Yolo-MLSAM: SAM Based Breast Cancer Microcalcification Cluster-Segmentation Method

Abstract

Although the HQ-SAM model has achieved some results in improving the accuracy of fuzzy boundary segmentation, it is still difficult to achieve accurate segmentation in medical image processing, especially in the face of small targets such as breast cancer microcalcification clusters, in addition, high labor costs make Prompt operation cumbersome, in order to solve these problems. A novel segmentation method of breast cancer microcalcification cluster based on SAM was proposed. The method first uses Yolov8 neural network model to accurately obtain the lesion region, then uses MLSAM model to perform more detailed semantic segmentation of the lesion region, and finally realizes semi-automatic annotation function, greatly reducing the cost and complexity of manual participation. The experimental results show that compared with the HQ-SAM model, the new method has significantly improved the segmentation performance, and the dice similarity coefficient reaches 81.78%.