Real-time gastric polyp detection using convolutional neural networks

• DOI: 10.1371/journal.pone.0214133
• Publication Date: 2019-03-25T17:26:10Z
• Authors: Xu Zhang; Fei Chen; Tao Yu; Jiye An; Zhengxing Huang; Jiquan Liu; Weiling Hu; Liangjing Wang; Huilong Duan; Jianmin Si
• Keywords: Male; Science; Q; R; 02 engineering and technology; Adenomatous Polyps; Stomach Neoplasms; Gastroscopy; Image Processing, Computer-Assisted; 0202 electrical engineering, electronic engineering, information engineering; Medicine; Humans; Female; Neural Networks, Computer; Research Article

Computer-aided polyp detection in gastric gastroscopy has been the subject of research over past few decades. However, despite significant advances, automatic real time is still an unsolved problem. In this paper, we report on a convolutional neural network (CNN) for that constructed based Single Shot MultiBox Detector (SSD) architecture and which call SSD Gastric Polyps (SSD-GPNet). To take full advantages feature maps' information from pyramid to acquire higher accuracy, re-use abandoned by Max-Pooling layers. other words, reuse lost data pooling layers concatenate as extra maps contribute classification detection. Meanwhile, pyramid, lower are deconvolved upper make explicit relationships between effectively increase number channels. The results show our enhanced can realize real-time with 50 frames per second (FPS) improve mean average precision (mAP) 88.5% 90.4%, only little loss time-performance. And further experiment shows SSD-GPNet excellent performance improving recalls 10% (p = 0.00053), especially small This help endoscopic physicians more easily find missed polyps decrease miss rate. It may be applicable daily clinical practice reduce burden physicians.