Accuracy of Segment-Anything Model (SAM) in medical image segmentation tasks

It is found that SAM without re-training on medical images do not perform as accurately as U-Net or other deep learning models trained onmedical images.