Nondestructive 3D pathology with analysis of nuclear features for prostate cancer risk assessment.

Prostate cancer treatment decisions rely heavily on subjective visual interpretation [assigning Gleason patterns or International Society of Urological Pathology (ISUP) grade groups] of limited numbers of two-dimensional (2D) histology sections. Under this paradigm, interobserver variance is high, with ISUP grades not correlating well with outcome for individual patients, and this contributes to the over- and undertreatment of patients. Recent studies have demonstrated improved prognostication of prostate cancer outcomes based on computational analyses of glands and nuclei within 2D whole slide images. Our group has also shown that the computational analysis of three-dimensional (3D) glandular features, extracted from 3D pathology datasets of whole intact biopsies, can allow for improved recurrence prediction compared to corresponding 2D features. Here we seek to expand on these prior studies by exploring the prognostic value of 3D shape-based nuclear features in prostate cancer (e.g. nuclear size, sphericity). 3D pathology datasets were generated using open-top light-sheet (OTLS) microscopy of 102 cancer-containing biopsies extracted ex vivo from the prostatectomy specimens of 46 patients. A deep learning-based workflow was developed for 3D nuclear segmentation within the glandular epithelium versus stromal regions of the biopsies. 3D shape-based nuclear features were extracted, and a nested cross-validation scheme was used to train a supervised machine classifier based on 5-year biochemical recurrence (BCR) outcomes. Nuclear features of the glandular epithelium were found to be more prognostic than stromal cell nuclear features (area under the ROC curve [AUC] = 0.72 versus 0.63). 3D shape-based nuclear features of the glandular epithelium were also more strongly associated with the risk of BCR than analogous 2D features (AUC = 0.72 versus 0.62). The results of this preliminary investigation suggest that 3D shape-based nuclear features are associated with prostate cancer aggressiveness and could be of value for the development of decision-support tools. © 2023 The Pathological Society of Great Britain and Ireland.

Computational Analysis of Routine Biopsies Improves Diagnosis and Prediction of Cardiac Allograft Vasculopathy.

BACKGROUND: Cardiac allograft vasculopathy (CAV) is a leading cause of morbidity and mortality for heart transplant recipients. Although clinical risk factors for CAV have been established, no personalized prognostic test exists to confidently identify patients at high versus low risk of developing aggressive CAV. This investigation aimed to leverage computational methods for analyzing digital pathology images from routine endomyocardial biopsies (EMBs) to develop a precision medicine tool for predicting CAV years before overt clinical presentation. METHODS: Clinical data from 1 year after transplant were collected on 302 transplant recipients from the University of Pennsylvania, including 53 patients with early-onset CAV and 249 no early-onset CAV controls. These data were used to generate a clinical model (Clinical Risk Factor Future Cardiac Allograft Vasculopathy Prediction Model [ClinCAV-Pr]) for predicting future CAV development. From this cohort, 183 archived EMBs were collected for CD31 and modified trichrome staining and then digitally scanned. These included 1-year posttransplant EMBs from 50 patients with early-onset CAV and 82 patients with no early-onset CAV, as well as 51 EMBs from disease control patients obtained at the time of definitive coronary angiography confirming CAV. Using biologically inspired, handcrafted features extracted from digitized EMBs, quantitative histological models for differentiating no early-onset CAV from disease controls (Histological Cardiac Allograft Vasculopathy Diagnostic Model [HistoCAV-Dx]) and for predicting future CAV from 1-year posttransplant EMBs were developed (Histological Future Cardiac Allograft Vasculopathy Prediction Model [HistoCAV-Pr]). The performance of histological and clinical models for predicting future CAV (ie, HistoCAV-Pr and ClinCAV-Pr, respectively) were compared in a held-out validation set before being combined to assess the added predictive value of an integrated predictive model (Integrated Histological/Clinical Risk Factor Future Cardiac Allograft Vasculopathy Prediction Model [iCAV-Pr]). RESULTS: ClinCAV-Pr achieved modest performance on the independent test set, with an area under the receiver operating curve (AUROC) of 0.70. The HistoCAV-Dx model for diagnosing CAV achieved excellent discrimination, with an AUROC of 0.91, whereas the HistoCAV-Pr model for predicting CAV achieved good performance with an AUROC of 0.80. The integrated iCAV-Pr model achieved excellent predictive performance, with an AUROC of 0.93 on the held-out test set. CONCLUSIONS: Prediction of future CAV development is greatly improved by incorporation of computationally extracted histological features. These results suggest morphological details contained within regularly obtained biopsy tissue have the potential to enhance precision and personalization of treatment plans for patients after heart transplant.

