The Gleason score is the most important prognostic marker for prostate cancer patients, but it suffers from significant observer variability. Artificial intelligence (AI) systems based on deep learning can achieve pathologist-level performance at Gleason grading. However, the performance of such systems can degrade in the presence of artifacts, foreign tissue, or other anomalies. Pathologists integrating their expertise with feedback from an AI system could result in a synergy that outperforms both the individual pathologist and the system. Despite the hype around AI assistance, existing literature on this topic within the pathology domain is limited. We investigated the value of AI assistance for grading prostate biopsies. A panel of 14 observers graded 160 biopsies with and without AI assistance. Using AI, the agreement of the panel with an expert reference standard increased significantly (quadratically weighted Cohens kappa, 0.799 vs. 0.872; p = 0.019). On an external validation set of 87 cases, the panel showed a significant increase in agreement with a panel of international experts in prostate pathology (quadratically weighted Cohens kappa, 0.733 vs. 0.786; p = 0.003). In both experiments, on a group-level, AI-assisted pathologists outperformed the unassisted pathologists and the standalone AI system. Our results show the potential of AI systems for Gleason grading, but more importantly, show the benefits of pathologist-AI synergy.

Purpose

Gastrointestinal stromal tumors (GIST) commonly arise in different regions of the stomach and are driven by various mutations (most often in KIT, PDGFRA, and SDHx). We hypothesized that the anatomic location of gastric GIST is associated with unique genomic profiles and distinct driver mutations.

Experimental design

We compared KIT versus non-KIT status with tumor location within the National Cancer Database (NCDB) for 2,418 patients with primary gastric GIST. Additionally, we compiled an international cohort (TransAtlantic GIST Collaborative, TAGC) of 236 patients and reviewed sequencing results, cross-sectional imaging, and operative reports. Subgroup analyses were performed for tumors located proximally versus distally. Risk factors for KIT versus non-KIT tumors were identified using multivariate regression analysis. A random forest machine learning model was then developed to determine feature importance.

Results

Within the NCDB cohort, non-KIT mutants dominated distal tumor locations (P < 0.03). Proximal GIST were almost exclusively KIT mutant (96%) in the TAGC cohort, whereas 100% of PDGFRA and SDH-mutant GIST occurred in the distal stomach. On multivariate regression analysis, tumor location was associated with KIT versus non-KIT mutations. Using random forest machine learning analysis, stomach location was the most important feature for predicting mutation status.

Conclusions

We provide the first evidence that the mutational landscape of gastric GIST is related to tumor location. Proximal gastric GIST are overwhelmingly KIT mutant, irrespective of morphology or age, whereas distal tumors display non-KIT genomic diversity. Anatomic location of gastric GIST may therefore provide immediate guidance for clinical treatment decisions and selective confirmatory genomic testing when resources are limited.

Artificial intelligence (AI) has shown promise for diagnosing prostate cancer in biopsies. However, results have been limited to individual studies, lacking validation in multinational settings. Competitions have been shown to be accelerators for medical imaging innovations, but their impact is hindered by lack of reproducibility and independent validation. With this in mind, we organized the PANDA challenge-the largest histopathology competition to date, joined by 1,290 developers-to catalyze development of reproducible AI algorithms for Gleason grading using 10,616 digitized prostate biopsies. We validated that a diverse set of submitted algorithms reached pathologist-level performance on independent cross-continental cohorts, fully blinded to the algorithm developers. On United States and European external validation sets, the algorithms achieved agreements of 0.862 (quadratically weighted κ, 95% confidence interval (CI), 0.840-0.884) and 0.868 (95% CI, 0.835-0.900) with expert uropathologists. Successful generalization across different patient populations, laboratories and reference standards, achieved by a variety of algorithmic approaches, warrants evaluating AI-based Gleason grading in prospective clinical trials.