Two-stage CNNs for computerized BI-RADS categorization in breast ultrasound images

• DOI: 10.1186/s12938-019-0626-5
• Publication Date: 2019-01-24T08:03:23Z
• Authors: Yunzhi Huang; Luyi Han; Haoran Dou; Honghao Luo; Zhen Yuan; Qi Liu; Jiang Zhang; Guangfu Yin
• Keywords: BI-RADS; Grading (engineering); Breast ultrasound; Breast imaging; Breast tumor

Quantizing the Breast Imaging Reporting and Data System (BI-RADS) criteria into different categories with single ultrasound modality has always been a challenge. To achieve this, we proposed two-stage grading system to automatically evaluate breast tumors from images five based on convolutional neural networks (CNNs).This new developed automatic was consisted of two stages, including tumor identification grading. The constructed network for identification, denoted as ROI-CNN, can identify region contained original images. following categorization network, G-CNN, generate effective features differentiating identified regions interest (ROIs) categories: Category "3", "4A", "4B", "4C", "5". Particularly, promote predictions by ROI-CNN better tailor tumor, refinement procedure Level-set leveraged joint between stage stage.We tested against 2238 cases in With accuracy an indicator, our computerized evaluation exhibited performance comparable that subjective determined physicians. Experimental results show framework 0.998 0.940 0.734 0.922 0.876 "5".The scheme extract final classification decoupling CNNs. Besides, extend diagnosing sub-categories according BI-RADS rather than merely distinguishing malignant benign.