Visual Assessment of 2-Dimensional Levels Within 3-Dimensional Pathology Data Sets of Prostate Needle Biopsies Reveals Substantial Spatial Heterogeneity.

Prostate cancer prognostication largely relies on visual assessment of a few thinly sectioned biopsy specimens under a microscope to assign a Gleason grade group (GG). Unfortunately, the assigned GG is not always associated with a patient's outcome in part because of the limited sampling of spatially heterogeneous tumors achieved by 2-dimensional histopathology. In this study, open-top light-sheet microscopy was used to obtain 3-dimensional pathology data sets that were assessed by 4 human readers. Intrabiopsy variability was assessed by asking readers to perform Gleason grading of 5 different levels per biopsy for a total of 20 core needle biopsies (ie, 100 total images). Intrabiopsy variability (Cohen κ) was calculated as the worst pairwise agreement in GG between individual levels within each biopsy and found to be 0.34, 0.34, 0.38, and 0.43 for the 4 pathologists. These preliminary results reveal that even within a 1-mm-diameter needle core, GG based on 2-dimensional images can vary dramatically depending on the location within a biopsy being analyzed. We believe that morphologic assessment of whole biopsies in 3 dimension has the potential to enable more reliable and consistent tumor grading.

Pathologist Computer-Aided Diagnostic Scoring of Tumor Cell Fraction: A Swiss National Study.

Tumor cell fraction (TCF) estimation is a common clinical task with well-established large interobserver variability. It thus provides an ideal test bed to evaluate potential impacts of employing a tumor cell fraction computer-aided diagnostic (TCFCAD) tool to support pathologists' evaluation. During a National Slide Seminar event, pathologists (n = 69) were asked to visually estimate TCF in 10 regions of interest (ROIs) from hematoxylin and eosin colorectal cancer images intentionally curated for diverse tissue compositions, cellularity, and stain intensities. Next, they re-evaluated the same ROIs while being provided a TCFCAD-created overlay highlighting predicted tumor vs nontumor cells, together with the corresponding TCF percentage. Participants also reported confidence levels in their assessments using a 5-tier scale, indicating no confidence to high confidence, respectively. The TCF ground truth (GT) was defined by manual cell-counting by experts. When assisted, interobserver variability significantly decreased, showing estimates converging to the GT. This improvement remained even when TCFCAD predictions deviated slightly from the GT. The standard deviation (SD) of the estimated TCF to the GT across ROIs was 9.9% vs 5.8% with TCFCAD (P < .0001). The intraclass correlation coefficient increased from 0.8 to 0.93 (95% CI, 0.65-0.93 vs 0.86-0.98), and pathologists stated feeling more confident when aided (3.67 ± 0.81 vs 4.17 ± 0.82 with the computer-aided diagnostic [CAD] tool). TCFCAD estimation support demonstrated improved scoring accuracy, interpathologist agreement, and scoring confidence. Interestingly, pathologists also expressed more willingness to use such a CAD tool at the end of the survey, highlighting the importance of training/education to increase adoption of CAD systems.

Assessment of a computerized quantitative quality control tool for whole slide images of kidney biopsies.

Inconsistencies in the preparation of histology slides and whole-slide images (WSIs) may lead to challenges with subsequent image analysis and machine learning approaches for interrogating the WSI. These variabilities are especially pronounced in multicenter cohorts, where batch effects (i.e. systematic technical artifacts unrelated to biological variability) may introduce biases to machine learning algorithms. To date, manual quality control (QC) has been the de facto standard for dataset curation, but remains highly subjective and is too laborious in light of the increasing scale of tissue slide digitization efforts. This study aimed to evaluate a computer-aided QC pipeline for facilitating a reproducible QC process of WSI datasets. An open source tool, HistoQC, was employed to identify image artifacts and compute quantitative metrics describing visual attributes of WSIs to the Nephrotic Syndrome Study Network (NEPTUNE) digital pathology repository. A comparison in inter-reader concordance between HistoQC aided and unaided curation was performed to quantify improvements in curation reproducibility. HistoQC metrics were additionally employed to quantify the presence of batch effects within NEPTUNE WSIs. Of the 1814 WSIs (458 H&E, 470 PAS, 438 silver, 448 trichrome) from n = 512 cases considered in this study, approximately 9% (163) were identified as unsuitable for subsequent computational analysis. The concordance in the identification of these WSIs among computational pathologists rose from moderate (Gwet's AC1 range 0.43 to 0.59 across stains) to excellent (Gwet's AC1 range 0.79 to 0.93 across stains) agreement when aided by HistoQC. Furthermore, statistically significant batch effects (p < 0.001) in the NEPTUNE WSI dataset were discovered. Taken together, our findings strongly suggest that quantitative QC is a necessary step in the curation of digital pathology cohorts. © 2020 The Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

