A Randomized Study Comparing Digital Imaging to Traditional Glass Slide Microscopy for Breast Biopsy and Cancer Diagnosis

Digital pathology for reporting histopathology samples, including cancer screening samples - definitive evidence from a multisite study

AIMS

To conduct a definitive multicentre comparison of digital pathology (DP) with light microscopy (LM) for reporting histopathology slides including breast and bowel cancer screening samples.

METHODS

A total of 2024 cases (608 breast, 607 GI, 609 skin, 200 renal) were studied, including 207 breast and 250 bowel cancer screening samples. Cases were examined by four pathologists (16 study pathologists across the four speciality groups), using both LM and DP, with the order randomly assigned and 6 weeks between viewings. Reports were compared for clinical management concordance (CMC), meaning identical diagnoses plus differences which do not affect patient management. Percentage CMCs were computed using logistic regression models with crossed random-effects terms for case and pathologist. The obtained percentage CMCs were referenced to 98.3% calculated from previous studies.

RESULTS

For all cases LM versus DP comparisons showed the CMC rates were 99.95% [95% confidence interval (CI) = 99.90-99.97] and 98.96 (95% CI = 98.42-99.32) for cancer screening samples. In speciality groups CMC for LM versus DP showed: breast 99.40% (99.06-99.62) overall and 96.27% (94.63-97.43) for cancer screening samples; [gastrointestinal (GI) = 99.96% (99.89-99.99)] overall and 99.93% (99.68-99.98) for bowel cancer screening samples; skin 99.99% (99.92-100.0); renal 99.99% (99.57-100.0). Analysis of clinically significant differences revealed discrepancies in areas where interobserver variability is known to be high, in reads performed with both modalities and without apparent trends to either.

CONCLUSIONS

Comparing LM and DP CMC, overall rates exceed the reference 98.3%, providing compelling evidence that pathologists provide equivalent results for both routine and cancer screening samples irrespective of the modality used.

An interpretable decision-support model for breast cancer diagnosis using histopathology images

Microscopic examination of biopsy tissue slides is perceived as the gold-standard methodology for the confirmation of presence of cancer cells. Manual analysis of an overwhelming inflow of tissue slides is highly susceptible to misreading of tissue slides by pathologists. A computerized framework for histopathology image analysis is conceived as a diagnostic tool that greatly benefits pathologists, augmenting definitive diagnosis of cancer. Convolutional Neural Network (CNN) turned out to be the most adaptable and effective technique in the detection of abnormal pathologic histology. Despite their high sensitivity and predictive power, clinical translation is constrained by a lack of intelligible insights into the prediction. A computer-aided system that can offer a definitive diagnosis and interpretability is therefore highly desirable. Conventional visual explanatory techniques, Class Activation Mapping (CAM), combined with CNN models offers interpretable decision making. The major challenge in CAM is, it cannot be optimized to create the best visualization map. CAM also decreases the performance of the CNN models. To address this challenge, we introduce a novel interpretable decision-support model using CNN with a trainable attention mechanism using response-based feed-forward visual explanation. We introduce a variant of DarkNet19 CNN model for the classification of histopathology images. In order to achieve visual interpretation as well as boost the performance of the DarkNet19 model, an attention branch is integrated with DarkNet19 network forming Attention Branch Network (ABN). The attention branch uses a convolution layer of DarkNet19 and Global Average Pooling (GAP) to model the context of the visual features and generate a heatmap to identify the region of interest. Finally, the perception branch is constituted using a fully connected layer to classify images. We trained and validated our model using more than 7000 breast cancer biopsy slide images from an openly available dataset and achieved 98.7% accuracy in the binary classification of histopathology images. The observations substantiated the enhanced clinical interpretability of the DarkNet19 CNN model, supervened by the attention branch, besides delivering a 3%-4% performance boost of the baseline model. The cancer regions highlighted by the proposed model correlate well with the findings of an expert pathologist. The coalesced approach of unifying attention branch with the CNN model capacitates pathologists with augmented diagnostic interpretability of histological images with no detriment to state-of-art performance. The model's proficiency in pinpointing the region of interest is an added bonus that can lead to accurate clinical translation of deep learning models that underscore clinical decision support.

