AI-based histologic scoring enables automated and reproducible assessment of enrollment criteria and endpoints in NASH clinical trials.

Clinical trials in nonalcoholic steatohepatitis (NASH) require histologic scoring for assessment of inclusion criteria and endpoints. However, guidelines for scoring key features have led to variability in interpretation, impacting clinical trial outcomes. We developed an artificial intelligence (AI)-based measurement (AIM) tool for scoring NASH histology (AIM-NASH). AIM-NASH predictions for NASH Clinical Research Network (CRN) grades of necroinflammation and stages of fibrosis aligned with expert consensus scores and were reproducible. Continuous scores produced by AIM-NASH for key histological features of NASH correlated with mean pathologist scores and with noninvasive biomarkers and strongly predicted patient outcomes. In a retrospective analysis of the ATLAS trial, previously unmet pathological endpoints were met when scored by the AIM-NASH algorithm alone. Overall, these results suggest that AIM-NASH may assist pathologists in histologic review of NASH clinical trials, reducing inter-rater variability on trial outcomes and offering a more sensitive and reproducible measure of patient therapeutic response.

Integration of deep learning-based histopathology and transcriptomics reveals key genes associated with fibrogenesis in patients with advanced NASH.

Nonalcoholic steatohepatitis (NASH) is the most common chronic liver disease globally and a leading cause for liver transplantation in the US. Its pathogenesis remains imprecisely defined. We combined two high-resolution modalities to tissue samples from NASH clinical trials, machine learning (ML)-based quantification of histological features and transcriptomics, to identify genes that are associated with disease progression and clinical events. A histopathology-driven 5-gene expression signature predicted disease progression and clinical events in patients with NASH with F3 (pre-cirrhotic) and F4 (cirrhotic) fibrosis. Notably, the Notch signaling pathway and genes implicated in liver-related diseases were enriched in this expression signature. In a validation cohort where pharmacologic intervention improved disease histology, multiple Notch signaling components were suppressed.

Artificial Intelligence Enables Quantitative Assessment of Ulcerative Colitis Histology.

Ulcerative colitis is a chronic inflammatory bowel disease that is characterized by a relapsing and remitting course. Assessment of disease activity critically informs treatment decisions. In addition to endoscopic remission, histologic remission is emerging as a treatment target and a key factor in the evaluation of disease activity and therapeutic efficacy. However, manual pathologist evaluation is semiquantitative and limited in granularity. Machine learning approaches are increasingly being developed to aid pathologists in accurate and reproducible scoring of histology, enabling precise quantitation of clinically relevant features. Here, we report the development and validation of convolutional neural network models that quantify histologic features pertinent to ulcerative colitis disease activity, directly from hematoxylin and eosin-stained whole slide images. Tissue and cell model predictions were used to generate quantitative human-interpretable features to fully characterize the histology samples. Tissue and cell predictions showed comparable agreement to pathologist annotations, and the extracted slide-level human-interpretable features demonstrated strong correlations with disease severity and pathologist-assigned Nancy histological index scores. Moreover, using a random forest classifier based on 13 human-interpretable features derived from the tissue and cell models, we were able to accurately predict Nancy histological index scores, with a weighted kappa (κ = 0.91) and Spearman correlation (⍴ = 0.89, P < .001) when compared with pathologist consensus Nancy histological index scores. We were also able to predict histologic remission, based on the absence of neutrophil extravasation, with a high accuracy of 0.97. This work demonstrates the potential of computer vision to enable a standardized and robust assessment of ulcerative colitis histopathology for translational research and improved evaluation of disease activity and prognosis.

A Machine Learning Approach to Liver Histological Evaluation Predicts Clinically Significant Portal Hypertension in NASH Cirrhosis.

