Search

Artificial intelligence (AI), including deep learning methods that leverage neural network-based algorithms, hold significant promise for dermatopathology and other areas of diagnostic pathology in research and clinical practice. There has been significant progress over past several years in applying AI to analyzing digital histopathology images for diagnosis. While much work in AI analysis of histopathology data remains investigational, recent regulatory agency approval in Europe and United States of AI-assisted tools for clinical use in histopathologic diagnosis of prostate and breast cancer herald broader movement of AI into the clinical diagnostic realm of anatomic pathology, including dermatopathology. However, significant challenges remain in translating AI from research into clinical practice, including algorithmic real-world performance, robustness to variation in data sets and practice settings, effective integration into clinical workflows, and cost effectiveness. This review introduces core concepts and terminology in AI, and assesses current progress and challenges in applying AI to dermatopathology. more...

Supplemental Figures

Supplemental Methods

Current regimens for the detection and surveillance of bladder cancer are invasive and have suboptimal sensitivity. Here, we present a novel high-throughput sequencing (HTS) method for detection of urine tumor DNA (utDNA) called utDNA CAPP-Seq (uCAPP-Seq) and apply it to 67 healthy adults and 118 patients with early-stage bladder cancer who had urine collected either prior to treatment or during surveillance. Using this targeted sequencing approach, we detected a median of 6 mutations per patient with bladder cancer and observed surprisingly frequent mutations of the PLEKHS1 promoter (46%), suggesting these mutations represent a useful biomarker for detection of bladder cancer. We detected utDNA pretreatment in 93% of cases using a tumor mutation–informed approach and in 84% when blinded to tumor mutation status, with 96% to 100% specificity. In the surveillance setting, we detected utDNA in 91% of patients who ultimately recurred, with utDNA detection preceding clinical progression in 92% of cases. uCAPP-Seq outperformed a commonly used ancillary test (UroVysion, P = 0.02) and cytology and cystoscopy combined (P ≤ 0.006), detecting 100% of bladder cancer cases detected by cytology and 82% that cytology missed. Our results indicate that uCAPP-Seq is a promising approach for early detection and surveillance of bladder cancer. Significance: This study shows that utDNA can be detected using HTS with high sensitivity and specificity in patients with early-stage bladder cancer and during post-treatment surveillance, significantly outperforming standard diagnostic modalities and facilitating noninvasive detection, genotyping, and monitoring. This article is highlighted in the In This Issue feature, p. 453 more...

Donor livers undergo subjective pathologist review of steatosis before transplantation to mitigate the risk for early allograft dysfunction (EAD). We developed an objective, computer vision artificial intelligence (CVAI) platform to score donor liver steatosis and compared its capability for predicting EAD against pathologist steatosis scores.Two pathologists scored digitized donor liver biopsy slides from 2014 to 2019. We trained four CVAI platforms with 1:99 training:prediction split. Mean intersection-over-union (IU) characterized CVAI model accuracy. We defined EAD using liver function tests within 1 week of transplantation. We calculated separate EAD logistic regression models with CVAI and pathologist steatosis and compared the models' discrimination and internal calibration.From 90 liver biopsies, 25,494 images trained CVAI models yielding peak mean IU = 0.80. CVAI steatosis scores were lower than pathologist scores (median 3% vs 20%, P 0.001). Among 41 transplanted grafts, 46% developed EAD. The median CVAI steatosis score was higher for those with EAD (2.9% vs 1.9%, P = 0.02). CVAI steatosis was independently associated with EAD after adjusting for donor age, donor diabetes, and MELD score (aOR = 1.34, 95%CI = 1.03-1.75, P = 0.03).The CVAI steatosis EAD model demonstrated slightly better calibration than pathologist steatosis, meriting further investigation into which modality most accurately and reliably predicts post-transplantation outcomes. more...

