Application of Image Segmentation and Convolutional Neural Network in Classification Algorithms for Mammary X-ray Molybdenum Target Image

摘要

Breast cancer is the second leading cause of death in women worldwide and of the methods for the detection of breast cancer, Mammography is consid-ered promising and effective. In order to improve the detection, the study explored automatic recognition of Mammary X-ray Molybdenum target images on the basis of the clustered distribution pleomorphic calcification images of Breast Imaging Reporting & Data System category 4 (BI-RADS 4) obtained from an open access database - Digital Database for Screening Mammography (DDSM). The region of interest (ROI) of molybdenum target images was firstly segmented into sub-images by coordinate matching technology, and then the sub-images were scanned row by row and subdivided into mini-images. Those mini-images containing lesions were thus screened out and used as the objects of neural network recognition. Pat-tern recognition was carried out via the classical convolutional neural networks such as VGGNet16, VGGNet11 and AlexNet, and the improved AlexNet network without LRN layer. The results showed that identification and subdivision of the ROIs together with the improved AlexNet network could significantly improve the performance of recognition. By comparison with other methods, the new meth-ods developed herein could provide additional and useful information for clinical diagnosis, and lay a technical foundation for refining classification of BI-RADS4 images into sub-categories and furthering accurate diagnosis.