Related Experiment Video
Updated: Jul 7, 2026
A Pipeline for 3D Multimodality Image Integration and Computer-assisted Planning in Epilepsy Surgery
Published on: May 20, 2016
The role of artificial intelligence in diagnostic neurosurgery: a systematic review
William Li1, Armand Gumera2, Shrushti Surya3
1Department of Ophthalmology, University of Sydney, Sydney, Australia. Wili0140@uni.sydney.edu.au.
Background:
Artificial intelligence (AI) is increasingly applied in diagnostic neurosurgery, enhancing precision and decision-making in neuro-oncology, vascular, functional, and spinal subspecialties. Despite its potential, variability in outcomes necessitates a systematic review of its performance and applicability.
Methods:
A comprehensive search of PubMed, Cochrane Library, Embase, CNKI, and ClinicalTrials.gov was conducted from January 2020 to January 2025. Inclusion criteria comprised studies utilizing AI for diagnostic neurosurgery, reporting quantitative performance metrics. Studies were excluded if they focused on non-human subjects, lacked clear performance metrics, or if they did not directly relate to AI applications in diagnostic neurosurgery. Risk of bias was assessed using the PROBAST tool. This study is registered on PROSPERO, number CRD42025631040 on January 26th, 2025.
Results:
Within the 193 studies, neural networks (30%) and hybrid models (48.2%) dominated. Studies were categorised into neuro-oncology (52.69%), vascular neurosurgery (19.89%), functional neurosurgery (16.67%), and spinal neurosurgery (11.83%). Median accuracies exceeded 85% in most categories, with neuro-oncology achieving high diagnostic accuracy for tumour detection, grading, and segmentation. Vascular neurosurgery models excelled in stroke and intracranial haemorrhage detection, with median AUC values of 88% and 97%, respectively. Functional and spinal applications showed promising results, though variability in sensitivity and specificity underscores the need for standardised datasets and validation.
Discussion:
The review's limitations include the lack of data weighting, absence of meta-analysis, limited data collection timeframe, variability in study quality, and risk of bias in some studies.
Conclusion:
AI in neurosurgery shows potential for improving diagnostic accuracy across neurosurgical domains. Models used for stroke, ICH, aneurysm detection, and functional conditions such as Parkinson's disease and epilepsy demonstrate promising results. However, variability in sensitivity, specificity, and AUC values across studies underscores the need for further research and model refinement to ensure clinical viability and effectiveness.
