PS-DeVCEM: Pathology-sensitive deep learning model for video capsule endoscopy based on weakly labeled data
作者：
Highlights：
• A novel weakly-supervised method for frame-level pathology inference with only video annotation in capsule endoscopy.
• An adaptive self-supervision mechanism that maximizes the distance between positive and negative video segments.
• A new colon capsule video endoscopy dataset containing 455 videos with the total of 28,304 frames and 14 classes of diseases from 40 patients.
• Extensive ablation study on weakly supervised capsule endoscopy video analysis.
• The proposed multi label pathology detection deep learning model achieves a better F1-score and precision on 14 classes of abnormalities. of abnormalities.
摘要
•A novel weakly-supervised method for frame-level pathology inference with only video annotation in capsule endoscopy.•An adaptive self-supervision mechanism that maximizes the distance between positive and negative video segments.•A new colon capsule video endoscopy dataset containing 455 videos with the total of 28,304 frames and 14 classes of diseases from 40 patients.•Extensive ablation study on weakly supervised capsule endoscopy video analysis.•The proposed multi label pathology detection deep learning model achieves a better F1-score and precision on 14 classes of abnormalities. of abnormalities.
论文关键词：Capsule endoscopy,Residual LSTM,Attention,Self-supervision,Multiple instance learning,Adaptive pooling
论文评审过程：Received 9 December 2019, Revised 1 August 2020, Accepted 5 August 2020, Available online 10 August 2020, Version of Record 25 August 2020.
论文官网地址：https://doi.org/10.1016/j.cviu.2020.103062