BACKGROUND AND AIMS: The hepatic venous pressure gradient (HVPG) is the standard for estimating portal pressure but requires expertise for interpretation. We hypothesized that HVPG could be extrapolated from liver histology using a machine learning (ML) algorithm. APPROACH AND RESULTS: Patients with NASH with compensated cirrhosis from a phase 2b trial were included. HVPG and biopsies from baseline and weeks 48 and 96 were reviewed centrally, and biopsies evaluated with a convolutional neural network (PathAI, Boston, MA). Using trichrome-stained biopsies in the training set (n = 130), an ML model was developed to recognize fibrosis patterns associated with HVPG, and the resultant ML HVPG score was validated in a held-out test set (n = 88). Associations between the ML HVPG score with measured HVPG and liver-related events, and performance of the ML HVPG score for clinically significant portal hypertension (CSPH) (HVPG ≥ 10 mm Hg), were determined. The ML-HVPG score was more strongly correlated with HVPG than hepatic collagen by morphometry (ρ = 0.47 vs. ρ = 0.28; P < 0.001). The ML HVPG score differentiated patients with normal (0-5 mm Hg) and elevated (5.5-9.5 mm Hg) HVPG and CSPH (median: 1.51 vs. 1.93 vs. 2.60; all P < 0.05). The areas under receiver operating characteristic curve (AUROCs) (95% CI) of the ML-HVPG score for CSPH were 0.85 (0.80, 0.90) and 0.76 (0.68, 0.85) in the training and test sets, respectively. Discrimination of the ML-HVPG score for CSPH improved with the addition of a ML parameter for nodularity, Enhanced Liver Fibrosis, platelets, aspartate aminotransferase (AST), and bilirubin (AUROC in test set: 0.85; 95% CI: 0.78, 0.92). Although baseline ML-HVPG score was not prognostic, changes were predictive of clinical events (HR: 2.13; 95% CI: 1.26, 3.59) and associated with hemodynamic response and fibrosis improvement. CONCLUSIONS: An ML model based on trichrome-stained liver biopsy slides can predict CSPH in patients with NASH with cirrhosis.

Human-interpretable image features derived from densely mapped cancer pathology slides predict diverse molecular phenotypes.

Computational methods have made substantial progress in improving the accuracy and throughput of pathology workflows for diagnostic, prognostic, and genomic prediction. Still, lack of interpretability remains a significant barrier to clinical integration. We present an approach for predicting clinically-relevant molecular phenotypes from whole-slide histopathology images using human-interpretable image features (HIFs). Our method leverages >1.6 million annotations from board-certified pathologists across >5700 samples to train deep learning models for cell and tissue classification that can exhaustively map whole-slide images at two and four micron-resolution. Cell- and tissue-type model outputs are combined into 607 HIFs that quantify specific and biologically-relevant characteristics across five cancer types. We demonstrate that these HIFs correlate with well-known markers of the tumor microenvironment and can predict diverse molecular signatures (AUROC 0.601-0.864), including expression of four immune checkpoint proteins and homologous recombination deficiency, with performance comparable to 'black-box' methods. Our HIF-based approach provides a comprehensive, quantitative, and interpretable window into the composition and spatial architecture of the tumor microenvironment.

A Machine Learning Approach Enables Quantitative Measurement of Liver Histology and Disease Monitoring in NASH.

BACKGROUND AND AIMS: Manual histological assessment is currently the accepted standard for diagnosing and monitoring disease progression in NASH, but is limited by variability in interpretation and insensitivity to change. Thus, there is a critical need for improved tools to assess liver pathology in order to risk stratify NASH patients and monitor treatment response. APPROACH AND RESULTS: Here, we describe a machine learning (ML)-based approach to liver histology assessment, which accurately characterizes disease severity and heterogeneity, and sensitively quantifies treatment response in NASH. We use samples from three randomized controlled trials to build and then validate deep convolutional neural networks to measure key histological features in NASH, including steatosis, inflammation, hepatocellular ballooning, and fibrosis. The ML-based predictions showed strong correlations with expert pathologists and were prognostic of progression to cirrhosis and liver-related clinical events. We developed a heterogeneity-sensitive metric of fibrosis response, the Deep Learning Treatment Assessment Liver Fibrosis score, which measured antifibrotic treatment effects that went undetected by manual pathological staging and was concordant with histological disease progression. CONCLUSIONS: Our ML method has shown reproducibility and sensitivity and was prognostic for disease progression, demonstrating the power of ML to advance our understanding of disease heterogeneity in NASH, risk stratify affected patients, and facilitate the development of therapies.

Combination Therapies Including Cilofexor and Firsocostat for Bridging Fibrosis and Cirrhosis Attributable to NASH.

