Summary
This new digital ecosystem, incorporating open surgery video capture and recording, automated anonymisation, and phase analysis, is the first of its kind in open surgery and will allow understanding and sharing of best practices, for example, between peers and trainees and mentors to improve education hence potentially drive further towards standardisation in orthopaedic surgery.
Abstract
Background
Review of surgical videos through provision of data analytics and insights is on the increase. Capturing and recording open orthopaedic procedures to gain novel quantitative data insights presents different challenges to those in endoscopic surgical procedures. Not only is the video more difficult to capture but issues with redaction and anonymisation need to be addressed.
Objectives
To assess the feasibility of recording, anonymising, and analysing surgical video data in open orthopaedic surgery to provide novel insights into surgical processes and preferences that can be used for education and post-operative review.
Design and Methods
Open total knee replacement (TKR) videos were recorded using a head-mounted camera on both the lead surgeon and scrub nurse in the operating room (OR). Videos were uploaded to an online web platform, Touch SurgeryTM, where a novel algorithm was applied to redact any faces or personal identifiable information e.g. name badges, white boards, imaging. The surgical workflow was mapped and a proportion of the videos manually annotated with surgical phase information. These annotations were used to develop a novel machine learning (ML) algorithm to automatically detect phases in TKR videos.
Results
A total of 48 procedures were captured. Videos were uploaded to Touch SurgeryTM and processed to redact sensitive information and partition into surgical phases. Our redactor model achieved the accuracy of 98% and our phase recognition model the accuracy of 95.4%. On average, surgical phase transitions were detected within 15 seconds of manual phase labels annotated by two experts.
Of 48 videos analysed, 89% had the same workflow (identical sequence of phases) with 4% of them marked as outlier based on case duration. Phase outliers were largely due to variable patient anatomy and the expertise/ level of the surgeon. Case duration for trainees was 40% longer than experts. The detailed breakdown of the phases enabled us to identify Medial Parapatellar Arthrotomy and Femoral AP and Chamfer Cuts as the most challenging phases for which more assistance and mentorship would be required.
Conclusion
This new digital ecosystem, incorporating open surgery video capture and recording, automated anonymisation, and phase analysis, is the first of its kind in open surgery and will allow understanding and sharing of best practices, for example, between peers and trainees and mentors to improve education hence potentially drive further towards standardisation in orthopaedic surgery. Extending this pipeline across more clinical sites will provide further insights in surgical workflow variation. This study presents the first step in developing novel technologies to process open orthopaedic surgical video that will be used to inform development of real-time advanced technologies, based on ML that we believe will provide novel information and guidance in the operating theatre in the future that will ultimately benefit patients in the long term.