An automated computational image analysis pipeline for histological grading of cardiac allograft rejection.

AIM: Allograft rejection is a serious concern in heart transplant medicine. Though endomyocardial biopsy with histological grading is the diagnostic standard for rejection, poor inter-pathologist agreement creates significant clinical uncertainty. The aim of this investigation is to demonstrate that cellular rejection grades generated via computational histological analysis are on-par with those provided by expert pathologists. METHODS AND RESULTS: The study cohort consisted of 2472 endomyocardial biopsy slides originating from three major US transplant centres. The 'Computer-Assisted Cardiac Histologic Evaluation (CACHE)-Grader' pipeline was trained using an interpretable, biologically inspired, 'hand-crafted' feature extraction approach. From a menu of 154 quantitative histological features relating the density and orientation of lymphocytes, myocytes, and stroma, a model was developed to reproduce the 4-grade clinical standard for cellular rejection diagnosis. CACHE-grader interpretations were compared with independent pathologists and the 'grade of record', testing for non-inferiority (δ = 6%). Study pathologists achieved a 60.7% agreement [95% confidence interval (CI): 55.2-66.0%] with the grade of record, and pair-wise agreement among all human graders was 61.5% (95% CI: 57.0-65.8%). The CACHE-Grader met the threshold for non-inferiority, achieving a 65.9% agreement (95% CI: 63.4-68.3%) with the grade of record and a 62.6% agreement (95% CI: 60.3-64.8%) with all human graders. The CACHE-Grader demonstrated nearly identical performance in internal and external validation sets (66.1% vs. 65.8%), resilience to inter-centre variations in tissue processing/digitization, and superior sensitivity for high-grade rejection (74.4% vs. 39.5%, P < 0.001). CONCLUSION: These results show that the CACHE-grader pipeline, derived using intuitive morphological features, can provide expert-quality rejection grading, performing within the range of inter-grader variability seen among human pathologists.

Development and evaluation of deep learning-based segmentation of histologic structures in the kidney cortex with multiple histologic stains.

The application of deep learning for automated segmentation (delineation of boundaries) of histologic primitives (structures) from whole slide images can facilitate the establishment of novel protocols for kidney biopsy assessment. Here, we developed and validated deep learning networks for the segmentation of histologic structures on kidney biopsies and nephrectomies. For development, we examined 125 biopsies for Minimal Change Disease collected across 29 NEPTUNE enrolling centers along with 459 whole slide images stained with Hematoxylin & Eosin (125), Periodic Acid Schiff (125), Silver (102), and Trichrome (107) divided into training, validation and testing sets (ratio 6:1:3). Histologic structures were manually segmented (30048 total annotations) by five nephropathologists. Twenty deep learning models were trained with optimal digital magnification across the structures and stains. Periodic Acid Schiff-stained whole slide images yielded the best concordance between pathologists and deep learning segmentation across all structures (F-scores: 0.93 for glomerular tufts, 0.94 for glomerular tuft plus Bowman's capsule, 0.91 for proximal tubules, 0.93 for distal tubular segments, 0.81 for peritubular capillaries, and 0.85 for arteries and afferent arterioles). Optimal digital magnifications were 5X for glomerular tuft/tuft plus Bowman's capsule, 10X for proximal/distal tubule, arteries and afferent arterioles, and 40X for peritubular capillaries. Silver stained whole slide images yielded the worst deep learning performance. Thus, this largest study to date adapted deep learning for the segmentation of kidney histologic structures across multiple stains and pathology laboratories. All data used for training and testing and a detailed online tutorial will be publicly available.

