Upper gastrointestinal anatomy detection with multi-task convolutional neural networks

Healthcare Technology Letters

article

Zhang Xu¹  Yu Tao¹  Zheng Wenfang²  Lin Ne²  Huang Zhengxing¹  Liu Jiquan¹  Hu Weiling²  Duan Huilong¹  Si Jianmin² 

[1] Key Laboratory for Biomedical Engineering of Ministry of Education, College of Biomedical Engineering & Instrument Science, Zhejiang University;Department of Gastroenterology, Sir Run Run Shaw Hospital, Medical School, Zhejiang University;Institute of Gastroenterology, Zhejiang University

关键词: image classification;    medical image processing;    biomedical optical imaging;    endoscopes;    learning (artificial intelligence);    patient diagnosis;    neural nets;    biological organs;    inspection;    diagnosis quality;    classification task;    gastroscopy examination process;    upper gastrointestinal anatomy detection;    multitask convolutional neural networks;    gastrointestinal examinations;    EGD inspection process;    authors design;    multitask anatomy detection convolutional neural network;    MT-AD-CNN;    EGD inspection quality;    upper digestive tract;    informative video frames;    noninformative frames;    gastroscopic videos;    anatomies;    detection network;    noninformative images;    detected box;    informative frames;   

DOI  :  10.1049/htl.2019.0066

学科分类：肠胃与肝脏病学

来源: Wiley

【 摘 要 】

Esophagogastroduodenoscopy (EGD) has been widely applied for gastrointestinal (GI) examinations. However, there is a lack of mature technology to evaluate the quality of the EGD inspection process. In this Letter, the authors design a multi-task anatomy detection convolutional neural network (MT-AD-CNN) to evaluate the EGD inspection quality by combining the detection task of the upper digestive tract with ten anatomical structures and the classification task of informative video frames. The authors’ model is able to eliminate non-informative frames of the gastroscopic videos and detect the anatomies in real time. Specifically, a sub-branch is added to the detection network to classify NBI images, informative and non-informative images. By doing so, the detected box will be only displayed on the informative frames, which can reduce the false-positive rate. They can determine the video frames on which each anatomical location is effectively examined, so that they can analyse the diagnosis quality. Their method reaches the performance of 93.74% mean average precision for the detection task and 98.77% accuracy for the classification task. Their model can reflect the detailed circumstance of the gastroscopy examination process, which shows application potential in improving the quality of examinations.

【 授权许可】

CC BY|CC BY-ND|CC BY-NC|CC BY-NC-ND