Pathologist pupil dilation reflects experience level and difficulty in diagnosing medical images

Purpose: Digital whole slide imaging allows pathologists to view slides on a computer screen instead of under a microscope. Digital viewing allows for real-time monitoring of pathologists' search behavior and neurophysiological responses during the diagnostic process. One particular neurophysiological measure, pupil diameter, could provide a basis for evaluating clinical competence during training or developing tools that support the diagnostic process. Prior research shows that pupil diameter is sensitive to cognitive load and arousal, and it switches between exploration and exploitation of a visual image. Different categories of lesions in pathology pose different levels of challenge, as indicated by diagnostic disagreement among pathologists. If pupil diameter is sensitive to the perceived difficulty in diagnosing biopsies, eye-tracking could potentially be used to identify biopsies that may benefit from a second opinion. Approach: We measured case onset baseline-corrected (phasic) and uncorrected (tonic) pupil diameter in 90 pathologists who each viewed and diagnosed 14 digital breast biopsy cases that cover the diagnostic spectrum from benign to invasive breast cancer. Pupil data were extracted from the beginning of viewing and interpreting of each individual case. After removing 122 trials ( < 10 % ) with poor eye-tracking quality, 1138 trials remained. We used multiple linear regression with robust standard error estimates to account for dependent observations within pathologists. Results: We found a positive association between the magnitude of phasic dilation and subject-centered difficulty ratings and between the magnitude of tonic dilation and untransformed difficulty ratings. When controlling for case diagnostic category, only the tonic-difficulty relationship persisted. Conclusions: Results suggest that tonic pupil dilation may indicate overall arousal differences between pathologists as they interpret biopsy cases and could signal a need for additional training, experience, or automated decision aids. Phasic dilation is sensitive to characteristics of biopsies that tend to elicit higher difficulty ratings and could indicate a need for a second opinion.

End-to-End Diagnosis of Breast Biopsy Images with Transformers

Diagnostic disagreements among pathologists occur throughout the spectrum of benign to malignant lesions. A computer-aided diagnostic system capable of reducing uncertainties would have important clinical impact. To develop a computer-aided diagnosis method for classifying breast biopsy images into a range of diagnostic categories (benign, atypia, ductal carcinoma in situ, and invasive breast cancer), we introduce a transformer-based hollistic attention network called HATNet. Unlike state-of-the-art histopathological image classification systems that use a two pronged approach, i.e., they first learn local representations using a multi-instance learning framework and then combine these local representations to produce image-level decisions, HATNet streamlines the histopathological image classification pipeline and shows how to learn representations from gigapixel size images end-to-end. HATNet extends the bag-of-words approach and uses self-attention to encode global information, allowing it to learn representations from clinically relevant tissue structures without any explicit supervision. It outperforms the previous best network Y-Net, which uses supervision in the form of tissue-level segmentation masks, by 8%. Importantly, our analysis reveals that HATNet learns representations from clinically relevant structures, and it matches the classification accuracy of 87 U.S. pathologists for this challenging test set.

DigiPatICS: Digital Pathology Transformation of the Catalan Health Institute Network of 8 Hospitals—Planification, Implementation, and Preliminary Results

Complete digital pathology transformation for primary histopathological diagnosis is a challenging yet rewarding endeavor. Its advantages are clear with more efficient workflows, but there are many technical and functional difficulties to be faced. The Catalan Health Institute (ICS) has started its DigiPatICS project, aiming to deploy digital pathology in an integrative, holistic, and comprehensive way within a network of 8 hospitals, over 168 pathologists, and over 1 million slides each year. We describe the bidding process and the careful planning that was required, followed by swift implementation in stages. The purpose of the DigiPatICS project is to increase patient safety and quality of care, improving diagnosis and the efficiency of processes in the pathological anatomy departments of the ICS through process improvement, digital pathology, and artificial intelligence tools.

