Google health, Radboud University Medical center and Karolinska Institute organised a machine learning (ML) global competition using artificial intelligence (AI) for diagnosis and Gleason grading of prostate cancer: the Prostate cANcer graDE Assessment (PANDA) challenge. Results from this challenge were recently published in the Journal Nature Medicine.1
In patients with prostate cancer, clinical grading (called Gleason grading) is quite tedious and thus has attracted substantial interest in ML. Therefore, developing large and well-annotated datasets using ML will aid in prostate cancer grading. The PANDA challenge is one such effort in this direction. The challenge was divided into two phases: the developmental phase involving 1,290 developers from 65 countries in building the best performing grading algorithm, and the validation phase which consisted of evaluating the top algorithms with internal and external validation datasets.
The researchers found that the various algorithms reached pathological-level performance when tested with independent cross-continental cohorts. Moreover, with external validation datasets, the algorithms achieved a high agreement 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.
The researchers were therefore able to conclude that a variety of algorithms successfully achieved the AI-based Gleason grading. Future clinical trials are needed to evaluate these AI-based grading tasks.