Fig. 8: Assessment of PAI.
From: A comprehensive AI model development framework for consistent Gleason grading

In this study, the Pathologist-AI interaction (PAI) strategy is evaluated, comprising two components: semi-automatic annotations to enhance pathologists’ annotation efficiency, and incremental learning for updating the existing AI model without full retraining. a illustrates the comparison between fully-manual annotation where pathologists annotate from scratch, and semi-automatic annotation where pathologists adjust pseudo annotations generated by the AI model. b presents a time comparison between fully-manual and semi-automatic annotation methods, while (c) compares the annotation area for each class before and after pathologists’ corrections. d demonstrates the workflow of incremental learning, where the base model is updated using semi-automatic annotations, leading to improved model performance after the 1st round of PAI, as shown in (e).