More scanning, but not zooming, is associated with diagnostic accuracy in evaluating digital breast pathology slides

Diagnoses of medical images can invite strikingly diverse strategies for image navigation and visual search. In computed tomography screening for lung nodules, distinct strategies, termed scanning and drilling, relate to both radiologists' clinical experience and accuracy in lesion detection. Here, we examined associations between search patterns and accuracy for pathologists (N = 92) interpreting a diverse set of breast biopsy images. While changes in depth in volumetric images reveal new structures through movement in the z-plane, in digital pathology changes in depth are associated with increased magnification. Thus, "drilling" in radiology may be more appropriately termed "zooming" in pathology. We monitored eye-movements and navigation through digital pathology slides to derive metrics of how quickly the pathologists moved through XY (scanning) and Z (zooming) space. Prior research on eye-movements in depth has categorized clinicians as either "scanners" or "drillers." In contrast, we found that there was no reliable association between a clinician's tendency to scan or zoom while examining digital pathology slides. Thus, in the current work we treated scanning and zooming as continuous predictors rather than categorizing as either a "scanner" or "zoomer." In contrast to prior work in volumetric chest images, we found significant associations between accuracy and scanning rate but not zooming rate. These findings suggest fundamental differences in the relative value of information types and review behaviors across two image formats. Our data suggest that pathologists gather critical information by scanning on a given plane of depth, whereas radiologists drill through depth to interrogate critical features.

Validation of digital microscopy: Review of validation methods and sources of bias

Digital microscopy (DM) is increasingly replacing traditional light microscopy (LM) for performing routine diagnostic and research work in human and veterinary pathology. The DM workflow encompasses specimen preparation, whole-slide image acquisition, slide retrieval, and the workstation, each of which has the potential (depending on the technical parameters) to introduce limitations and artifacts into microscopic examination by pathologists. Performing validation studies according to guidelines established in human pathology ensures that the best-practice approaches for patient care are not deteriorated by implementing DM. Whereas current publications on validation studies suggest an overall high reliability of DM, each laboratory is encouraged to perform an individual validation study to ensure that the DM workflow performs as expected in the respective clinical or research environment. With the exception of validation guidelines developed by the College of American Pathologists in 2013 and its update in 2021, there is no current review of the application of methods fundamental to validation. We highlight that there is high methodological variation between published validation studies, each having advantages and limitations. The diagnostic concordance rate between DM and LM is the most relevant outcome measure, which is influenced (regardless of the viewing modality used) by different sources of bias including complexity of the cases examined, diagnostic experience of the study pathologists, and case recall. Here, we review 3 general study designs used for previous publications on DM validation as well as different approaches for avoiding bias.

Integrated digital pathology at scale: A solution for clinical diagnostics and cancer research at a large academic medical center

OBJECTIVE

Broad adoption of digital pathology (DP) is still lacking, and examples for DP connecting diagnostic, research, and educational use cases are missing. We blueprint a holistic DP solution at a large academic medical center ubiquitously integrated into clinical workflows; researchapplications including molecular, genetic, and tissue databases; and educational processes.

MATERIALS AND METHODS

We built a vendor-agnostic, integrated viewer for reviewing, annotating, sharing, and quality assurance of digital slides in a clinical or research context. It is the first homegrown viewer cleared by New York State provisional approval in 2020 for primary diagnosis and remote sign-out during the COVID-19 (coronavirus disease 2019) pandemic. We further introduce an interconnected Honest Broker for BioInformatics Technology (HoBBIT) to systematically compile and share large-scale DP research datasets including anonymized images, redacted pathology reports, and clinical data of patients with consent.

RESULTS

The solution has been operationally used over 3 years by 926 pathologists and researchers evaluating 288 903 digital slides. A total of 51% of these were reviewed within 1 month after scanning. Seamless integration of the viewer into 4 hospital systems clearly increases the adoption of DP. HoBBIT directly impacts the translation of knowledge in pathology into effective new health measures, including artificial intelligence-driven detection models for prostate cancer, basal cell carcinoma, and breast cancer metastases, developed and validated on thousands of cases.

CONCLUSIONS

We highlight major challenges and lessons learned when going digital to provide orientation for other pathologists. Building interconnected solutions will not only increase adoption of DP, but also facilitate next-generation computational pathology at scale for enhanced cancer research.