T1 and T2 MR fingerprinting measurements of prostate cancer and prostatitis correlate with deep learning-derived estimates of epithelium, lumen, and stromal composition on corresponding whole mount histopathology.

OBJECTIVES: To explore the associations between T1 and T2 magnetic resonance fingerprinting (MRF) measurements and corresponding tissue compartment ratios (TCRs) on whole mount histopathology of prostate cancer (PCa) and prostatitis. MATERIALS AND METHODS: A retrospective, IRB-approved, HIPAA-compliant cohort consisting of 14 PCa patients who underwent 3 T multiparametric MRI along with T1 and T2 MRF maps prior to radical prostatectomy was used. Correspondences between whole mount specimens and MRI and MRF were manually established. Prostatitis, PCa, and normal peripheral zone (PZ) regions of interest (ROIs) on pathology were segmented for TCRs of epithelium, lumen, and stroma using two U-net deep learning models. Corresponding ROIs were mapped to T2-weighted MRI (T2w), apparent diffusion coefficient (ADC), and T1 and T2 MRF maps. Their correlations with TCRs were computed using Pearson's correlation coefficient (R). Statistically significant differences in means were assessed using one-way ANOVA. RESULTS: Statistically significant differences (p < 0.01) in means of TCRs and T1 and T2 MRF were observed between PCa, prostatitis, and normal PZ. A negative correlation was observed between T1 and T2 MRF and epithelium (R = - 0.38, - 0.44, p < 0.05) of PCa. T1 MRF was correlated in opposite directions with stroma of PCa and prostatitis (R = 0.35, - 0.44, p < 0.05). T2 MRF was positively correlated with lumen of PCa and prostatitis (R = 0.57, 0.46, p < 0.01). Mean T2 MRF showed significant differences (p < 0.01) between PCa and prostatitis across both transition zone (TZ) and PZ, while mean T1 MRF was significant (p = 0.02) in TZ. CONCLUSION: Significant associations between MRF (T1 in the TZ and T2 in the PZ) and tissue compartments on corresponding histopathology were observed. KEY POINTS: • Mean T2 MRF measurements and ADC within cancerous regions of interest dropped with increasing ISUP prognostic groups (IPG). • Mean T1 and T2 MRF measurements were significantly different (p < 0.001) across IPGs, prostatitis, and normal peripheral zone (NPZ). • T2 MRF showed stronger correlations in the peripheral zone, while T1 MRF showed stronger correlations in the transition zone with histopathology for prostate cancer.

Biogeography of microbial bile acid transformations along the murine gut.

Bile acids, which are synthesized from cholesterol by the liver, are chemically transformed along the intestinal tract by the gut microbiota, and the products of these transformations signal through host receptors, affecting overall host health. These transformations include bile acid deconjugation, oxidation, and 7α-dehydroxylation. An understanding of the biogeography of bile acid transformations in the gut is critical because deconjugation is a prerequisite for 7α-dehydroxylation and because most gut microorganisms harbor bile acid transformation capacity. Here, we used a coupled metabolomic and metaproteomic approach to probe in vivo activity of the gut microbial community in a gnotobiotic mouse model. Results revealed the involvement of Clostridium scindens in 7α-dehydroxylation, of the genera Muribaculum and Bacteroides in deconjugation, and of six additional organisms in oxidation (the genera Clostridium, Muribaculum, Bacteroides, Bifidobacterium, Acutalibacter, and Akkermansia). Furthermore, the bile acid profile in mice with a more complex microbiota, a dysbiosed microbiota, or no microbiota was considered. For instance, conventional mice harbor a large diversity of bile acids, but treatment with an antibiotic such as clindamycin results in the complete inhibition of 7α-dehydroxylation, underscoring the strong inhibition of organisms that are capable of carrying out this process by this compound. Finally, a comparison of the hepatic bile acid pool size as a function of microbiota revealed that a reduced microbiota affects host signaling but not necessarily bile acid synthesis. In this study, bile acid transformations were mapped to the associated active microorganisms, offering a systematic characterization of the relationship between microbiota and bile acid composition.

Nuclear shape and orientation features from H&E images predict survival in early-stage estrogen receptor-positive breast cancers.