BACKGROUND AND AIMS: Advanced fibrosis attributable to NASH is a leading cause of end-stage liver disease. APPROACH AND RESULTS: In this phase 2b trial, 392 patients with bridging fibrosis or compensated cirrhosis (F3-F4) were randomized to receive placebo, selonsertib 18 mg, cilofexor 30 mg, or firsocostat 20 mg, alone or in two-drug combinations, once-daily for 48 weeks. The primary endpoint was a ≥1-stage improvement in fibrosis without worsening of NASH between baseline and 48 weeks based on central pathologist review. Exploratory endpoints included changes in NAFLD Activity Score (NAS), liver histology assessed using a machine learning (ML) approach, liver biochemistry, and noninvasive markers. The majority had cirrhosis (56%) and NAS ≥5 (83%). The primary endpoint was achieved in 11% of placebo-treated patients versus cilofexor/firsocostat (21%; P = 0.17), cilofexor/selonsertib (19%; P = 0.26), firsocostat/selonsertib (15%; P = 0.62), firsocostat (12%; P = 0.94), and cilofexor (12%; P = 0.96). Changes in hepatic collagen by morphometry were not significant, but cilofexor/firsocostat led to a significant decrease in ML NASH CRN fibrosis score (P = 0.040) and a shift in biopsy area from F3-F4 to ≤F2 fibrosis patterns. Compared to placebo, significantly higher proportions of cilofexor/firsocostat patients had a ≥2-point NAS reduction; reductions in steatosis, lobular inflammation, and ballooning; and significant improvements in alanine aminotransferase (ALT), aspartate aminotransferase (AST), bilirubin, bile acids, cytokeratin-18, insulin, estimated glomerular filtration rate, ELF score, and liver stiffness by transient elastography (all P ≤ 0.05). Pruritus occurred in 20%-29% of cilofexor versus 15% of placebo-treated patients. CONCLUSIONS: In patients with bridging fibrosis and cirrhosis, 48 weeks of cilofexor/firsocostat was well tolerated, led to improvements in NASH activity, and may have an antifibrotic effect. This combination offers potential for fibrosis regression with longer-term therapy in patients with advanced fibrosis attributable to NASH.

Report on computational assessment of Tumor Infiltrating Lymphocytes from the International Immuno-Oncology Biomarker Working Group.

Assessment of tumor-infiltrating lymphocytes (TILs) is increasingly recognized as an integral part of the prognostic workflow in triple-negative (TNBC) and HER2-positive breast cancer, as well as many other solid tumors. This recognition has come about thanks to standardized visual reporting guidelines, which helped to reduce inter-reader variability. Now, there are ripe opportunities to employ computational methods that extract spatio-morphologic predictive features, enabling computer-aided diagnostics. We detail the benefits of computational TILs assessment, the readiness of TILs scoring for computational assessment, and outline considerations for overcoming key barriers to clinical translation in this arena. Specifically, we discuss: 1. ensuring computational workflows closely capture visual guidelines and standards; 2. challenges and thoughts standards for assessment of algorithms including training, preanalytical, analytical, and clinical validation; 3. perspectives on how to realize the potential of machine learning models and to overcome the perceptual and practical limits of visual scoring.

Pitfalls in assessing stromal tumor infiltrating lymphocytes (sTILs) in breast cancer.

Stromal tumor-infiltrating lymphocytes (sTILs) are important prognostic and predictive biomarkers in triple-negative (TNBC) and HER2-positive breast cancer. Incorporating sTILs into clinical practice necessitates reproducible assessment. Previously developed standardized scoring guidelines have been widely embraced by the clinical and research communities. We evaluated sources of variability in sTIL assessment by pathologists in three previous sTIL ring studies. We identify common challenges and evaluate impact of discrepancies on outcome estimates in early TNBC using a newly-developed prognostic tool. Discordant sTIL assessment is driven by heterogeneity in lymphocyte distribution. Additional factors include: technical slide-related issues; scoring outside the tumor boundary; tumors with minimal assessable stroma; including lymphocytes associated with other structures; and including other inflammatory cells. Small variations in sTIL assessment modestly alter risk estimation in early TNBC but have the potential to affect treatment selection if cutpoints are employed. Scoring and averaging multiple areas, as well as use of reference images, improve consistency of sTIL evaluation. Moreover, to assist in avoiding the pitfalls identified in this analysis, we developed an educational resource available at www.tilsinbreastcancer.org/pitfalls.