Diagnostic concordance and discordance in digital pathology: a systematic review and meta-analysis

BACKGROUND

Digital pathology (DP) has the potential to fundamentally change the way that histopathology is practised, by streamlining the workflow, increasing efficiency, improving diagnostic accuracy and facilitating the platform for implementation of artificial intelligence-based computer-assisted diagnostics. Although the barriers to wider adoption of DP have been multifactorial, limited evidence of reliability has been a significant contributor. A meta-analysis to demonstrate the combined accuracy and reliability of DP is still lacking in the literature.

OBJECTIVES

We aimed to review the published literature on the diagnostic use of DP and to synthesise a statistically pooled evidence on safety and reliability of DP for routine diagnosis (primary and secondary) in the context of validation process.

METHODS

A comprehensive literature search was conducted through PubMed, Medline, EMBASE, Cochrane Library and Google Scholar for studies published between 2013 and August 2019. The search protocol identified all studies comparing DP with light microscopy (LM) reporting for diagnostic purposes, predominantly including H&E-stained slides. Random-effects meta-analysis was used to pool evidence from the studies.

RESULTS

Twenty-five studies were deemed eligible to be included in the review which examined a total of 10 410 histology samples (average sample size 176). For overall concordance (clinical concordance), the agreement percentage was 98.3% (95% CI 97.4 to 98.9) across 24 studies. A total of 546 major discordances were reported across 25 studies. Over half (57%) of these were related to assessment of nuclear atypia, grading of dysplasia and malignancy. These were followed by challenging diagnoses (26%) and identification of small objects (16%).

CONCLUSION

The results of this meta-analysis indicate equivalent performance of DP in comparison with LM for routine diagnosis. Furthermore, the results provide valuable information concerning the areas of diagnostic discrepancy which may warrant particular attention in the transition to DP.

Deep Feature Representations for Variable-Sized Regions of Interest in Breast Histopathology

OBJECTIVE

Modeling variable-sized regions of interest (ROIs) in whole slide images using deep convolutional networks is a challenging task, as these networks typically require fixed-sized inputs that should contain sufficient structural and contextual information for classification. We propose a deep feature extraction framework that builds an ROI-level feature representation via weighted aggregation of the representations of variable numbers of fixed-sized patches sampled from nuclei-dense regions in breast histopathology images.

METHODS

First, the initial patch-level feature representations are extracted from both fully-connected layer activations and pixel-level convolutional layer activations of a deep network, and the weights are obtained from the class predictions of the same network trained on patch samples. Then, the final patch-level feature representations are computed by concatenation of weighted instances of the extracted feature activations. Finally, the ROI-level representation is obtained by fusion of the patch-level representations by average pooling.

RESULTS

Experiments using a well-characterized data set of 240 slides containing 437 ROIs marked by experienced pathologists with variable sizes and shapes result in an accuracy score of 72.65% in classifying ROIs into four diagnostic categories that cover the whole histologic spectrum.

CONCLUSION

The results show that the proposed feature representations are superior to existing approaches and provide accuracies that are higher than the average accuracy of another set of pathologists.

SIGNIFICANCE

The proposed generic representation that can be extracted from any type of deep convolutional architecture combines the patch appearance information captured by the network activations and the diagnostic relevance predicted by the class-specific scoring of patches for effective modeling of variable-sized ROIs.

Telepathology Practice in Cancer Diagnosis in Saint Jean de Dieu Hospital - Tanguieta, Benin

CONTEXT.—

Both the incidence of cancer and cancer-related mortality rates are high in sub-Saharan Africa, while resources for diagnosis and management are inadequate. In Benin, there is an extreme shortage of pathology services. Because of this shortage we built a histopathology laboratory equipped with an automated immunohistochemistry and a whole-slide imaging and telepathology system.

OBJECTIVE.—

To report our experience of telepathology practice in the improvement of cancer diagnosis.

