Background: Biomarkers in colorectal cancer are scarce, especially for patients with Stage 2 disease. The aim of our study was to identify potential prognostic biomarkers from colorectal cancers using a novel combination of approaches, whereby digital pathology is coupled to shotgun proteomics followed by validation of candidates by immunohistochemistry (IHC) using digital image analysis (DIA). Methods and Results: Tissue cores were punched from formalin-fixed paraffin-embedded colorectal cancers from patients with Stage 2 and 3 disease (n = 26, each). Protein extraction and liquid chromatography-mass spectrometry (MS) followed by analysis using three different methods were performed. Fold changes were evaluated. The candidate biomarker was validated by IHC on a series of 413 colorectal cancers from surgically treated patients using a next-generation tissue microarray. DIA was performed by using a pan-cytokeratin serial alignment and quantifying staining within the tumor and normal tissue epithelium. Analysis was done in QuPath and Brightness_Max scores were used for statistical analysis and clinicopathological associations. MS identified 1947 proteins with at least two unique peptides. To reinforce the validity of the biomarker candidates, only proteins showing a significant (P < 0.05) fold-change using all three analysis methods were considered. Eight were identified, and of these, cathepsin B was selected for further validation. DIA revealed strong associations between higher cathepsin B expression and less aggressive tumor features, including tumor node metastasis stage and lymphatic vessel and venous vessel invasion (P < 0.001, all). Cathepsin B was associated with more favorable survival in univariate analysis only. Conclusions: Our results present a novel approach to biomarker discovery that includes MS and digital pathology. Cathepsin B expression analyzed by DIA within the tumor epithelial compartment was identified as a strong feature of less aggressive tumor behavior and favorable outcome, a finding that should be further investigated on a more functional level.

Background: Digital pathology enables remote access or consults and powerful image analysis algorithms. However, the slide digitization process can create artifacts such as out-of-focus (OOF). OOF is often only detected on careful review, potentially causing rescanning, and workflow delays. Although scan time operator screening for whole-slide OOF is feasible, manual screening for OOF affecting only parts of a slide is impractical. Methods: We developed a convolutional neural network (ConvFocus) to exhaustively localize and quantify the severity of OOF regions on digitized slides. ConvFocus was developed using our refined semi-synthetic OOF data generation process and evaluated using seven slides spanning three different tissue and three different stain types, each of which were digitized using two different whole-slide scanner models ConvFocus's predictions were compared with pathologist-annotated focus quality grades across 514 distinct regions representing 37,700 35 μm × 35 μm image patches, and 21 digitized “z-stack” WSIs that contain known OOF patterns. Results: When compared to pathologist-graded focus quality, ConvFocus achieved Spearman rank coefficients of 0.81 and 0.94 on two scanners and reproduced the expected OOF patterns from z-stack scanning. We also evaluated the impact of OOF on the accuracy of a state-of-the-art metastatic breast cancer detector and saw a consistent decrease in performance with increasing OOF. Conclusions: Comprehensive whole-slide OOF categorization could enable rescans before pathologist review, potentially reducing the impact of digitization focus issues on the clinical workflow. We show that the algorithm trained on our semi-synthetic OOF data generalizes well to real OOF regions across tissue types, stains, and scanners. Finally, quantitative OOF maps can flag regions that might otherwise be misclassified by image analysis algorithms, preventing OOF-induced errors.

Background: Grading of prostatic adenocarcinoma is based on the Gleason scoring system and the more recently established prognostic grade groups. Typically, prostate cancer grading is performed by pathologists based on the morphology of the tumor on hematoxylin and eosin (H and E) slides. In this study, we investigated the histopathological image features with various survival models and attempted to study their correlations. Methods: Three texture methods (speeded-up robust features, histogram of oriented gradient, and local binary pattern) and two convolutional neural network (CNN)-based methods were applied to quantify histopathological image features. Five survival models were assessed on those image features in the context with other prostate clinical prognostic factors, including primary and secondary Gleason patterns, prostate-specific antigen levels, age, and clinical tumor stages. Results: Based on statistical comparisons among different image features with survival models, image features from CNN-based method with a recurrent neural network called CNN-long-short-term memory provided the highest hazard ratio of prostate cancer recurrence under Cox regression with an elastic net penalty. Conclusions: This approach outperformed the other image quantification methods listed above. Using this approach, patient outcomes were highly correlated with the histopathological image features of the tissue samples. In future studies, we plan to investigate the potential use of this approach for predicting recurrence in a wider range of cancer types.

