Deep learning for tracing esophageal motility function over time

Zheng Wang, Muzhou Hou, Lu Yan, Yuzhuo Dai, Yani Yin, Xiaowei Liu

Abstract: Highlights•A high-efficiency feature manifestation method integrated high-resolution manometry with the bounding box coordinates of swallowing features from each patient to fully reconstruct the sequence information of motor function.•5-fold cross-validation was employed to estimate how accurately the proposed model will perform in practice.•A deep model was implemented to automatically learn how to identify and predict minor motility disorders, major motility disorders and normal motility.•This is the first attempt to exploit deep Conv3D and BiConvLSTM models to predict normal motility and motility disorders.