Early-stage estrogen receptor-positive (ER+) breast cancer (BCa) is the most common type of BCa in the United States. One critical question with these tumors is identifying which patients will receive added benefit from adjuvant chemotherapy. Nuclear pleomorphism (variance in nuclear shape and morphology) is an important constituent of breast grading schemes, and in ER+ cases, the grade is highly correlated with disease outcome. This study aimed to investigate whether quantitative computer-extracted image features of nuclear shape and orientation on digitized images of hematoxylin-stained and eosin-stained tissue of lymph node-negative (LN-), ER+ BCa could help stratify patients into discrete (<10 years short-term vs. >10 years long-term survival) outcome groups independent of standard clinical and pathological parameters. We considered a tissue microarray (TMA) cohort of 276 ER+, LN- patients comprising 150 patients with long-term and 126 patients with short-term overall survival, wherein 177 randomly chosen cases formed the modeling set, and 99 remaining cases the test set. Segmentation of individual nuclei was performed using multiresolution watershed; subsequently, 615 features relating to nuclear shape/texture and orientation disorder were extracted from each TMA spot. The Wilcoxon's rank-sum test identified the 15 most prognostic quantitative histomorphometric features within the modeling set. These features were then subsequently combined via a linear discriminant analysis classifier and evaluated on the test set to assign a probability of long-term vs. short-term disease-specific survival. In univariate survival analysis, patients identified by the image classifier as high risk had significantly poorer survival outcome: hazard ratio (95% confident interval) = 2.91(1.23-6.92), p = 0.02786. Multivariate analysis controlling for T-stage, histology grade, and nuclear grade showed the classifier to be independently predictive of poorer survival: hazard ratio (95% confident interval) = 3.17(0.33-30.46), p = 0.01039. Our results suggest that quantitative histomorphometric features of nuclear shape and orientation are strongly and independently predictive of patient survival in ER+, LN- BCa.

Quantitative nuclear histomorphometry predicts oncotype DX risk categories for early stage ER+ breast cancer.

BACKGROUND: Gene-expression companion diagnostic tests, such as the Oncotype DX test, assess the risk of early stage Estrogen receptor (ER) positive (+) breast cancers, and guide clinicians in the decision of whether or not to use chemotherapy. However, these tests are typically expensive, time consuming, and tissue-destructive. METHODS: In this paper, we evaluate the ability of computer-extracted nuclear morphology features from routine hematoxylin and eosin (H&E) stained images of 178 early stage ER+ breast cancer patients to predict corresponding risk categories derived using the Oncotype DX test. A total of 216 features corresponding to the nuclear shape and architecture categories from each of the pathologic images were extracted and four feature selection schemes: Ranksum, Principal Component Analysis with Variable Importance on Projection (PCA-VIP), Maximum-Relevance, Minimum Redundancy Mutual Information Difference (MRMR MID), and Maximum-Relevance, Minimum Redundancy - Mutual Information Quotient (MRMR MIQ), were employed to identify the most discriminating features. These features were employed to train 4 machine learning classifiers: Random Forest, Neural Network, Support Vector Machine, and Linear Discriminant Analysis, via 3-fold cross validation. RESULTS: The four sets of risk categories, and the top Area Under the receiver operating characteristic Curve (AUC) machine classifier performances were: 1) Low ODx and Low mBR grade vs. High ODx and High mBR grade (Low-Low vs. High-High) (AUC = 0.83), 2) Low ODx vs. High ODx (AUC = 0.72), 3) Low ODx vs. Intermediate and High ODx (AUC = 0.58), and 4) Low and Intermediate ODx vs. High ODx (AUC = 0.65). Trained models were tested independent validation set of 53 cases which comprised of Low and High ODx risk, and demonstrated per-patient accuracies ranging from 75 to 86%. CONCLUSION: Our results suggest that computerized image analysis of digitized H&E pathology images of early stage ER+ breast cancer might be able predict the corresponding Oncotype DX risk categories.

An oral cavity squamous cell carcinoma quantitative histomorphometric-based image classifier of nuclear morphology can risk stratify patients for disease-specific survival.

