, and the authors' own experience with solutions.

Pathology Artificial Intelligence Platform (PAIP) is a free research platform in support of pathological artificial intelligence (AI). The main goal of the platform is to construct a high-quality pathology learning data set that will allow greater accessibility. The PAIP Liver Cancer Segmentation Challenge, organized in conjunction with the Medical Image Computing and Computer Assisted Intervention Society (MICCAI 2019), is the first image analysis challenge to apply PAIP datasets. The goal of the challenge was to evaluate new and existing algorithms for automated detection of liver cancer in whole-slide images (WSIs). Additionally, the PAIP of this year attempted to address potential future problems of AI applicability in clinical settings. In the challenge, participants were asked to use analytical data and statistical metrics to evaluate the performance of automated algorithms in two different tasks. The participants were given the two different tasks: Task 1 involved investigating Liver Cancer Segmentation and Task 2 involved investigating Viable Tumor Burden Estimation. There was a strong correlation between high performance of teams on both tasks, in which teams that performed well on Task 1 also performed well on Task 2. After evaluation, we summarized the top 11 team's algorithms. We then gave pathological implications on the easily predicted images for cancer segmentation and the challenging images for viable tumor burden estimation. Out of the 231 participants of the PAIP challenge datasets, a total of 64 were submitted from 28 team participants. The submitted algorithms predicted the automatic segmentation on the liver cancer with WSIs to an accuracy of a score estimation of 0.78. The PAIP challenge was created in an effort to combat the lack of research that has been done to address Liver cancer using digital pathology. It remains unclear of how the applicability of AI algorithms created during the challenge can affect clinical diagnoses. However, the results of this dataset and evaluation metric provided has the potential to aid the development and benchmarking of cancer diagnosis and segmentation.

Accumulation of abnormal tau in neurofibrillary tangles (NFT) occurs in Alzheimer disease (AD) and a spectrum of tauopathies. These tauopathies have diverse and overlapping morphological phenotypes that obscure classification and quantitative assessments. Recently, powerful machine learning-based approaches have emerged, allowing the recognition and quantification of pathological changes from digital images. Here, we applied deep learning to the neuropathological assessment of NFT in postmortem human brain tissue to develop a classifier capable of recognizing and quantifying tau burden. The histopathological material was derived from 22 autopsy brains from patients with tauopathies. We used a custom web-based informatics platform integrated with an in-house information management system to manage whole slide images (WSI) and human expert annotations as ground truth. We utilized fully annotated regions to train a deep learning fully convolutional neural network (FCN) implemented in PyTorch against the human expert annotations. We found that the deep learning framework is capable of identifying and quantifying NFT with a range of staining intensities and diverse morphologies. With our FCN model, we achieved high precision and recall in naive WSI semantic segmentation, correctly identifying tangle objects using a SegNet model trained for 200 epochs. Our FCN is efficient and well suited for the practical application of WSIs with average processing times of 45 min per WSI per GPU, enabling reliable and reproducible large-scale detection of tangles. We measured performance on test data of 50 pre-annotated regions on eight naive WSI across various tauopathies, resulting in the recall, precision, and an F1 score of 0.92, 0.72, and 0.81, respectively. Machine learning is a useful tool for complex pathological assessment of AD and other tauopathies. Using deep learning classifiers, we have the potential to integrate cell- and region-specific annotations with clinical, genetic, and molecular data, providing unbiased data for clinicopathological correlations that will enhance our knowledge of the neurodegeneration.

Postoperative risk assessment remains an important variable in the effective treatment of prostate cancer. There is an unmet clinical need for a test with the potential to enhance the Gleason grading system with novel features that more accurately reflect a personalized prediction of clinical failure. A prospectively designed retrospective study utilizing 892 patients, post radical prostatectomy, followed for a median of 8 years. In training, using digital image analysis to combine microscopic pattern analysis/machine learning with biomarkers, we evaluated Precise Post-op model results to predict clinical failure in 446 patients. The derived prognostic score was validated in 446 patients. Eligible subjects required complete clinical-pathologic variables and were excluded if they had received neoadjuvant treatment including androgen deprivation, radiation or chemotherapy prior to surgery. No patients were enrolled with metastatic disease prior to surgery. Evaluate the assay using time to event concordance index (C-index), Kaplan–Meier, and hazards ratio. In the training cohort (n = 306), the Precise Post-op test predicted significant clinical failure with a C-index of 0.82, [95% CI: 0.76–0.86], HR:6.7, [95% CI: 3.59–12.45], p

Accurate analysis of tissue structures in breast cancer histopathology slides is crucial for staging treatments and predicting outcome. Such analysis depends on identification of tissue architecture in different regions, and determining the different types of cancer morphology which includes in-situ carcinoma, invasive tumor, and benign tumor. We propose an automated classification method for identifying these micro-architectures using an ensemble of convolutional neural networks. This ensemble is constructed by combining multiple networks, trained using different data subset sampling and image perturbation models. Our proposed approach results in a high performing detector with robustness to data variations.

