To train, or not to train: exploring foundation models for TBAD segmentation

Recent advancements in biomedical image analysis have been significantly influenced by vision foundation models (FMs) originally designed for general computer vision tasks. These models, such as the Segment Anything Model (SAM) have quickly transitioned into applications in medical image processing. This study evaluates three cutting-edge segmentation FMs using Type B Aortic Dissection (TBAD) CTA images. Our experiments critically assess their strengths and weaknesses compared to traditional methods, emphasizing their potential to enhance efficiency in medical image analysis. However, our findings also indicate that these FMs have not surpassed traditional segmentation models in terms of performance, suggesting limitations in their adaptability for medical image segmentation tasks straight out of the box.