In an external validation study, the algorithm achieved an overall high agreement with the reference standard created by the consensus of the two uropathologists, with quadratic weighted kappa of 0.79.
It performed especially well at clinically meaningful thresholds in evaluating benign vs malignant (κ 0.927), and clinically low-risk (benign, GG 1-2) versus high-risk (GG 3-5) disease (κ 0.858).
In another study, the algorithm achieved an IoU of 0.8 for cancer coverage.
* All kappa results were measured at the patch level.
