A deep learning model leveraging SAM improved segmentation speed and matched radiologist performance in classifying ovarian lesions on MRI.
Key Details
- 1Researchers from Johns Hopkins developed an MRI-based, end-to-end DL pipeline incorporating Meta’s Segment Anything Model (SAM) and DenseNet-121.
- 2Integrated model reduced segmentation time by 4 minutes per lesion compared to manual segmentation.
- 3SAM achieved a Dice coefficient of 0.86 to 0.88 for lesion segmentation.
- 4DL model achieved an AUC of 0.85 internally and 0.79 externally, on par with radiologists’ AUC of 0.84 (p > 0.05).
- 5Training data included 534 lesions (internal) and 87 lesions (external) from the US and Taiwan.
Why It Matters
This study demonstrates that foundation AI models can efficiently streamlines lesion segmentation and classification with accuracy rivaling radiologists, suggesting a path toward more automated and collaborative clinical workflows in radiology.
Source
AuntMinnie
Related News
•Radiology Business
New AI Flags Uncertainty in Breast Cancer Imaging Predictions
University of Pittsburgh researchers developed an AI model that quantifies uncertainty in breast cancer diagnostic predictions to enhance trustworthiness.
•Radiology Business
FDA Approves SimBioSys AI Tool for Breast Cancer Risk via MRI
SimBioSys received FDA clearance for TumorSight Risk, an AI tool assessing breast cancer recurrence risk using MRI.
•AuntMinnie
LLMs Rarely Recommend Unnecessary Imaging, Except for Thyroid Ultrasound
Large language models seldom over-recommend imaging, but thyroid ultrasound guidance shows substantial discrepancies.