The Combined Effect of FGFR Inhibition and PD-1 Blockade Promotes Tumor-Intrinsic Induction of Antitumor Immunity.

The success of targeted or immune therapies is often hampered by the emergence of resistance and/or clinical benefit in only a subset of patients. We hypothesized that combining targeted therapy with immune modulation would show enhanced antitumor responses. Here, we explored the combination potential of erdafitinib, a fibroblast growth factor receptor (FGFR) inhibitor under clinical development, with PD-1 blockade in an autochthonous FGFR2K660N/p53mut lung cancer mouse model. Erdafitinib monotherapy treatment resulted in substantial tumor control but no significant survival benefit. Although anti-PD-1 alone was ineffective, the erdafitinib and anti-PD-1 combination induced significant tumor regression and improved survival. For both erdafitinib monotherapy and combination treatments, tumor control was accompanied by tumor-intrinsic, FGFR pathway inhibition, increased T-cell infiltration, decreased regulatory T cells, and downregulation of PD-L1 expression on tumor cells. These effects were not observed in a KRASG12C-mutant genetically engineered mouse model, which is insensitive to FGFR inhibition, indicating that the immune changes mediated by erdafitinib may be initiated as a consequence of tumor cell killing. A decreased fraction of tumor-associated macrophages also occurred but only in combination-treated tumors. Treatment with erdafitinib decreased T-cell receptor (TCR) clonality, reflecting a broadening of the TCR repertoire induced by tumor cell death, whereas combination with anti-PD-1 led to increased TCR clonality, suggesting a more focused antitumor T-cell response. Our results showed that the combination of erdafitinib and anti-PD-1 drives expansion of T-cell clones and immunologic changes in the tumor microenvironment to support enhanced antitumor immunity and survival.

Computational pathology definitions, best practices, and recommendations for regulatory guidance: a white paper from the Digital Pathology Association.

In this white paper, experts from the Digital Pathology Association (DPA) define terminology and concepts in the emerging field of computational pathology, with a focus on its application to histology images analyzed together with their associated patient data to extract information. This review offers a historical perspective and describes the potential clinical benefits from research and applications in this field, as well as significant obstacles to adoption. Best practices for implementing computational pathology workflows are presented. These include infrastructure considerations, acquisition of training data, quality assessments, as well as regulatory, ethical, and cyber-security concerns. Recommendations are provided for regulators, vendors, and computational pathology practitioners in order to facilitate progress in the field. © 2019 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.

EN1 Is a Transcriptional Dependency in Triple-Negative Breast Cancer Associated with Brain Metastasis.

To define transcriptional dependencies of triple-negative breast cancer (TNBC), we identified transcription factors highly and specifically expressed in primary TNBCs and tested their requirement for cell growth in a panel of breast cancer cell lines. We found that EN1 (engrailed 1) is overexpressed in TNBCs and its downregulation preferentially and significantly reduced viability and tumorigenicity in TNBC cell lines. By integrating gene expression changes after EN1 downregulation with EN1 chromatin binding patterns, we identified genes involved in WNT and Hedgehog signaling, neurogenesis, and axonal guidance as direct EN1 transcriptional targets. Quantitative proteomic analyses of EN1-bound chromatin complexes revealed association with transcriptional repressors and coactivators including TLE3, TRIM24, TRIM28, and TRIM33. High expression of EN1 correlated with short overall survival and increased risk of developing brain metastases in patients with TNBC. Thus, EN1 is a prognostic marker and a potential therapeutic target in TNBC. SIGNIFICANCE: These findings show that the EN1 transcription factor regulates neurogenesis-related genes and is associated with brain metastasis in triple-negative breast cancer.

Predicting breast tumor proliferation from whole-slide images: The TUPAC16 challenge.

