Semi-automated segmentation of magnitude images in 4D flow MR scans using segment anything model 2 (SAM 2)
Amirkhosro Kazemi, Aryan Ghazipour, Tyler Settle, Marcus F. Stoddard, Amir A. Amini · 2025
Accurate segmentation of 4D flow MRI is essential for assessing hemodynamic biomarkers such as flow patterns and pressure gradients. We analyzed magnitude images in 5,440 4D flow MR images from 34 patients diagnosed with aortic stenosis using the Segment Anything Model 2 (SAM 2) with point and rectangle box prompting. SAM 2 consistently delivered strong performance. Compared to manual segmentations, SAM 2 achieved a Dice Similarity Coefficient (DSC) of 0.85 and an Intersection over Union (IoU) of 0.82, indicating the model’s robustness in segmenting magnitude images of 4D flow MR scans. SAM 2’s segmentation performance was particularly enhanced in regions with clearer vessel background contrast, highlighting the importance of contrast in accurate segmentation. The model’s ability to handle various cardiac phases reinforces its adaptability to different clinical conditions, demonstrating the potential utility of SAM 2 in 4D flow MRI segmentation within clinical settings.