Practical application of SAM for breast nodules segmentation

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology
• Computer-Aided Diagnosis

Background:

• Breast cancer is a leading cause of death globally, making early detection crucial.
• Accurate breast nodule segmentation is vital for timely diagnosis and treatment.
• Training medical image segmentation models is challenging due to limited annotated data.

Purpose of the Study:

• To investigate the effectiveness of the Segment Anything Model (SAM) for breast nodule segmentation.
• To explore the impact of initial weights, organ masks, and prompt boxes on SAM's performance in medical imaging.
• To assess the feasibility of applying SAM for practical breast cancer detection.

Main Methods:

• Evaluated SAM's performance on breast nodule segmentation using various configurations.
• Investigated the influence of initial weights, specifically MedSAM.
• Assessed the utility of organ (breast) masks and different prompt box strategies.

Main Results:

• The use of MedSAM initial weights demonstrated superior segmentation results compared to standard SAM.
• Employing single, fixed prompt boxes yielded optimal segmentation outcomes.
• The proposed approach showed promise for practical application in breast cancer screening.

Conclusions:

• MedSAM initial weights and fixed prompt boxes enhance breast nodule segmentation accuracy.
• SAM shows potential for improving early breast cancer detection and diagnosis.
• Further research can optimize SAM for clinical integration in breast imaging.