Tumor proliferation is an important biomarker indicative of the prognosis of breast cancer patients. Assessment of tumor proliferation in a clinical setting is a highly subjective and labor-intensive task. Previous efforts to automate tumor proliferation assessment by image analysis only focused on mitosis detection in predefined tumor regions. However, in a real-world scenario, automatic mitosis detection should be performed in whole-slide images (WSIs) and an automatic method should be able to produce a tumor proliferation score given a WSI as input. To address this, we organized the TUmor Proliferation Assessment Challenge 2016 (TUPAC16) on prediction of tumor proliferation scores from WSIs. The challenge dataset consisted of 500 training and 321 testing breast cancer histopathology WSIs. In order to ensure fair and independent evaluation, only the ground truth for the training dataset was provided to the challenge participants. The first task of the challenge was to predict mitotic scores, i.e., to reproduce the manual method of assessing tumor proliferation by a pathologist. The second task was to predict the gene expression based PAM50 proliferation scores from the WSI. The best performing automatic method for the first task achieved a quadratic-weighted Cohen's kappa score of κ = 0.567, 95% CI [0.464, 0.671] between the predicted scores and the ground truth. For the second task, the predictions of the top method had a Spearman's correlation coefficient of r = 0.617, 95% CI [0.581 0.651] with the ground truth. This was the first comparison study that investigated tumor proliferation assessment from WSIs. The achieved results are promising given the difficulty of the tasks and weakly-labeled nature of the ground truth. However, further research is needed to improve the practical utility of image analysis methods for this task.

SPOP Promotes Nanog Destruction to Suppress Stem Cell Traits and Prostate Cancer Progression.

Frequent SPOP mutation defines the molecular feature underlying one of seven sub-types of human prostate cancer (PrCa). However, it remains largely elusive how SPOP functions as a tumor suppressor in PrCa. Here, we report that SPOP suppresses stem cell traits of both embryonic stem cells and PrCa cells through promoting Nanog poly-ubiquitination and subsequent degradation. Mechanistically, Nanog, but not other pluripotency-determining factors including Oct4, Sox2, and Klf4, specifically interacts with SPOP via a conservative degron motif. Importantly, cancer-derived mutations in SPOP or at the Nanog-degron (S68Y) disrupt SPOP-mediated destruction of Nanog, leading to elevated cancer stem cell traits and PrCa progression. Notably, we identify the Pin1 oncoprotein as an upstream Nanog regulator that impairs its recognition by SPOP and thereby stabilizes Nanog. Thus, Pin1 inhibitors promote SPOP-mediated destruction of Nanog, which provides the molecular insight and rationale to use Pin1 inhibitor(s) for targeted therapies of PrCa patients with wild-type SPOP.

Androgen Receptor Expression and Breast Cancer Survival: Results From the Nurses' Health Studies.

BACKGROUND: Hormone receptor signaling is critical in the progression of breast cancers, although the role of the androgen receptor (AR) remains unclear, particularly for estrogen receptor (ER)-negative tumors. This study assessed AR protein expression as a prognostic marker for breast cancer mortality. METHODS: This study included 4147 pre- and postmenopausal women with invasive breast cancer from the Nurses' Health Study (diagnosed 1976-2008) and Nurses' Health Study II (1989-2008) cohorts. AR protein expression was evaluated by immunohistochemistry and scored through pathologist review and as a digitally quantified continuous measure. Hazard ratios (HR) and 95% confidence intervals (CI) of breast cancer mortality were estimated from Cox proportional hazards models, adjusting for patient, tumor, and treatment covariates. RESULTS: Over a median 16.5 years of follow-up, there were 806 deaths due to breast cancer. In the 7 years following diagnosis, AR expression was associated with a 27% reduction in breast cancer mortality overall (multivariable HR = 0.73, 95% CI = 0.58 to 0.91) a 47% reduction for ER+ cancers (HR = 0.53, 95% CI = 0.41 to 0.69), and a 62% increase for ER- cancers (HR = 1.62, 95% CI = 1.18 to 2.22) (P heterogeneity < .001). A log-linear association was observed between AR expression and breast cancer mortality among ER- cancers (HR = 1.14, 95% CI = 1.02 to 1.26 per each 10% increase in AR), although no log-linear association was observed among ER+ cancers. CONCLUSIONS: AR expression was associated with improved prognosis in ER+ tumors and worse prognosis in ER- tumors in the first 5-10 years postdiagnosis. These findings support the continued evaluation of AR-targeted therapies for AR+/ER- breast cancers.

Molecular mechanisms linking high body mass index to breast cancer etiology in post-menopausal breast tumor and tumor-adjacent tissues.