Background: Calculated panel reactive antibody (cPRA) scoring is used to assess whether platelet refractoriness is mediated by human leukocyte antigen (HLA) antibodies in the recipient. cPRA testing uses a national sample of US kidney donors to estimate the population frequency of HLA antigens, which may be different than HLA frequencies within local platelet inventories. We aimed to determine the impact on patient cPRA scores of using HLA frequencies derived from typing local platelet donations rather than national HLA frequencies. Methods: We built an open-source web service to calculate cPRA scores based on national frequencies or custom-derived frequencies. We calculated cPRA scores for every hematopoietic stem cell transplantation (HSCT) patient at our institution based on the United Network for Organ Sharing (UNOS) frequencies and local frequencies. We compared frequencies and correlations between the calculators, segmented by gender. Finally, we put all scores into three buckets (mild, moderate, and high sensitizations) and looked at intergroup movement. Results: 2531 patients that underwent HSCT at our institution had at least 1 antibody and were included in the analysis. Overall, the difference in medians between each group's UNOS cPRA and local cPRA was statistically significant, but highly correlated (UNOS vs. local total: 0.249 and 0.243, ρ = 0.994; UNOS vs. local female: 0.474 and 0.463, ρ = 0.987, UNOS vs. local male: 0.165 and 0.141, ρ = 0.996;P< 0.001 for all comparisons). The median difference between UNOS and cPRA scores for all patients was low (male: 0.014, interquartile range [IQR]: 0.004–0.029; female: 0.0013, IQR: 0.003–0.028). Placement of patients into three groups revealed little intergroup movement, with 2.96% (75/2531) of patients differentially classified. Conclusions: cPRA scores using local frequencies were modestly but significantly different than those obtained using national HLA frequencies. We released our software as open source, so other groups can calculate cPRA scores from national or custom-derived frequencies. Further investigation is needed to determine whether a local-HLA frequency approach can improve outcomes in patients who are immune-refractory to platelets.

Objectives: Barcode-driven workflows reduce patient identification errors. Missing process timestamp data frequently confound our health system's pending lists and appear as actions left undone. Anecdotally, it was noted that missing data could be found when there is procedure noncompliance. This project was developed to determine if missing timestamp data in the histology barcode drive workflow correlated with other process variations, procedure noncompliance, or is an indicator of workflows needing focus for improvement projects.Materials and Methods: Data extracts of timestamp data from January 1, 2018, to December 15, 2018 for the major histology process steps were analyzed for missing data. Case level analysis to determine the presence or absence of expected barcoding events was performed on 1031 surgical pathology cases to determine the cause of the missing data and determine if additional data variations or procedure noncompliance events were present. The data variations were classified according to a scheme defined in the study. Results: Of 70,085, there were 7218 cases (10.3%) with missing process timestamp data. Missing histology process step data was associated with other additional data variations in case-level deep dives (P < 0.0001). Of the cases missing timestamp data in the initial review, 18.4% of the cases had no identifiable cause for the missing data (all expected events took place in the case-level deep dive). Conclusions: Operationally, valuable information can be obtained by reviewing the types and causes of missing data in the anatomic pathology laboratory information system, but only in conjunction with user input and feedback.

