Deep Learning-Based Fully Automated Detection and Segmentation of Breast Mass

In the field of breast mass detection, there are many of small-scale masses in the image. However, most of the existing target detection models have low accuracy in detecting small-scale masses, which is prone to error detection and missing detection. In order to improve the detection accuracy of small-scale masses, this paper proposed a small scale target detection model Dense-Mask R-CNN based on Mask R-CNN, which is suitable for breast masses detection. Firstly, this paper improves the internal structure of FPN, and modifies the lateral connection mode in the original FPN structure to dense connection. Secondly, modify the size of the anchor of RPN to improve the location accuracy of small-scale masses. This paper uses the CBIS-DDSM dataset for all experiments. The results show that the AP value of the improved model for detecting breast masses reached 0.65 in the test set, which was 0.04 higher than that of the original Mask R-CNN.