Deep learning generates synthetic cancer histology for explainability and education.

Artificial intelligence methods including deep neural networks (DNN) can provide rapid molecular classification of tumors from routine histology with accuracy that matches or exceeds human pathologists. Discerning how neural networks make their predictions remains a significant challenge, but explainability tools help provide insights into what models have learned when corresponding histologic features are poorly defined. Here, we present a method for improving explainability of DNN models using synthetic histology generated by a conditional generative adversarial network (cGAN). We show that cGANs generate high-quality synthetic histology images that can be leveraged for explaining DNN models trained to classify molecularly-subtyped tumors, exposing histologic features associated with molecular state. Fine-tuning synthetic histology through class and layer blending illustrates nuanced morphologic differences between tumor subtypes. Finally, we demonstrate the use of synthetic histology for augmenting pathologist-in-training education, showing that these intuitive visualizations can reinforce and improve understanding of histologic manifestations of tumor biology.

Accuracy and Reproducibility of Frozen Section Diagnosis in Ovarian Tumors.

CONTEXT.­: Intraoperative consultation-frozen section diagnosis (FSD)-determines tumor pathology and guides the optimal surgical management of ovarian neoplasms intraoperatively. OBJECTIVE.­: To evaluate the diagnostic accuracy of the FSD and analyze the discrepancy between the FSD and final diagnosis. DESIGN.­: This is a retrospective study of 618 ovarian neoplasm FSDs from 2009 to 2018 at a tertiary health care center. The discrepant cases were reviewed and reevaluated by gynecologic and general surgical pathologists. The outcomes of interest were performing unnecessary procedure, returning for a second surgery, and 30-day postoperative mortality. RESULTS.­: The sensitivity and the positive predictive value of the FSD were lower in borderline tumors than in benign and malignant epithelial ovarian tumors. Major and minor discrepancies were identified in 5.3% (33 of 618) and 12.3% of (76 of 618) cases, respectively. A root cause analysis of the major discrepant cases showed that sampling error accounted for 43% (14 of 33). The discrepancy distributions of gynecologic and general surgical pathologists were statistically similar in the overall cohort (P = .65). The overall κ for diagnostic agreement among gynecologic pathologists, general surgical pathologists, and final diagnosis was 0.18 (0.10-0.26, P < .001), implying only a slight overall agreement. Of the major discrepant cases, only 3 had a clinical implication. One overdiagnosed patient underwent an unecessary procedure, and 2 underdiagnosed patients were recommended to return for a second surgery. No patient had 30-day postoperative mortality. CONCLUSIONS.­: Frozen section diagnosis remains a definitive diagnostic tool in ovarian neoplasms and plays a crucial role in guiding intraoperative surgical management.