Background: Tumor programmed death-ligand 1 (PD-L1) status is useful in determining which patients may benefit from programmed death-1 (PD-1)/PD-L1 inhibitors. However, little is known about the association between PD-L1 status and tumor histopathological patterns. Using deep learning, we predicted PD-L1 status from hematoxylin and eosin (H and E) whole-slide images (WSIs) of nonsmall cell lung cancer (NSCLC) tumor samples. Materials and Methods: One hundred and thirty NSCLC patients were randomly assigned to training (n = 48) or test (n = 82) cohorts. A pair of H and E and PD-L1-immunostained WSIs was obtained for each patient. A pathologist annotated PD-L1 positive and negative tumor regions on the training samples using immunostained WSIs for reference. From the H and E WSIs, over 145,000 training tiles were generated and used to train a multi-field-of-view deep learning model with a residual neural network backbone. Results: The trained model accurately predicted tumor PD-L1 status on the held-out test cohort of H and E WSIs, which was balanced for PD-L1 status (area under the receiver operating characteristic curve [AUC] =0.80, P << 0.01). The model remained effective over a range of PD-L1 cutoff thresholds (AUC = 0.67–0.81, P ≤ 0.01) and when different proportions of the labels were randomly shuffled to simulate interpathologist disagreement (AUC = 0.63–0.77, P ≤ 0.03). Conclusions: A robust deep learning model was developed to predict tumor PD-L1 status from H and E WSIs in NSCLC. These results suggest that PD-L1 expression is correlated with the morphological features of the tumor microenvironment.

Objective: Digital pathology is today a widely used technology, and the digitalization of microscopic slides into whole slide images (WSIs) allows the use of machine learning algorithms as a tool in the diagnostic process. In recent years, “deep learning” algorithms for image analysis have been applied to digital pathology with great success. The training of these algorithms requires a large volume of high-quality images and image annotations. These large image collections are a potent source of information, and to use and share the information, standardization of the content through a consistent terminology is essential. The aim of this project was to develop a pilot dataset of exhaustive annotated WSI of normal and abnormal human tissue and link the annotations to appropriate ontological information. Materials and Methods: Several biomedical ontologies and controlled vocabularies were investigated with the aim of selecting the most suitable ontology for this project. The selection criteria required an ontology that covered anatomical locations, histological subcompartments, histopathologic diagnoses, histopathologic terms, and generic terms such as normal, abnormal, and artifact. WSIs of normal and abnormal tissue from 50 colon resections and 69 skin excisions, diagnosed 2015-2016 at the Department of Clinical Pathology in Linköping, were randomly collected. These images were manually and exhaustively annotated at the level of major subcompartments, including normal or abnormal findings and artifacts. Results: Systemized nomenclature of medicine clinical terms (SNOMED CT) was chosen, and the annotations were linked to its codes and terms. Two hundred WSI were collected and annotated, resulting in 17,497 annotations, covering a total area of 302.19 cm², equivalent to 107,7 gigapixels. Ninety-five unique SNOMED CT codes were used. The time taken to annotate a WSI varied from 45 s to over 360 min, a total time of approximately 360 h. Conclusion: This work resulted in a dataset of 200 exhaustive annotated WSIs of normal and abnormal tissue from the colon and skin, and it has informed plans to build a comprehensive library of annotated WSIs. SNOMED CT was found to be the best ontology for annotation labeling. This project also demonstrates the need for future development of annotation tools in order to make the annotation process more efficient.

Background: Online “e-modules” integrated into medical education may enhance traditional learning. Medical students use e-modules during clinical rotations, but these often lack histopathology correlates of diseases and minimal time is devoted to pathology teaching. To address this gap, we created pediatric pathology case-based e-modules to complement the clinical pediatric curriculum and enhance students' understanding of pediatric diseases. Methods: Philips Tutor is an interactive web-based program in which pediatric pathology e-modules were created with pre-/post-test questions. Each e-module contains a clinical vignette, virtual microscopy, and links to additional resources. Topics were selected based on established learning objectives for pediatric clinical rotations. Pre- and post-tests were administered at the beginning/end of each rotation. Test group had access to the e-modules, but control group did not. Both groups completed the pre/post-tests. Posttest was followed by a feedback survey. Results: Overall, 7% (9/123) in the control group and 8% (13/164) in the test group completed both tests and were included in the analysis. Test group improved their posttest scores by about one point on a 5-point scale (P = 0.01); control group did not (P = 1.00). Students responded that test questions were helpful in assessing their knowledge of pediatric pathology (90%) and experienced relative ease of use with the technology (80%). Conclusions: Students responded favorably to the new technology, but cited time constraints as a significant barrier to study participation. Access to the e-modules suggested an improved posttest score compared to the control group, but pilot data were limited by the small sample size. Incorporating pediatric case-based e-modules with anatomic and clinical pathology topics into the clinical medical education curriculum may heighten students' understanding of important diseases. Our model may serve as a pilot for other medical education platforms.

