Current measures of assessing surgical performance are resource-intensive, prone to rater bias, and inconsistent. Machine learning (ML) offers the opportunity to capitalise on the wealth of datasets emerging from operating rooms to provide rapid, automatic, and objective surgical performance assessment.

This thesis aims to develop and validate a novel ML-based surgical performance assessment tool and identify the existing barriers that limit development and translation of such digital surgical tools. Firstly, existing deficiencies in the literature for ML-based surgical performance are explored. A Delphi exercise is then conducted to identify barriers to translation of digital technology through engagement with a wide range of stakeholders including clinicians, technologists, lawyers, and patients and builds upon this review of the literature. A consensus definition of the umbrella term ‘digital surgery’ and a roadmap for overcoming identified barriers is achieved. Specific focus is then given to patients as a key stakeholder group. Two questionnaire studies detail current awareness, perceptions, and priorities of existing surgeon performance assessment methods and examine perceptions surrounding the future role of AI for performance assessment and patient-perceived barriers to achieving this.

Results of this exploratory work shape the design of a novel ML surgical performance assessment tool. An instrument detection model is trained and outputs used to design instrument motion metrics which are validated against OSATS scores and clinical outcomes. Generated ML metrics are shown to correlate to independently scored OSATS scores globally and across all 5 individual domains.

Collectively, this examination of ML-based surgeon performance assessment demonstrates how digital surgical applications have the potential to improve surgical training. However, successful translation of this technology will require engagement of multidisciplinary stakeholders to overcome not only the technological barriers but also the significant regulatory, ethical, and cultural factors identified within this thesis.