Organ-Specific Accuracy Threshold Determination in Medical Image Segmentation Using Bootstrapping

Recognizing the heterogeneous nature of segmentation performance across diverse anatomical structures, this study posits that a uniform accuracy threshold is inadequate for evaluating model efficacy on different organs. Therefore, we propose an organ-specific approach to threshold determination. Initially, the accuracy scores derived from all model segmentations of a specific organ are aggregated. Subsequently, this aggregated data is subjected to the Bootstrapping statistical method, as previously described, to establish a robust and organ-specific accuracy threshold.