Background: Professionals and trainees in the medical and scientific fields may receive high e-mail volumes for conferences and journals. In this report, we analyze the amount and characteristics of unsolicited e-mails for journals, conferences, and webinars received by faculty and trainees in a pathology department at an academic medical center. Methods: With informed consent, we analyzed 7 consecutive days of e-mails from faculty and trainees who voluntarily participated in the study and saved unsolicited e-mails from their institutional e-mail address (including junk e-mail folder) for medical/scientific journals, conferences, and webinars. All e-mails were examined for characteristics such as reply receipts, domain name, and spam likelihood. Journal e-mails were specifically analyzed for claims in the message body (for example, peer review, indexing in databases/resources, rapid publication) and actual inclusion in recognized journal databases/resources. Results: A total of 17 faculty (4 assistant, 4 associate, and 9 full professors) and 9 trainees (5 medical students, 2 pathology residents, and 2 pathology fellows) completed the study. A total of 755 e-mails met study criteria (417 e-mails from 328 unique journals, 244 for conferences, and 94 for webinars). Overall, 44.4% of e-mails were flagged as potential spam by the institutional default settings, and 13.8% requested reply receipts. The highest burden of e-mails in 7 days was by associate and full professors (maximum 158 or approximately 8200 per year), although some trainees and assistant professors had over 30 e-mails in 7 days (approximately 1560 per year). Common characteristics of journal e-mails were mention of “peer review” in the message body and low rates of inclusion in recognized journal databases/resources, with 76.4% not found in any of 9 journal databases/resources. The location for conferences in e-mails included 31 different countries, with the most common being the United States (33.2%), Italy (9.8%), China (4.9%), United Kingdom (4.9%), and Canada (4.5%). Conclusions: The present study in an academic pathology department shows a high burden of unsolicited e-mails for medical/scientific journals, conferences, and webinars, especially to associate and full professors. We also demonstrate that some pathology trainees and junior faculty are receiving an estimated 1500 unsolicited e-mails per year.

Background: To assess reproducibility and accuracy of overall Nottingham grade and component scores using digital whole slide images (WSIs) compared to glass slides. Methods: Two hundred and eight pathologists were randomized to independently interpret 1 of 4 breast biopsy sets using either glass slides or digital WSI. Each set included 5 or 6 invasive carcinomas (22 total invasive cases). Participants interpreted the same biopsy set approximately 9 months later following a second randomization to WSI or glass slides. Nottingham grade, including component scores, was assessed on each interpretation, providing 2045 independent interpretations of grade. Overall grade and component scores were compared between pathologists (interobserver agreement) and for interpretations by the same pathologist (intraobserver agreement). Grade assessments were compared when the format (WSI vs. glass slides) changed or was the same for the two interpretations. Results: Nottingham grade intraobserver agreement was highest using glass slides for both interpretations (73%, 95% confidence interval [CI]: 68%, 78%) and slightly lower but not statistically different using digital WSI for both interpretations (68%, 95% CI: 61%, 75%; P= 0.22). The agreement was lowest when the format changed between interpretations (63%, 95% CI: 59%, 68%). Interobserver agreement was significantly higher (P < 0.001) using glass slides versus digital WSI (68%, 95% CI: 66%, 70% versus 60%, 95% CI: 57%, 62%, respectively). Nuclear pleomorphism scores had the lowest inter- and intra-observer agreement. Mitotic scores were higher on glass slides in inter- and intra-observer comparisons. Conclusions: Pathologists' intraobserver agreement (reproducibility) is similar for Nottingham grade using glass slides or WSI. However, slightly lower agreement between pathologists suggests that verification of grade using digital WSI may be more challenging.

