Endo3D: Online Workflow Analysis for Endoscopic Surgeries Based on 3D CNN and LSTM

Surgical workflow analysis is an important topic of computer-assisted intervention and phase recognition is one of its important tasks. Features extracted from video frames by 2D convolutional networks were proved feasible for online phase analysis in former publications. In this paper, we propose to extract fine-level temporal features from video clips using 3D convolutional networks (CNN) and use Long Short-Term Memory (LSTM) networks to capture coarse-level information. By combining fine-level and coarse-level information, our proposed method outperforms state-of-the-art online methods without using specific knowledge of surgeries and almost reaches the state-of-the-art offline performance.