2020/08/26 by Francisco Luongo, Luongo, Francisco, Ryan Hakim +7 · 1 citation
Medicine · #Computer Vision and Pattern Recognition (cs.CV) #Digital Imaging in Medicine #FOS: Computer and information sciences #Intraocular Surgery and Lenses #J.3 #Robotics (cs.RO) #Surgical Simulation and Training
paper · pdf · doi:10.48550/arxiv.2008.11833
openalex publication_date 2020/08/26 · openalex created_date 2022/07/24 · openalex updated_date 2026/07/28
Our previous work classified a taxonomy of suturing gestures during a\nvesicourethral anastomosis of robotic radical prostatectomy in association with\ntissue tears and patient outcomes. Herein, we train deep-learning based\ncomputer vision (CV) to automate the identification and classification of\nsuturing gestures for needle driving attempts. Using two independent raters, we\nmanually annotated live suturing video clips to label timepoints and gestures.\nIdentification (2395 videos) and classification (511 videos) datasets were\ncompiled to train CV models to produce two- and five-class label predictions,\nrespectively. Networks were trained on inputs of raw RGB pixels as well as\noptical flow for each frame. Each model was trained on 80/20 train/test splits.\nIn this study, all models were able to reliably predict either the presence of\na gesture (identification, AUC: 0.88) as well as the type of gesture\n(classification, AUC: 0.87) at significantly above chance levels. For both\ngesture identification and classification datasets, we observed no effect of\nrecurrent classification model choice (LSTM vs. convLSTM) on performance. Our\nresults demonstrate CV's ability to recognize features that not only can\nidentify the action of suturing but also distinguish between different\nclassifications of suturing gestures. This demonstrates the potential to\nutilize deep learning CV towards future automation of surgical skill\nassessment.\n