Oral cavity squamous cell carcinoma is the most common type of head and neck carcinoma. Its incidence is increasing worldwide, and it is associated with major morbidity and mortality. It is often unclear which patients have aggressive, treatment refractory tumors vs those whose tumors will be more responsive to treatment. Better identification of patients with high- vs low-risk cancers could help provide more tailored treatment approaches and could improve survival rates while decreasing treatment-related morbidity. This study investigates computer-extracted image features of nuclear shape and texture on digitized images of H&E-stained tissue sections for risk stratification of oral cavity squamous cell carcinoma patients compared with standard clinical and pathologic parameters. With a tissue microarray cohort of 115 retrospectively identified oral cavity squamous cell carcinoma patients, 50 were randomly chosen as the modeling set, and the remaining 65 constituted the test set. Following nuclear segmentation and feature extraction, the Wilcoxon rank sum test was used to identify the five most prognostic quantitative histomorphometric features from the modeling set. These top ranked features were then combined via a machine learning classifier to construct the oral cavity histomorphometric-based image classifier (OHbIC). The classifier was then validated for its ability to risk stratify patients for disease-specific outcomes on the test set. On the test set, the classifier yielded an area under the receiver operating characteristic curve of 0.72 in distinguishing disease-specific outcomes. In univariate survival analysis, high-risk patients predicted by the classifier had significantly poorer disease-specific survival (P=0.0335). In multivariate analysis controlling for T/N-stage, resection margins, and smoking status, positive classifier results were independently predictive of poorer disease-specific survival: hazard ratio (95% confidence interval)=11.023 (2.62-46.38) and P=0.001. Our results suggest that quantitative histomorphometric features of local nuclear architecture derived from digitized H&E slides of oral cavity squamous cell carcinomas are independently predictive of patient survival.

A deep learning based strategy for identifying and associating mitotic activity with gene expression derived risk categories in estrogen receptor positive breast cancers.

The treatment and management of early stage estrogen receptor positive (ER+) breast cancer is hindered by the difficulty in identifying patients who require adjuvant chemotherapy in contrast to those that will respond to hormonal therapy. To distinguish between the more and less aggressive breast tumors, which is a fundamental criterion for the selection of an appropriate treatment plan, Oncotype DX (ODX) and other gene expression tests are typically employed. While informative, these gene expression tests are expensive, tissue destructive, and require specialized facilities. Bloom-Richardson (BR) grade, the common scheme employed in breast cancer grading, has been shown to be correlated with the Oncotype DX risk score. Unfortunately, studies have also shown that the BR grade determined experiences notable inter-observer variability. One of the constituent categories in BR grading is the mitotic index. The goal of this study was to develop a deep learning (DL) classifier to identify mitotic figures from whole slides images of ER+ breast cancer, the hypothesis being that the number of mitoses identified by the DL classifier would correlate with the corresponding Oncotype DX risk categories. The mitosis detector yielded an average F-score of 0.556 in the AMIDA mitosis dataset using a 6-fold validation setup. For a cohort of 174 whole slide images with early stage ER+ breast cancer for which the corresponding Oncotype DX score was available, the distributions of the number of mitoses identified by the DL classifier was found to be significantly different between the high vs low Oncotype DX risk groups (P < 0.01). Comparisons of other risk groups, using both ODX score and histological grade, were also found to present significantly different automated mitoses distributions. Additionally, a support vector machine classifier trained to separate low/high Oncotype DX risk categories using the mitotic count determined by the DL classifier yielded a 83.19% classification accuracy. © 2017 International Society for Advancement of Cytometry.

Engineering the future of 3D pathology.

In recent years, technological advances in tissue preparation, high-throughput volumetric microscopy, and computational infrastructure have enabled rapid developments in nondestructive 3D pathology, in which high-resolution histologic datasets are obtained from thick tissue specimens, such as whole biopsies, without the need for physical sectioning onto glass slides. While 3D pathology generates massive datasets that are attractive for automated computational analysis, there is also a desire to use 3D pathology to improve the visual assessment of tissue histology. In this perspective, we discuss and provide examples of potential advantages of 3D pathology for the visual assessment of clinical specimens and the challenges of dealing with large 3D datasets (of individual or multiple specimens) that pathologists have not been trained to interpret. We discuss the need for artificial intelligence triaging algorithms and explainable analysis methods to assist pathologists or other domain experts in the interpretation of these novel, often complex, large datasets.

CohortFinder: an open-source tool for data-driven partitioning of digital pathology and imaging cohorts to yield robust machine-learning models.