DESIGN.—

The study was performed in our histopathology laboratory from January 1, 2016, to December 31, 2018. Resident laboratory technicians were trained in the preparation of microscopic and virtual slides by European pathologists. Virtual slides were stored on a Web-accessible server area for reading by 21 telepathologists in Benin and Europe. All patients with a histologic diagnosis of cancer were included in this study. Demographic data of patients, anatomic site of cancer, its histologic type, and its histologic grade were recorded.

RESULTS.—

We registered 399 patients diagnosed with cancer of 1593 patients whose surgical specimens had been analyzed. There were 349 adults including 160 males and 189 females, and 50 children (both sexes) with a mean age of 53.40 years, 46.92 years, and 9.72 years, respectively. Eighty-three of 211 females (39.34%) had infiltrating breast carcinoma, and 34 of 188 males (18.09%) had prostatic carcinoma. Infiltrating carcinoma of no special type represented 51 (91.07%) of all infiltrating breast carcinomas. Prostatic carcinoma and infiltrating breast carcinoma were of high grade in 13 of 23 males (56.52%) and 34 of 56 females (60.71%), respectively.

CONCLUSIONS.—

Telepathology is enabling a great improvement in cancer diagnosis in our hospital.

Assessment of Machine Learning of Breast Pathology Structures for Automated Differentiation of Breast Cancer and High-Risk Proliferative Lesions

IMPORTANCE

Following recent US Food and Drug Administration approval, adoption of whole slide imaging in clinical settings may be imminent, and diagnostic accuracy, particularly among challenging breast biopsy specimens, may benefit from computerized diagnostic support tools.

OBJECTIVE

To develop and evaluate computer vision methods to assist pathologists in diagnosing the full spectrum of breast biopsy samples, from benign to invasive cancer.

DESIGN, SETTING, AND PARTICIPANTS

In this diagnostic study, 240 breast biopsies from Breast Cancer Surveillance Consortium registries that varied by breast density, diagnosis, patient age, and biopsy type were selected, reviewed, and categorized by 3 expert pathologists as benign, atypia, ductal carcinoma in situ (DCIS), and invasive cancer. The atypia and DCIS cases were oversampled to increase statistical power. High-resolution digital slide images were obtained, and 2 automated image features (tissue distribution feature and structure feature) were developed and evaluated according to the consensus diagnosis of the expert panel. The performance of the automated image analysis methods was compared with independent interpretations from 87 practicing US pathologists. Data analysis was performed between February 2017 and February 2019.

MAIN OUTCOMES AND MEASURES

Diagnostic accuracy defined by consensus reference standard of 3 experienced breast pathologists.

RESULTS

The accuracy of machine learning tissue distribution features, structure features, and pathologists for classification of invasive cancer vs noninvasive cancer was 0.94, 0.91, and 0.98, respectively; the accuracy of classification of atypia and DCIS vs benign tissue was 0.70, 0.70, and 0.81, respectively; and the accuracy of classification of DCIS vs atypia was 0.83, 0.85, and 0.80, respectively. The sensitivity of both machine learning features was lower than that of the pathologists for the invasive vs noninvasive classification (tissue distribution feature, 0.70; structure feature, 0.49; pathologists, 0.84) but higher for the classification of atypia and DCIS vs benign cases (tissue distribution feature, 0.79; structure feature, 0.85; pathologists, 0.72) and the classification of DCIS vs atypia (tissue distribution feature, 0.88; structure feature, 0.89; pathologists, 0.70). For the DCIS vs atypia classification, the specificity of the machine learning feature classification was similar to that of the pathologists (tissue distribution feature, 0.78; structure feature, 0.80; pathologists, 0.82).

CONCLUSION AND RELEVANCE

The computer-based automated approach to interpreting breast pathology showed promise, especially as a diagnostic aid in differentiating DCIS from atypical hyperplasia.