PURPOSE: In post-menopausal women, high body mass index (BMI) is an established breast cancer risk factor and is associated with worse breast cancer prognosis. We assessed the associations between BMI and gene expression of both breast tumor and adjacent tissue in estrogen receptor-positive (ER+) and estrogen receptor-negative (ER-) diseases to help elucidate the mechanisms linking obesity with breast cancer biology in 519 post-menopausal women from the Nurses' Health Study (NHS) and NHSII. METHODS: Differential gene expression was analyzed separately in ER+ and ER- disease both comparing overweight (BMI ≥ 25 to < 30) or obese (BMI ≥ 30) women to women with normal BMI (BMI < 25), and per 5 kg/m2 increase in BMI. Analyses controlled for age and year of diagnosis, physical activity, alcohol consumption, and hormone therapy use. Gene set enrichment analyses were performed and validated among a subset of post-menopausal cases in The Cancer Genome Atlas (for tumor) and Polish Breast Cancer Study (for tumor-adjacent). RESULTS: No gene was differentially expressed by BMI (FDR < 0.05). BMI was significantly associated with increased cellular proliferation pathways, particularly in ER+ tumors, and increased inflammation pathways in ER- tumor and ER- tumor-adjacent tissues (FDR < 0.05). High BMI was associated with upregulation of genes involved in epithelial-mesenchymal transition in ER+ tumor-adjacent tissues. CONCLUSIONS: This study provides insights into molecular mechanisms of BMI influencing post-menopausal breast cancer biology. Tumor and tumor-adjacent tissues provide independent information about potential mechanisms.

Androgen receptor expression in normal breast tissue and subsequent breast cancer risk.

Sex steroid hormone signaling is critical in the development of breast cancers, although the role of the androgen receptor remains unclear. This study evaluated androgen receptor (AR) expression in normal breast tissue as a potential marker of breast cancer risk. We conducted a nested case-control study of women with benign breast disease (BBD) within the Nurses' Health Studies. Epithelial AR expression was assessed by immunohistochemistry in normal tissue from the BBD biopsy and the percent of positive nuclei was estimated in ordinal categories of 10% for 78 breast cancer cases and 276 controls. Logistic regression models adjusting for the matching factors and BBD lesion type were used to calculate odds ratios (ORs) for the association between AR expression (tertiles: ≤10%, 11-30%, and >30%) and breast cancer risk. AR expression in normal breast tissue was not associated with subsequent breast cancer risk (ORT3vsT1 = 0.9, 95% CI = 0.4-1.8, p trend = 0.68). In comparison with low AR/low ER women, ORs of 0.4 (95% CI = 0.1-1.2) for high AR/high ER women, 1.8 (95% CI = 0.4-7.8) for low AR/high ER women, and 0.7 (95% CI = 0.3-1.6) for high AR/low ER women were observed (p interaction = 0.21). Ki67 did not modify the association between AR expression and breast cancer risk (p interaction = 0.75). There was little evidence for an overall association between AR expression in normal breast tissue and breast cancer risk. These findings did not show that the AR association varied by Ki67 expression in normal breast tissue, though there was suggestive heterogeneity by ER expression.

Using deep convolutional neural networks to identify and classify tumor-associated stroma in diagnostic breast biopsies.

The breast stromal microenvironment is a pivotal factor in breast cancer development, growth and metastases. Although pathologists often detect morphologic changes in stroma by light microscopy, visual classification of such changes is subjective and non-quantitative, limiting its diagnostic utility. To gain insights into stromal changes associated with breast cancer, we applied automated machine learning techniques to digital images of 2387 hematoxylin and eosin stained tissue sections of benign and malignant image-guided breast biopsies performed to investigate mammographic abnormalities among 882 patients, ages 40-65 years, that were enrolled in the Breast Radiology Evaluation and Study of Tissues (BREAST) Stamp Project. Using deep convolutional neural networks, we trained an algorithm to discriminate between stroma surrounding invasive cancer and stroma from benign biopsies. In test sets (928 whole-slide images from 330 patients), this algorithm could distinguish biopsies diagnosed as invasive cancer from benign biopsies solely based on the stromal characteristics (area under the receiver operator characteristics curve = 0.962). Furthermore, without being trained specifically using ductal carcinoma in situ as an outcome, the algorithm detected tumor-associated stroma in greater amounts and at larger distances from grade 3 versus grade 1 ductal carcinoma in situ. Collectively, these results suggest that algorithms based on deep convolutional neural networks that evaluate only stroma may prove useful to classify breast biopsies and aid in understanding and evaluating the biology of breast lesions.