The Gleason score is the most common architectural and morphological assessment of prostate cancer severity and prognosis. There have been numerous quantitative techniques developed to approximate and duplicate the Gleason scoring system. Most of these approaches have been developed in standard H and E brightfield microscopy. Immunofluorescence (IF) image analysis of tissue pathology has recently been proven to be extremely valuable and robust in developing prognostic assessments of disease, particularly in prostate cancer. There have been significant advances in the literature in quantitative biomarker expression as well as characterization of glandular architectures in discrete gland rings. In this work we leverage a new method of segmenting gland rings in IF images for predicting the pathological Gleason; both the clinical and the image specific grade, which may not necessarily be the same. We combine these measures with nuclear specific characteristics as assessed by the MST algorithm. Our individual features correlate well univariately with the Gleason grades, and in a multivariate setting have an accuracy of 85% in predicting the Gleason grade. Additionally, these features correlate strongly with clinical progression outcomes (CI of 0.89), significantly outperforming the clinical Gleason grades (CI of 0.78). This work presents the first assessment of morphological gland unit features from IF images for predicting the Gleason grade.

Background: Postoperative risk assessment remains an important variable in the treatment of prostate cancer. We aimed to develop and validate a tissue-based on-slide prognostic risk model test to identify which patients would benefit most from early intervention. Methods: Prospectively designed retrospective study utilizing 892 patients, post radical prostatectomy, followed for a median of 8 years. We constructed a matched 50:50 training and test cohort to identify compound features using protein multiplex immunofluorescence combined with novel morphometry generated by advanced image analysis and machine learning tools. The primary endpoint was the development of a postoperative clinical failure (CF) including PSA rise post-adjuvant treatment, or metastasis or prostate cancer specific mortality. Time to event C-index and Kaplan-Meier were used to assess accuracy. Results: The Precise postop model in training (n=306) selected 4 compound imaging features and one clinical variable to predict CF with a performance C-index of 0.82, [95%CI: 0.76-0.86], HR:6.7, [95%CI: 3.59-12.45], p Conclusion: Patients with high Precise postop scores had a higher likelihood of having disease progression within 8 years with a reduced durable response to androgen-deprivation therapy. The Precise postop algorithm-based model guided by machine learning replaces subjective Gleason grading with novel image features that combine morphometry with biologic attributes that more accurately reflect disease potential. Citation Format: Michael Joseph Donovan, Gerardo Fernandez, Richard Scott, Jack Zeineh, Giovanni Koll, Faisal Khan, Nataliya Gladoun, Elizabeth Charytonowicz, Ash Tewari, Carlos Cordon-Cardo. Development and validation of a novel automated Gleason grade and molecular profile that define a highly predictive prostate cancer progression algorithm-based test [abstract]. In: Proceedings of the AACR Special Conference: Prostate Cancer: Advances in Basic, Translational, and Clinical Research; 2017 Dec 2-5; Orlando, Florida. Philadelphia (PA): AACR; Cancer Res 2018;78(16 Suppl):Abstract nr B093.

170 Background: Postoperative risk assessment remains an important variable in the treatment of prostate cancer (PCA). Advances in genomic risk classifiers have aided clinical-decision making; however, clinical-pathologic variables such as Gleason grade and pathologic stage remain significant comparators for accurate prognostication. We aimed to standardize the descriptive pathology of PCA through automation of Gleason grading with artificial intelligence and image analysis feature selection Methods: Retrospective study using radical prostatectomy (RP) tissue microarrays from Henry Ford Hospital and Roswell Park Cancer Center with 8-year median follow-up. Samples were stained with a multiplex immunofluorescent assay: Androgen Receptor (AR), Ki67, Cytokeratin 18, Cytokeratin 5/6 and Alpha-methylacyl-CoA racemase); imaged with a CRI Nuance FX camera and then analyzed with proprietary software to generate a suite of morphometric - attributes that quantitatively characterize the Gleason spectrum. Derived features were univariately correlated with disease progression using the concordance index (CI) along with the hazards ratio and p-value. Results: Starting with a training cohort of 306 patients and a 15% event rate, MIF PCA images were subjected to a machine learning analysis program which incorporates a graph theory-based approach for characterization of gland / ring fusion and fragmentation of tumor architecture (TA) with biomarker quantitation (BQ) (i.e. AR and Ki67). 19 unique image features with 7 TA and 12 TA+BQ were identified, by univariate CI, all TA features were strongly associated with Gleason grading with CI’s reflecting degree of tumor differentiation (CI 0.29-.33, p-value = 0.005). Four TA+BQ features were selected in a training risk model and effectively replaced the clinical Gleason features. By comparison, dominant RP Gleason had a CI of 0.31. Conclusions: Image-based feature selection guided by principles of machine learning has the potential to automate and replace traditional Gleason grading. Such approaches provide the necessary foundation for next generation risk assessment assays.