A validation study of whole slide imaging for primary diagnosis of lymphoma

Whole slide imaging (WSI) is being increasingly used worldwide. Although previous studies have asserted the validity of WSI diagnosis, they have primarily targeted only small specimens and excluded cases requiring immunohistochemistry or special staining, such as lymphoma. The purpose of this study was to evaluate the accuracy of WSI diagnosis of lymphoma, for which 240 biopsies and resections of lymphoma cases were selected from the study set of lymphomas. All slides including H&E, immunohistochemical and special staining were digitized using a WSI image scanner. An experienced pathologist performed the WSI diagnoses, which were compared with original diagnoses based on light microscopic examinations. Discrepancy between the two interpretations were classified into three categories: concordance, minor discrepancy (no clinical significance), and major discrepancy (with clinical significance). Overall concordance between the light microscopic and WSI diagnosis was found in 223 cases (92.92%; 95%CI = 88.90-95.82), minor discrepancy in fifteen (6.25%; 95%CI = 3.54-10.10), and major discrepancy in two (0.83%; 95%CI = 0.10-2.98). Diagnosis of lymphoma using WSI appeared to be mostly accurate, suggesting that WSI may be a reliable technology for the diagnosis of lymphoma.

Characterizing Diagnostic Search Patterns in Digital Breast Pathology: Scanners and Drillers

Following a baseline demographic survey, 87 pathologists interpreted 240 digital whole slide images of breast biopsy specimens representing a range of diagnostic categories from benign to atypia, ductal carcinoma in situ, and invasive cancer. A web-based viewer recorded pathologists' behaviors while interpreting a subset of 60 randomly selected and randomly ordered slides. To characterize diagnostic search patterns, we used the viewport location, time stamp, and zoom level data to calculate four variables: average zoom level, maximum zoom level, zoom level variance, and scanning percentage. Two distinct search strategies were confirmed: scanning is characterized by panning at a constant zoom level, while drilling involves zooming in and out at various locations. Statistical analysis was applied to examine the associations of different visual interpretive strategies with pathologist characteristics, diagnostic accuracy, and efficiency. We found that females scanned more than males, and age was positively correlated with scanning percentage, while the facility size was negatively correlated. Throughout 60 cases, the scanning percentage and total interpretation time per slide decreased, and these two variables were positively correlated. The scanning percentage was not predictive of diagnostic accuracy. Increasing average zoom level, maximum zoom level, and zoom variance were correlated with over-interpretation.

Breast Cancer Prognostic Factors in the Digital Era: Comparison of Nottingham Grade using Whole Slide Images and Glass Slides

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.

Reliability of deceased-donor procurement kidney biopsy images uploaded in United Network for Organ Sharing

Prior studies demonstrate poor agreement among pathologists' interpretation of kidney biopsy slides. Reliability of representative images of these slides uploaded to the United Network of Organ Sharing (UNOS) web portal for clinician review has not been studied. We hypothesized high agreement among pathologists' image interpretation, since static images eliminate variation induced by viewing different areas of movable slides. To test our hypothesis, we compared the assessments of UNOS-uploaded images recorded in standardized forms by three pathologists. We selected 100 image sets, each having at least two images from kidneys of deceased donors. Weighted Cohen's kappa was used for inter-rater agreement. Mean (SD) donor age was 50 (13). Acute tubular injury had kappas of 0.12, 0.14, and 0.19; arteriolar hyalinosis 0.16, 0.27, and 0.38; interstitial inflammation 0.30, 0.33, and 0.49; interstitial fibrosis 0.28, 0.32, and 0.67; arterial intimal fibrosis 0.34, 0.42, and 0.59; tubular atrophy 0.35, 0.41, and 0.52; glomeruli thrombi 0.32, 0.53, and 0.85; and global glomerulosclerosis 0.68, 0.70, and 0.77. Pathologists' agreement demonstrated kappas of 0.12 to 0.77. The lower values raise concern about the reliability of using images. Although further research is needed to understand how uploaded images are used clinically, the field may consider higher-quality standards for biopsy photomicrographs.