Artificial intelligence (AI) algorithms continue to rival human performance on a variety of clinical tasks, while their actual impact on human diagnosticians, when incorporated into clinical workflows, remains relatively unexplored. In this study, we developed a deep learning-based assistant to help pathologists differentiate between two subtypes of primary liver cancer, hepatocellular carcinoma and cholangiocarcinoma, on hematoxylin and eosin-stained whole-slide images (WSI), and evaluated its effect on the diagnostic performance of 11 pathologists with varying levels of expertise. Our model achieved accuracies of 0.885 on a validation set of 26 WSI, and 0.842 on an independent test set of 80 WSI. Although use of the assistant did not change the mean accuracy of the 11 pathologists (p = 0.184, OR = 1.281), it significantly improved the accuracy (p = 0.045, OR = 1.499) of a subset of nine pathologists who fell within well-defined experience levels (GI subspecialists, non-GI subspecialists, and trainees). In the assisted state, model accuracy significantly impacted the diagnostic decisions of all 11 pathologists. As expected, when the model’s prediction was correct, assistance significantly improved accuracy (p = 0.000, OR = 4.289), whereas when the model’s prediction was incorrect, assistance significantly decreased accuracy (p = 0.000, OR = 0.253), with both effects holding across all pathologist experience levels and case difficulty levels. Our results highlight the challenges of translating AI models into the clinical setting, and emphasize the importance of taking into account potential unintended negative consequences of model assistance when designing and testing medical AI-assistance tools. more...

We report a case of decidual perivascular non-necrotizing granulomas in a placenta from a pregnancy complicated by severe preeclampsia with no evidence of infection. The mother was a 20-year-old primigravida with severe preeclampsia diagnosed in the third trimester with subsequent delivery of a healthy baby boy at 37 weeks 5 days gestation. Pathologic examination of the placenta showed scattered non-necrotizing granulomas in decidua, often adjacent to remodeled decidual arteries without fibrinoid necrosis. These were well-formed, non-necrotizing granulomas with scant lymphoid cuffs. Polarization microscopy did not show foreign material. There were no histopathologic or clinical findings suggestive of maternal-fetal infection or systemic vasculitis at the time of delivery, and the mother had no other reported conditions associated with granulomatous inflammation. Our case demonstrates that granulomatous reaction may be seen in the placenta from a pregnancy complicated by severe preeclampsia, although work-up for infection may be indicated. more...

Smooth-Muscle Myopathy in Systemic Lupus Erythematosus Presenting With Intestinal Pseudo-Obstruction

Photoswitchable distance constraints in the form of photoisomerizable chemical cross-links offer a general approach to the design of reversibly photocontrolled proteins. To apply these effectively, however, one must have guidelines for the choice of cross-linker structure and cross-linker attachment sites. Here we investigate the effects of varying cross-linker structure on the photocontrol of folding of the Fyn SH3 domain, a well-studied model protein. We develop a theoretical framework based on an explicit-chain model of protein folding, modified to include detailed model linkers, that allows prediction of the effect of a given linker on the free energy of folding of a protein. Using this framework, we were able to quantitatively explain the experimental result that a longer, but somewhat flexible, cross-linker is less destabilizing to the folded state than a shorter more rigid cross-linker. The models also suggest how misfolded states may be generated by cross-linking, providing a rationale for altered dynamics seen in nuclear magnetic resonance analyses of these proteins. The theoretical framework is readily portable to any protein of known folded state structure and thus can be used to guide the design of photoswitchable proteins generally. more...

Objectives Controversy persists regarding appropriate radiographic surveillance strategies after lung cancer resection. We compared the impact of surveillance computed tomography scan versus chest radiography in patients who underwent resection for stage I lung cancer. Methods A retrospective analysis was performed of all patients undergoing resection for pathologic stage I lung cancer from January 2000 to April 2013. After resection, follow-up included routine history and physical examination in conjunction with chest radiography or computed tomography at the discretion of the treating physician. Identification of successive lung malignancy (ie, recurrence at any new site or new primary) and survival were recorded. Results There were 554 evaluable patients, with 232 receiving routine postoperative computed tomography and 322 receiving routine chest radiography. Postoperative 5-year survival was 67.8% in the computed tomography group versus 74.8% in the chest radiography group ( P = .603). Successive lung malignancy was found in 27% (63/232) of patients receiving computed tomography versus 22% (72/322) receiving chest radiography ( P = .19). The mean time from surgery to diagnosis of successive malignancy was 1.93 years for computed tomography versus 2.56 years for chest radiography ( P = .046). For the computed tomography group, 41% (26/63) of successive malignancies were treated with curative intent versus 40% (29/72) in the chest radiography group ( P = .639). Cox proportional hazard analysis indicated imaging modality (computed tomography vs chest radiography) was not associated with survival ( P = .958). Conclusions Surveillance computed tomography may result in earlier diagnosis of successive malignancy versus chest radiography in stage I lung cancer, although no difference in survival was demonstrated. A randomized trial would help determine the impact of postoperative surveillance strategies on survival. more...