Immunofluorescent (IF) image analysis of tissue pathology has proven to be extremely valuable and robust in developing prognostic assessments of disease, particularly in prostate cancer. There have been significant advances in the literature in quantitative biomarker expression as well as characterization of glandular architectures in discrete gland rings. However, while biomarker and glandular morphometric features have been combined as separate predictors in multivariate models, there is a lack of integrative features for biomarkers co-localized within specific morphological sub-types; for example the evaluation of androgen receptor (AR) expression within Gleason 3 glands only. In this work we propose a novel framework employing multiple techniques to generate integrated metrics of morphology and biomarker expression. We demonstrate the utility of the approaches in predicting clinical disease progression in images from 326 prostate biopsies and 373 prostatectomies. Our proposed integrative approaches yield significant improvements over existing IF image feature metrics. This work presents some of the first algorithms for generating innovative characteristics in tissue diagnostics that integrate co-localized morphometry and protein biomarker expression.

236 Background: Quantitative analysis of formalin-fixed paraffin-embedded (FFPE) prostate samples post radical-prostatectomy (RP) can help guide management of post-operative prostate cancer (PCa) patients. We have previously published on the results of our systems (precise) pathology platform which includes integration of clinical data with immunofluorescent (IF) morphology and biomarkers to predict disease progression in post-RP patients. We extended this approach by developing quantitative image features that further refine risk discrimination by directly competing with Gleason through the incorporation of biological attributes. Methods: Tissue microarrays representing a 892 RP population from Roswell Park Cancer Institute and the Henry Ford Hospital, with median 8 years follow-up were stained with a multiplex IF assay incorporating CK18 (epithelial cells), CK5/6 (basal cells), AMACR, the Androgen Receptor (AR), and Ki67. Images were captured with a Nikon i90 microscope / CRI multiplex camera and a Bayesian probabilistic approach was used to derive micro-anatomic quantitative image attributes. We generated a 50:50 balanced training / test cohort and optimized feature construction using clinical variables. Multivariate models were produced on the 324 training set to predict disease progression (i.e. clinical failure) which included a PSA rise post-adjuvant therapy, metastasis or death from PCa. The concordance index (CI), sensitivity and specificity were used to assess performance of the model. Results: A multivariate training model yielded a CI of 0.80, sensitivity 85% and specificity 74% with a hazard ratio of 9.8. By comparison a clinical only model produced a CI of 0.77, sensitivity 80% and specificity 73%, with a hazard ratio of 6. The IF model consisted of 3 features including pathologic stage and two imaging features that combine morphometry integrated with levels of AR and Ki67. Neither the Gleason score nor PSA were selected. Conclusions: Tissue-based models which incorporate objective and biological attributes accurately predict disease progression. Once validated, such approaches may replace the subjective Gleason grading system and provide an additional tool for pathologists when assigning risk.

Morphometric features characterizing the fusion and fragmentation of the glandular architecture of advanced prostate cancer have not previously been based upon the automated segmentation of discrete gland rings, due in part to the difficulty of extracting these structures from the HE stained tissues. We present a novel approach for segmenting gland rings in multi-spectral immunofluorescence (IF) images and demonstrate the utility of the resultant features in predicting cancer recurrence in a cohort of 1956 images of prostate biopsies and prostatectomies from 679 patients. The proposed approach is evaluated for prediction of actual clinical outcomes of interest to physicians in comparison with previously published gland-unit features, yielding a concordance index (CI) of 0.67. This compares favorably to the CI of 0.66 obtained using a semi-automated segmentation of the corresponding HE images from the same patients. This work presents the first algorithms for segmentation of gland rings lacking a central lumen, and for separation of touching epithelial units, and introduces new gland adjacency features for predicting prostate cancer clinical progression across both biopsy and prostatectomy images.

