New AI System Captures Uncertainty in Medical Imaging to Rival Human Judgment
-
New AI system, Tyche, captures uncertainty in medical image segmentation by providing multiple plausible options instead of just one answer.
-
Tyche doesn't need retraining for new tasks, making it easier to use for clinicians and researchers.
-
It modifies a neural network to output multiple candidate segmentations that capture diversity like human annotators.
-
In tests, Tyche's predictions matched human uncertainty and its best predictions outperformed baseline models.
-
Tyche runs faster than most models and can outperform some complex models without specialized training.