Computationally-Guided Development of a Stromal Inflammation Histologic Biomarker in Lung Squamous Cell Carcinoma.

The goal of this study is to use computational pathology to help guide the development of human-based prognostic H&E biomarker(s) suitable for research and potential clinical use in lung squamous cell carcinoma (SCC). We started with high-throughput computational image analysis with tissue microarrays (TMAs) to screen for histologic features associated with patient overall survival, and found that features related to stromal inflammation were the most strongly prognostic. Based on this, we developed an H&E stromal inflammation (SI) score. The prognostic value of the SI score was validated by two blinded human observers on two large cohorts from a single institution. The SI score was found to be reproducible on TMAs (Spearman rho = 0.88 between the two observers), and highly prognostic (e.g. hazard ratio = 0.32; 95% confidence interval: 0.19-0.54; p-value = 2.5 × 10-5 in multivariate analyses), particularly in comparison to established histologic biomarkers. Guided by downstream molecular/biomarker correlation studies starting with TCGA cases, we investigated the hypothesis that epithelial PD-L1 expression modified the prognostic value of SI. Our research demonstrates that computational pathology can be an efficient hypothesis generator for human pathology research, and support the histologic evaluation of SI as a prognostic biomarker in lung SCCs.

Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer.

Importance: Application of deep learning algorithms to whole-slide pathology images can potentially improve diagnostic accuracy and efficiency. Objective: Assess the performance of automated deep learning algorithms at detecting metastases in hematoxylin and eosin-stained tissue sections of lymph nodes of women with breast cancer and compare it with pathologists' diagnoses in a diagnostic setting. Design, Setting, and Participants: Researcher challenge competition (CAMELYON16) to develop automated solutions for detecting lymph node metastases (November 2015-November 2016). A training data set of whole-slide images from 2 centers in the Netherlands with (n = 110) and without (n = 160) nodal metastases verified by immunohistochemical staining were provided to challenge participants to build algorithms. Algorithm performance was evaluated in an independent test set of 129 whole-slide images (49 with and 80 without metastases). The same test set of corresponding glass slides was also evaluated by a panel of 11 pathologists with time constraint (WTC) from the Netherlands to ascertain likelihood of nodal metastases for each slide in a flexible 2-hour session, simulating routine pathology workflow, and by 1 pathologist without time constraint (WOTC). Exposures: Deep learning algorithms submitted as part of a challenge competition or pathologist interpretation. Main Outcomes and Measures: The presence of specific metastatic foci and the absence vs presence of lymph node metastasis in a slide or image using receiver operating characteristic curve analysis. The 11 pathologists participating in the simulation exercise rated their diagnostic confidence as definitely normal, probably normal, equivocal, probably tumor, or definitely tumor. Results: The area under the receiver operating characteristic curve (AUC) for the algorithms ranged from 0.556 to 0.994. The top-performing algorithm achieved a lesion-level, true-positive fraction comparable with that of the pathologist WOTC (72.4% [95% CI, 64.3%-80.4%]) at a mean of 0.0125 false-positives per normal whole-slide image. For the whole-slide image classification task, the best algorithm (AUC, 0.994 [95% CI, 0.983-0.999]) performed significantly better than the pathologists WTC in a diagnostic simulation (mean AUC, 0.810 [range, 0.738-0.884]; P < .001). The top 5 algorithms had a mean AUC that was comparable with the pathologist interpreting the slides in the absence of time constraints (mean AUC, 0.960 [range, 0.923-0.994] for the top 5 algorithms vs 0.966 [95% CI, 0.927-0.998] for the pathologist WOTC). Conclusions and Relevance: In the setting of a challenge competition, some deep learning algorithms achieved better diagnostic performance than a panel of 11 pathologists participating in a simulation exercise designed to mimic routine pathology workflow; algorithm performance was comparable with an expert pathologist interpreting whole-slide images without time constraints. Whether this approach has clinical utility will require evaluation in a clinical setting.