C2U

We collected CT and ultrasound liver images with corresponding reports, applying inclusion criteria of >3 ultrasound images per case, CT slice thickness <1.25mm, a time interval \u22647 days between scans, and CT reports indicating significant lesions. Expert radiologists annotated liver and lesion regions on ultrasound images, training segmentation models using a boundary-sensitive algorithm with Dice score thresholds of >0.9 for liver and >0.95 for lesions, with corrections by other radiologists for discrepancies. For CT images, pre-trained models directly predicted liver and lesion regions, followed by radiologist review and correction. This process yielded a meticulously annotated dataset for both ultrasound and CT liver and lesion segmentation.