This multicenter study develops a hybrid AI pipeline combining foundation segmentation models and multimodal diagnostic analysis to improve classification of ovarian lesions as benign or malignant from MRI scans.
This study used three independent datasets from large academic medical centers in the United States and Taiwan (ChangGung Memorial Hospital).
In this retrospective study of 532 patients with ovarian lesions on MRI scans from three institutions, the Segment Anything Model (SAM) showed good segmentation accuracy (Dice coefficient, 0.86–0.88) and shorter segmentation time than manual methods (2.9–6.4 minutes vs 7.0–10.6 minutes; P < .001).
The deep learning model's classification performance using SAMsegmented MRI scans and clinical and lesion characteristics was comparable with radiologists' performance (AUC, 0.79 vs 0.84–0.93; P= .72–.95).
These results describe an accurate, efficient pipeline that integrates SAM with DL-based classification for differentiating malignant from benign ovarian lesions on MRI scans.
It reduced segmentation time and achieved classification performance comparable with that of radiologists.
【MORE】