Bulten et al. developed an artificial intelligence (AI) system that would use deep learning to remove pathologist-level observer variability from calculation of Gleason scores in prostate cancer patient prognosis. Acknowledging that artifacts, foreign tissue, and other anomalies can reduce the efficacy of systems such as these, they proposed the synergistic integration of AI systems with pathologists’ experience. Using AI across 14 observers and 160 biopsies, they demonstrated that agreement of the panel with an expert reference standard increased significantly. Even in an external validation panel, the group demonstrated that AI- assisted pathologists outperformed unassisted pathologists as well as stand-alone AI systems. While acknowledging the limitations of their data sets, they indicate that no other studies have been performed to investigate this potentially important tool in prostate cancer prognosis.
展开▼