Morphometric features characterizing the fusion and fragmentation of the glandular architecture of advanced prostate cancer have not previously been based upon the automated segmentation of discrete gland rings, due in part to the difficulty of extracting these structures from the H&E stained tissues. We present a novel approach for segmenting gland rings in multi-spectral immunofluorescence (IF) images and demonstrate the utility of the resultant features in predicting cancer recurrence in a cohort of 1956 images of prostate biopsies and prostatectomies from 679 patients. The proposed approach is evaluated for prediction of actual clinical outcomes of interest to physicians in comparison with previously published gland-unit features, yielding a concordance index (CI) of 0.67. This compares favorably to the CI of 0.66 obtained using a semi-automated segmentation of the corresponding H&E images from the same patients. This work presents the first algorithms for segmentation of gland rings lacking a central lumen, and for separation of touching epithelial units, and introduces new gland adjacency features for predicting prostate cancer clinical progression across both biopsy and prostatectomy images.

47 Background: We have previously validated the importance of utilizing the intact tissue section and quantitative biomarker strategies for developing prognostic prostate cancer outcome models. We sought to further expand upon these constructs by applying next-gen advanced image analysis tools using multiplex immunofluorescence (IF) integrated with clinical variables on an independent cohort. Methods: Utilizing a 686 patient radical prostatectomy tissue microarray population with median 8 years follow-up from Roswell Park Cancer Institute, NY, we selected an initial subset of 98 patients to construct a training model; applying multiplex immunofluorescence with CK-18 (epithelial cells), CK 5/6 (basal cells), androgen receptor (AR), Ki67, AMACR and DAPI (nuclear marker). Images were captured with an Nikon i90 microscope equipped with a CRI multiplex camera, a Bayesian probabilistic approach was used to derive features that reflect tumor grade and co-locate - quantify IF biomarker expression within various glandular sub-types. Multivariate models were constructed to predict disease progression (i.e., clinical failure (CF) = PSA rise post-adjuvant therapy, or metastasis, or death) with c-index (CI) to estimate performance. Results: A 98 patient with the following clinical features: mean PSA 6.6 ng/mL, prostatectomy Gleason score, PGS 6, 34%; >/=PGS 7, 55%; LN+ 2%, +SM 19%; + ECE 49% ; + SVI 3%. CFevent rate 13%. Two models (i.e., +/- LN status) employing 47 features with 7 clinical variables (e.g., PSA and Gleason) had comparable outcomes, CI 0.84. The number one feature selected in both models was an imaging feature reflecting prostate cancer gland architecture and differentiation. Combined AR levels with gland phenotype and CK-18 composition characterized a relevant phenotype dependent upon feature inclusion in the model generated. Conclusions: We have established the importance of utilizing the intact tissue section when developing accurate prognostic prostate cancer models. Additional studies with expanded training and test patient cohorts are underway to confirm and validate these initial findings.

Accurate segmentation of overlapping nuclei is essential in determining nuclei count and evaluating the sub-cellular localization of protein biomarkers in image Cytometry and Histology. Current cellular segmentation algorithms generally lack fast and reliable methods for disambiguating clumped nuclei. In immuno-fluorescence segmentation, solutions to challenges including nuclei misclassification, irregular boundaries, and under-segmentation require reliable separation of clumped nuclei. This paper presents a fast and accurate algorithm for joint segmentation of cellular cytoplasm and nuclei incorporating procedures for reliably separating overlapping nuclei. The algorithm utilizes a combination of ideas and is a significant improvement on state-of-the-art algorithms for this application. First, an adaptive process that includes top-hat filtering, blob detection and distance transforms estimates the inverse illumination field and corrects for intensity non-uniformity. Minimum-error-thresholding based binarization augmented by statistical stability estimation is applied prior to seed-detection constrained by a distance-map-based scale-selection to identify candidate seeds for nuclei segmentation. The nuclei clustering step also incorporates error estimation based on statistical stability. This enables the algorithm to perform localized error correction. Final steps include artifact removal and reclassification of nuclei objects near the cytoplasm boundary as epithelial or stroma. Evaluation using 48 realistic phantom images with known ground-truth shows overall segmentation accuracy exceeding 96%. It significantly outperformed two state-of-the-art algorithms in clumped nuclei separation. Tests on 926 prostate biopsy images (326 patients) show that the segmentation improvement improves the predictive power of nuclei architecture features based on the minimum spanning tree algorithm. The algorithm has been deployed in a large scale pathology application.