Pituitary surgeons will test an artificial intelligence system that watches live video from the operating microscope and highlights critical structures in real time, aiming to reduce the 40–70% rate of incomplete tumour removal and the 1–2% risk of vascular injury. Current surgical navigation tools cannot account for the brain shifting during surgery and disrupt the surgeon’s workflow. This project trains a computer vision AI to recognise anatomical landmarks directly from the surgical video feed, overlaying guidance without those limitations. The team has already shown the approach works in preclinical tests. The research moves through four stages: refining the AI on a surgical simulator with ten neurosurgeons, then testing it against standard practice in a cross-over trial with 30 surgeons, followed by first-in-human studies in six to 30 patients, and finally gathering patient views through a focus group. If successful, this would be the first real-time computer vision system used in neurosurgery, providing a platform ready for definitive multicentre trials on safety, effectiveness, and cost.
View original technical description
This proposal seeks to improve patient safety in pituitary surgery by leveraging advances in artificial intelligence (AI) and integrating it into the operating theatre in a systematic stepwise process. Background Pituitary tumours are among the most common brain tumours, often causing significant morbidity. Transsphenoidal surgery is the primary treatment but it is difficult due to critical structures near the tumour, contributing to incomplete resections (40-70%) or vascular injuries (1-2%). These complications are potentially devasting to patients - potentially resulting in death or significant morbidity and quality of life impact (e.g. stroke, need for radiotherapy). To reduce these risks, surgeons rely on adjuncts to help recognise and protect important anatomical structures. However current adjuncts (image guidance and Doppler ultrasound) have significant limitations (e.g., don't address intra-operative shifts and disrupt workflow). Computer vision AI offers real-time anatomical recognition, overlaying it directly on surgical video, without these limitations. Our preclinical work demonstrates its potential for augmenting surgeon performance. Question Can computer vision technology improve real-time anatomy recognition in pituitary surgery, thereby enhancing surgical safety and performance? Methodology Four work packages (WPs) progress from preclinical simulation to first-in-human studies, aligning with surgical technology evaluation guidelines (IDEAL) and early clinical AI reporting guidelines (DECIDE-AI). Preparatory work including AI model development, ethical approvals, systematic review and user surveys is already underway. WP1: Simulator-based Exploratory Study Aim: Test real-time AI assistance on an existing high-fidelity pituitary simulator to refine the prototype design. Design: Ten neurosurgeons perform simulated procedures with the AI prototype. Outcomes: AI platform performance (accuracy, errors), structured usability questionnaires, Think Aloud feedback, workload assessment. WP2: Simulator-based Comparative Study Aim: Evaluate the prototype's impact on safety, performance and training in a simulated setting. Design: Cross-over trial (with vs without AI) involving 30 neurosurgeons using a surgical simulator. Outcomes: AI platform performance (accuracy, errors), comparative surgical performance (OSATS, time), confidence changes, educational impact. WP3: Early Clinical Evaluation Aim: Implement the platform on a secondary monitor in live pituitary surgeries at a single centre to iteratively improve platform design and assess its safety and feasibility. Design: Non-comparative, single-centre, proof-of-concept study (n=6; IDEAL Stage 1) followed by single-centre cases series (n=20-30; IDEAL Stage 2a). Design, recruitment and reporting will be guided by DECIDE AI guidelines for early clinical AI evaluation. Outcomes: Safety (surgeon distraction observation, AI output errors, workflow disruption), AI platform performance (accuracy, functionality), structured usability questionnaires, tracking of surgeon trust, and clinical & patient-reported outcomes. WP4: Patient Perspectives Evaluation Aim: To explore patient acceptability and safety concerns to inform improvement in platform design and implementation. Design: WP3 clinical data presentation to the UCL Pituitary Patient Advisory Group with a structured focus group session. Outcomes: Structured acceptability questionnaires, and thematic analysis of qualitative data. Anticipated impact This proposal seeks to deliver the first real-time perioperative computer vision system in neurosurgery, with the potential to improve surgical safety and improve patient outcomes. Through systematic multi-stakeholder evaluation, it will provide an optimised and stable AI platform, ready for definitive multicentre clinical trials assessing effectiveness, safety and economic impact.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know