An observational analysis of the current regulatory-approved Artificial Intelligence Medical Devices and Softwares used in intraoperative settings: Where is AI in Neurosurgery?

Introduction: Artificial Intelligence (AI) is revolutionizing medicine, but its adoption in intraoperative settings remains slow. To facilitate the expansion of intraoperative AI (iAI) use in neurosurgery, it is crucial to first understand the current landscape of iAI technologies and the barriers to implementation that exist. We aim to provide a comprehensive overview of FDA/CE approved iAI devices and software. We suggest ways to accelerate iAI adoption and usability to fulfill iAI’s potential in transforming neurosurgical practice.

Methods: All medical devices and software up to July 2023 were extracted from 6 repositories (AI/ML-enabled Device FDA list, AI Central, AI Tools Database, Medical Futurist, articles published on Nature and Lancet). After excluding duplicate entries, n = 680 devices were screened. iAI was defined to be technology which computes and outputs data within the intraoperative setting. Device name, company, country, specialty, procedures, regulatory approval status, function of AI, AI model, end-user, end-user interface, amount of training data, and publications/reports related to device, were extracted for descriptive analysis.

Results: Out of 13 iAI devices, only 2 were for neurosurgical procedures (1 cranial, 1 spine). The cranial device utilized AI to detect brain abnormalities in T1+C MR images, enabling easier view centering and aiding registration with pre-registration based on guide points. The spine device had AI implemented in an ultrasound imaging system to provide real-time soft tissue imaging and differentiation during procedures. Both of these devices used computer vision models and were used by the surgeons. None of these devices disclosed how much data was used to train or validate the AI architecture.

Conclusion : Despite the tremendous interest in AI in surgery, only few iAI devices in neurosurgery are available on the market. iAI remains mostly in the diagnostic realm and few have been adopted into the treatment paradigm. Transparency regarding AI use and the amount of training is poor. A systematic review (OSF registration: 10.17605/OSF.IO/8XAFY) will be conducted to further identify the gap between translation and research of neurosurgical application of AI.