Influence of Video-Preprocessing on Deep Learning-based Surgical Skill Assessment

Abstract Automatic surgical skill assessment in robotic surgery based on video data is essential facilitating faster learning curves for trainees while relieving expert surgeons from the time- and cost-intensive feedback process. Recent years have shown several advancements this area by utilizing deep learning. While current research focuses novel architectures, influence of video-preprocessing their performance remains unknown. In work, we present first investigation learning-based assessment. Thus, integrated four preprocessing modules, i.e. Deblurring, Segmentbased Sampling, Optical Flow Combination all them, into JIGSAWS dataset using a well-established network architecture. Despite single steps showing no clear improvement, showed higher median lower variance. Furthermore, performed frame-wise investigations optical flow artifacts reduction combined setting. Our results highlight potential well-calibrated automatic