Doga Demirel1, Bryce Palmer2, Gunnar Sundberg2, Bayazit Karaman2, Tansel Halic3, Sinan Kockara3, Nizamettin Kockara4, Mark Edward Rogers5, Shahryar Ahmadi6. 1. Computer Science Department, Florida Polytechnic University, 4700 Research Way, Lakeland, FL, 33805, USA. ddemirel@floridapoly.edu. 2. Computer Science Department, Florida Polytechnic University, 4700 Research Way, Lakeland, FL, 33805, USA. 3. Computer Science Department, University of Central Arkansas, Conway, AR, USA. 4. Department of Orthopedics and Traumatology, Erzincan University Medical School, Erzincan, Turkey. 5. Alabama Ortho Spine and Sports in Birmingham, Birmingham, AL, USA. 6. Memorial Orthopaedic Surgical Group, Long Beach, CA, USA.
Abstract
PURPOSE: We aim to develop quantitative performance metrics and a deep learning model to objectively assess surgery skills between the novice and the expert surgeons for arthroscopic rotator cuff surgery. These proposed metrics can be used to give the surgeon an objective and a quantitative self-assessment platform. METHODS: Ten shoulder arthroscopic rotator cuff surgeries were performed by two novices, and fourteen were performed by two expert surgeons. These surgeries were statistically analyzed. Two existing evaluation systems: Basic Arthroscopic Knee Skill Scoring System (BAKSSS) and the Arthroscopic Surgical Skill Evaluation Tool (ASSET), were used to validate our proposed metrics. In addition, a deep learning-based model called Automated Arthroscopic Video Evaluation Tool (AAVET) was developed toward automating quantitative assessments. RESULTS: The results revealed that novice surgeons used surgical tools approximately 10% less effectively and identified and stopped bleeding less swiftly. Our results showed a notable difference in the performance score between the experts and novices, and our metrics successfully identified these at the task level. Moreover, the F1-scores of each class are found as 78%, 87%, and 77% for classifying cases with no-tool, electrocautery, and shaver tool, respectively. CONCLUSION: We have constructed quantitative metrics that identified differences in the performances of expert and novice surgeons. Our ultimate goal is to validate metrics further and incorporate these into our virtual rotator cuff surgery simulator (ViRCAST), which has been under development. The initial results from AAVET show that the capability of the toolbox can be extended to create a fully automated performance evaluation platform.
PURPOSE: We aim to develop quantitative performance metrics and a deep learning model to objectively assess surgery skills between the novice and the expert surgeons for arthroscopic rotator cuff surgery. These proposed metrics can be used to give the surgeon an objective and a quantitative self-assessment platform. METHODS: Ten shoulder arthroscopic rotator cuff surgeries were performed by two novices, and fourteen were performed by two expert surgeons. These surgeries were statistically analyzed. Two existing evaluation systems: Basic Arthroscopic Knee Skill Scoring System (BAKSSS) and the Arthroscopic Surgical Skill Evaluation Tool (ASSET), were used to validate our proposed metrics. In addition, a deep learning-based model called Automated Arthroscopic Video Evaluation Tool (AAVET) was developed toward automating quantitative assessments. RESULTS: The results revealed that novice surgeons used surgical tools approximately 10% less effectively and identified and stopped bleeding less swiftly. Our results showed a notable difference in the performance score between the experts and novices, and our metrics successfully identified these at the task level. Moreover, the F1-scores of each class are found as 78%, 87%, and 77% for classifying cases with no-tool, electrocautery, and shaver tool, respectively. CONCLUSION: We have constructed quantitative metrics that identified differences in the performances of expert and novice surgeons. Our ultimate goal is to validate metrics further and incorporate these into our virtual rotator cuff surgery simulator (ViRCAST), which has been under development. The initial results from AAVET show that the capability of the toolbox can be extended to create a fully automated performance evaluation platform.
Authors: Ryan J Koehler; Simon Amsdell; Elizabeth A Arendt; Leslie J Bisson; Jonathan P Braman; Jonathan P Bramen; Aaron Butler; Andrew J Cosgarea; Christopher D Harner; William E Garrett; Tyson Olson; Winston J Warme; Gregg T Nicandri Journal: Am J Sports Med Date: 2013-04-02 Impact factor: 6.202