Background: Proliferation markers, especially Ki67, are increasingly important in diagnosis and prognosis. The best method for calculating Ki67 is still the subject of debate. Materials and Methods: We evaluated an image analysis tool for quantitative interpretation of Ki67 in neuroendocrine tumors and compared it to manual counts. We expanded a primary digital pathology platform to include the Leica Biosystems image analysis nuclear algorithm. Slides were digitized using a Leica Aperio AT2 Scanner and accessed through the Cerner CoPath LIS interfaced with Aperio eSlideManager through Aperio ImageScope. Selected regions of interest (ROIs) were manually defined and annotated to include tumor cells only; they were then analyzed with the algorithm and by four pathologists counting on printed images. After validation, the algorithm was used to examine the impact of the size and number of areas selected as ROIs. Results: The algorithm provided reproducible results that were obtained within seconds, compared to up to 55 min of manual counting that varied between users. Benefits of image analysis identified by users included accuracy, time savings, and ease of viewing. Access to the algorithm allowed rapid comparisons of Ki67 counts in ROIs that varied in numbers of cells and selection of fields, the outputs demonstrated that the results vary around defined cutoffs that provide tumor grade depending on the number of cells and ROIs counted. Conclusions: Digital image analysis provides accurate and reproducible quantitative data faster than manual counts. However, access to this tool allows multiple analyses of a single sample to use variable numbers of cells and selection of variable ROIs that can alter the result in clinically significant ways. This study highlights the potential risk of hard cutoffs of continuous variables and indicates that standardization of number of cells and number of regions selected for analysis should be incorporated into guidelines for Ki67 calculations.

Background: Accurate diagnosis of melanocytic lesions is challenging, even for expert pathologists. Nowadays, whole-slide imaging (WSI) is used for routine clinical pathology diagnosis in several laboratories. One of the limitations of WSI, as it is most often used, is the lack of a multiplanar focusing option. In this study, we aim to establish the diagnostic accuracy of WSI for melanocytic lesions and investigate the potential accuracy increase of z-stack scanning. Z-stack enables pathologists to use a software focus adjustment, comparable to the fine-focus knob of a conventional light microscope. Materials and Methods: Melanocytic lesions (n = 102) were selected from our pathology archives: 35 nevi, 5 spitzoid tumors of unknown malignant potential, and 62 malignant melanomas, including 10 nevoid melanomas. All slides were scanned at a magnification comparable to use of a ×40 objective, in z-stack mode. A ground truth diagnosis was established on the glass slides by four academic dermatopathologists with a special interest in the diagnosis of melanoma. Six nonacademic surgical pathologists subspecialized in dermatopathology examined the cases by WSI. Results: An expert consensus diagnosis was achieved in 99 (97%) of cases. Concordance rates between surgical pathologists and the ground truth varied between 75% and 90%, excluding nevoid melanoma cases. Concordance rates of nevoid melanoma varied between 10% and 80%. Pathologists used the software focusing option in 7%–28% of cases, which in 1 case of nevoid melanoma resulted in correcting a misdiagnosis after finding a dermal mitosis. Conclusion: Diagnostic accuracy of melanocytic lesions based on glass slides and WSI is comparable with previous publications. A large variability in diagnostic accuracy of nevoid melanoma does exist. Our results show that z-stack scanning, in general, does not increase the diagnostic accuracy of melanocytic.

Whole-slide images (WSIs) are a rich new source of biomedical imaging data. The use of automated systems to classify and segment WSIs has recently come to forefront of the pathology research community. While digital slides have obvious educational and clinical uses, their most exciting potential lies in the application of quantitative computational tools to automate search tasks, assist in classic diagnostic classification tasks, and improve prognosis and theranostics. An essential step in enabling these advancements is to apply advances in machine learning and artificial intelligence from other fields to previously inaccessible pathology datasets, thereby enabling the application of new technologies to solve persistent diagnostic challenges in pathology. Here, we applied convolutional neural networks to differentiate between two forms of melanocytic lesions (Spitz and conventional). Classification accuracy at the patch level was 99.0%–2% when applied to WSI. Importantly, when the model was trained without careful image curation by a pathologist, the training took significantly longer and had lower overall performance. These results highlight the utility of augmented human intelligence in digital pathology applications, and the critical role pathologists will play in the evolution of computational pathology algorithms.