Contextual AI for tool-tissue interaction-related surgical tasks
Research type
Research Study
Full title
Contextual AI for tool-tissue interaction-related surgical tasks
IRAS ID
314395
Contact name
Jonathan Shapey
Contact email
Sponsor organisation
King's College London
Duration of Study in the UK
0 years, 7 months, 31 days
Research summary
Neurosurgery is one the most challenging and complex surgical disciplines, reliant on significant human expertise and skill. In the UK over 28,250 cranial neurosurgery operations are performed every year (GIRFT 2020), some of which are associated with a particularly high risk of procedural complications and surgical morbidity. When complications arise, they are most often due to errors in judgement or technique, or other forms of system and communication errors that lead to adverse patient outcomes (Vedula 2017). This project will focus on two different neurosurgical procedures, each of which is associated with a significant risk of intraoperative complications (Endonasal endoscopic transsphenoidal surgery and aneurysm clipping).
This is a pilot retrospective observational study that will focus video data of 100 consecutive patients undergoing endoscopic endonasal transsphenoidal pituitary surgery (n = 50) and open aneurysm clipping (n = 50). Intraoperative surgical video data will be collected from the endoscope and microscope system for pituitary surgery and aneurysm surgery, respectively. Patients’ pre- and post-operative clinical and imaging data will also be collected in order to correlate with intraoperative events. No additional procedures will be required as part of this study.
Data analysis will be performed for surgical phase recognition and identifying risks of complications.
Methods for surgical phase recognition will be designed using deep learning methods based on recent Transformers technology for spatial and temporal information. Following this we will evaluate different models that can help identify the risks of intraoperative complications (e.g. CSF leak and intraoperative aneurysm rupture) from surgical actions using linear regression, support vector machines, and deep neural networks (Stam 2022). Once models are built, we will use Shapley Additive Explanations (SHAP) to evaluate the contribution of each feature to a higher risk of complication (Xue 2021).
REC name
East of England - Essex Research Ethics Committee
REC reference
22/EE/0306
Date of REC Opinion
21 Dec 2022
REC opinion
Favourable Opinion