Batch effects (BEs) refer to systematic technical differences in data collection unrelated to biological variations whose noise is shown to negatively impact machine learning (ML) model generalizability. Here we release CohortFinder (http://cohortfinder.com), an open-source tool aimed at mitigating BEs via data-driven cohort partitioning. We demonstrate CohortFinder improves ML model performance in downstream digital pathology and medical image processing tasks. CohortFinder is freely available for download at cohortfinder.com.

PatchSorter: a high throughput deep learning digital pathology tool for object labeling.

The discovery of patterns associated with diagnosis, prognosis, and therapy response in digital pathology images often requires intractable labeling of large quantities of histological objects. Here we release an open-source labeling tool, PatchSorter, which integrates deep learning with an intuitive web interface. Using >100,000 objects, we demonstrate a >7x improvement in labels per second over unaided labeling, with minimal impact on labeling accuracy, thus enabling high-throughput labeling of large datasets.

An end-to-end workflow for nondestructive 3D pathology.

Recent advances in 3D pathology offer the ability to image orders of magnitude more tissue than conventional pathology methods while also providing a volumetric context that is not achievable with 2D tissue sections, and all without requiring destructive tissue sectioning. Generating high-quality 3D pathology datasets on a consistent basis, however, is not trivial and requires careful attention to a series of details during tissue preparation, imaging and initial data processing, as well as iterative optimization of the entire process. Here, we provide an end-to-end procedure covering all aspects of a 3D pathology workflow (using light-sheet microscopy as an illustrative imaging platform) with sufficient detail to perform well-controlled preclinical and clinical studies. Although 3D pathology is compatible with diverse staining protocols and computationally generated color palettes for visual analysis, this protocol focuses on the use of a fluorescent analog of hematoxylin and eosin, which remains the most common stain used for gold-standard pathological reports. We present our guidelines for a broad range of end users (e.g., biologists, clinical researchers and engineers) in a simple format. The end-to-end workflow requires 3-6 d to complete, bearing in mind that data analysis may take longer.

Multi-site cross-organ calibrated deep learning (MuSClD): Automated diagnosis of non-melanoma skin cancer.

Although deep learning (DL) has demonstrated impressive diagnostic performance for a variety of computational pathology tasks, this performance often markedly deteriorates on whole slide images (WSI) generated at external test sites. This phenomenon is due in part to domain shift, wherein differences in test-site pre-analytical variables (e.g., slide scanner, staining procedure) result in WSI with notably different visual presentations compared to training data. To ameliorate pre-analytic variances, approaches such as CycleGAN can be used to calibrate visual properties of images between sites, with the intent of improving DL classifier generalizability. In this work, we present a new approach termed Multi-Site Cross-Organ Calibration based Deep Learning (MuSClD) that employs WSIs of an off-target organ for calibration created at the same site as the on-target organ, based off the assumption that cross-organ slides are subjected to a common set of pre-analytical sources of variance. We demonstrate that by using an off-target organ from the test site to calibrate training data, the domain shift between training and testing data can be mitigated. Importantly, this strategy uniquely guards against potential data leakage introduced during calibration, wherein information only available in the testing data is imparted on the training data. We evaluate MuSClD in the context of the automated diagnosis of non-melanoma skin cancer (NMSC). Specifically, we evaluated MuSClD for identifying and distinguishing (a) basal cell carcinoma (BCC), (b) in-situ squamous cell carcinomas (SCC-In Situ), and (c) invasive squamous cell carcinomas (SCC-Invasive), using an Australian (training, n = 85) and a Swiss (held-out testing, n = 352) cohort. Our experiments reveal that MuSCID reduces the Wasserstein distances between sites in terms of color, contrast, and brightness metrics, without imparting noticeable artifacts to training data. The NMSC-subtyping performance is statistically improved as a result of MuSCID in terms of one-vs. rest AUC: BCC (0.92 vs 0.87, p = 0.01), SCC-In Situ (0.87 vs 0.73, p = 0.15) and SCC-Invasive (0.92 vs 0.82, p = 1e-5). Compared to baseline NMSC-subtyping with no calibration, the internal validation results of MuSClD (BCC (0.98), SCC-In Situ (0.92), and SCC-Invasive (0.97)) suggest that while domain shift indeed degrades classification performance, our on-target calibration using off-target tissue can safely compensate for pre-analytical variabilities, while improving the robustness of the model.