Fig. 2: Subgroup performance on the two datasets.

All the results are evaluated on the respective test set. The U-Net and TransUnet are trained on the TUSC / QDUS training set. SAM and MedSAM are used as zero-shot segmentators with the ground truth bounding box as the prompt. a Result on TUSC-Sex; b Result on TUSC-Age; c Result on QDUS-Sex; d Result on QDUS-Age.