Detection and classification of cancer in whole slide breast histopathology images using deep convolutional networks

Generalizability of algorithms for binary cancer vs. no cancer classification is unknown for clinically more significant multi-class scenarios where intermediate categories have different risk factors and treatment strategies. We present a system that classifies whole slide images (WSI) of breast biopsies into five diagnostic categories. First, a saliency detector that uses a pipeline of four fully convolutional networks, trained with samples from records of pathologists' screenings, performs multi-scale localization of diagnostically relevant regions of interest in WSI. Then, a convolutional network, trained from consensus-derived reference samples, classifies image patches as non-proliferative or proliferative changes, atypical ductal hyperplasia, ductal carcinoma in situ, and invasive carcinoma. Finally, the saliency and classification maps are fused for pixel-wise labeling and slide-level categorization. Experiments using 240 WSI showed that both saliency detector and classifier networks performed better than competing algorithms, and the five-class slide-level accuracy of 55% was not statistically different from the predictions of 45 pathologists. We also present example visualizations of the learned representations for breast cancer diagnosis.

Accuracy of Digital Pathologic Analysis vs Traditional Microscopy in the Interpretation of Melanocytic Lesions

Importance

Use of digital whole-slide imaging (WSI) for dermatopathology in general has been noted to be similar to traditional microscopy (TM); however, concern has been noted that WSI is inferior for interpretation of melanocytic lesions. Since approximately 1 of every 4 skin biopsies is of a melanocytic lesion, the use of WSI requires verification before use in clinical practice.

Objective

To compare pathologists' accuracy and reproducibility in diagnosing melanocytic lesions using Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis (MPATH-Dx) categories when analyzing by TM vs WSI.

Design, Setting, and Participants

A total of 87 pathologists in community-based and academic settings from 10 US states were randomized with stratification based on clinical experience to interpret in TM format 180 skin biopsy cases of melanocytic lesions, including 90 invasive melanoma, divided into 5 sets of 36 cases (phase 1). The pathologists were then randomized via stratified permuted block randomization with block size 2 to interpret cases in either TM (n = 46) or WSI format (n = 41), with each pathologist interpreting the same 36 cases on 2 separate occasions (phase 2). Diagnoses were categorized as MPATH-Dx categories I through V, with I indicating the least severe and V the most severe.

Main Outcomes and Measures

Accuracy with respect to a consensus reference diagnosis and the reproducibility of repeated interpretations of the same cases.

Results

Of the 87 pathologists in the study, 46% (40) were women and the mean (SD) age was 50.7 (10.2) years. Except for class III melanocytic lesions, the diagnostic categories showed no significant differences in diagnostic accuracy between TM and WSI interpretation. Discordance was lower among class III lesions for the TM interpretation arm (51%; 95% CI, 46%-57%) than for the WSI arm (61%; 95% CI, 53%-69%) (P = .05). This difference is likely to have clinical significance, because 6% of TM vs 11% of WSI class III lesions were interpreted as invasive melanoma. Reproducibility was similar between the traditional and digital formats overall (66.4%; 95% CI, 63.3%-69.3%; and 62.7%; 95% CI, 59.5%-65.7%, respectively), and for all classes, although class III showed a nonsignificant lower intraobserver agreement for digital. Significantly more mitotic figures were detected with TM compared with WSI: mean (SD) TM, 6.72 (2.89); WSI, 5.84 (2.56); P = .002.

Conclusions and Relevance

Interpretive accuracy for melanocytic lesions was similar for WSI and TM slides except for class III lesions. We found no clinically meaningful differences in reproducibility for any of the diagnostic classes.

Multi-Instance Multi-Label Learning for Multi-Class Classification of Whole Slide Breast Histopathology Images

Digital pathology has entered a new era with the availability of whole slide scanners that create the high-resolution images of full biopsy slides. Consequently, the uncertainty regarding the correspondence between the image areas and the diagnostic labels assigned by pathologists at the slide level, and the need for identifying regions that belong to multiple classes with different clinical significances have emerged as two new challenges. However, generalizability of the state-of-the-art algorithms, whose accuracies were reported on carefully selected regions of interest (ROIs) for the binary benign versus cancer classification, to these multi-class learning and localization problems is currently unknown. This paper presents our potential solutions to these challenges by exploiting the viewing records of pathologists and their slide-level annotations in weakly supervised learning scenarios. First, we extract candidate ROIs from the logs of pathologists' image screenings based on different behaviors, such as zooming, panning, and fixation. Then, we model each slide with a bag of instances represented by the candidate ROIs and a set of class labels extracted from the pathology forms. Finally, we use four different multi-instance multi-label learning algorithms for both slide-level and ROI-level predictions of diagnostic categories in whole slide breast histopathology images. Slide-level evaluation using 5-class and 14-class settings showed average precision values up to 81% and 69%, respectively, under different weakly labeled learning scenarios. ROI-level predictions showed that the classifier could successfully perform multi-class localization and classification within whole slide images that were selected to include the full range of challenging diagnostic categories.

Diagnostic Reproducibility: What Happens When the Same Pathologist Interprets the Same Breast Biopsy Specimen at Two Points in Time?

BACKGROUND

Surgeons may receive a different diagnosis when a breast biopsy is interpreted by a second pathologist. The extent to which diagnostic agreement by the same pathologist varies at two time points is unknown.

METHODS

Pathologists from eight U.S. states independently interpreted 60 breast specimens, one glass slide per case, on two occasions separated by ≥9 months. Reproducibility was assessed by comparing interpretations between the two time points; associations between reproducibility (intraobserver agreement rates); and characteristics of pathologists and cases were determined and also compared with interobserver agreement of baseline interpretations.

RESULTS

Sixty-five percent of invited, responding pathologists were eligible and consented; 49 interpreted glass slides in both study phases, resulting in 2940 interpretations. Intraobserver agreement rates between the two phases were 92% [95% confidence interval (CI) 88-95] for invasive breast cancer, 84% (95% CI 81-87) for ductal carcinoma-in-situ, 53% (95% CI 47-59) for atypia, and 84% (95% CI 81-86) for benign without atypia. When comparing all study participants' case interpretations at baseline, interobserver agreement rates were 89% (95% CI 84-92) for invasive cancer, 79% (95% CI 76-81) for ductal carcinoma-in-situ, 43% (95% CI 41-45) for atypia, and 77% (95% CI 74-79) for benign without atypia.

CONCLUSIONS

Interpretive agreement between two time points by the same individual pathologist was low for atypia and was similar to observed rates of agreement for atypia between different pathologists. Physicians and patients should be aware of the diagnostic challenges associated with a breast biopsy diagnosis of atypia when considering treatment and surveillance decisions.

Accuracy is in the eyes of the pathologist: The visual interpretive process and diagnostic accuracy with digital whole slide images

Digital whole slide imaging is an increasingly common medium in pathology, with application to education, telemedicine, and rendering second opinions. It has also made it possible to use eye tracking devices to explore the dynamic visual inspection and interpretation of histopathological features of tissue while pathologists review cases. Using whole slide images, the present study examined how a pathologist's diagnosis is influenced by fixed case-level factors, their prior clinical experience, and their patterns of visual inspection. Participating pathologists interpreted one of two test sets, each containing 12 digital whole slide images of breast biopsy specimens. Cases represented four diagnostic categories as determined via expert consensus: benign without atypia, atypia, ductal carcinoma in situ (DCIS), and invasive cancer. Each case included one or more regions of interest (ROIs) previously determined as of critical diagnostic importance. During pathologist interpretation we tracked eye movements, viewer tool behavior (zooming, panning), and interpretation time. Models were built using logistic and linear regression with generalized estimating equations, testing whether variables at the level of the pathologists, cases, and visual interpretive behavior would independently and/or interactively predict diagnostic accuracy and efficiency. Diagnostic accuracy varied as a function of case consensus diagnosis, replicating earlier research. As would be expected, benign cases tended to elicit false positives, and atypia, DCIS, and invasive cases tended to elicit false negatives. Pathologist experience levels, case consensus diagnosis, case difficulty, eye fixation durations, and the extent to which pathologists' eyes fixated within versus outside of diagnostic ROIs, all independently or interactively predicted diagnostic accuracy. Higher zooming behavior predicted a tendency to over-interpret benign and atypia cases, but not DCIS cases. Efficiency was not predicted by pathologist- or visual search-level variables. Results provide new insights into the medical interpretive process and demonstrate the complex interactions between pathologists and cases that guide diagnostic decision-making. Implications for training, clinical practice, and computer-aided decision aids are considered.