Artificial intelligence in the operating room: A systematic review of AI models for surgical phase, instruments and anatomical structure identification.

Abstract

Introduction: This systematic review examines the application of multiple deep learning algorithms in the analysis of intraoperative videos to enable feature extraction and pattern recognition of surgical phases, anatomical structures, and surgical instruments.

Material and methods: A comprehensive literature search was conducted across PubMed, Web of Science, and EBSCO, covering studies published until March 2024. This review includes studies that applied AI models in the operating room for surgical-phase recognition and/or anatomical structures and instruments. Only studies utilizing machine learning or deep learning for surgical video analysis were considered. The primary outcome measures were accuracy, precision, recall, and F1 score.

Results: A total of 21 studies were included. Multilayer architecture of interconnected neural networks was predominantly used. The deep learning models demonstrated promising results, with accuracy ranging from 81% to 93.2% for surgical-phase recognition. Anatomical structure recognition models achieved accuracy between 71.4% and 98.1%.

Conclusions: Artificial intelligence has the potential to significantly improve surgical precision and workflow, with demonstrated success in phase recognition and anatomical structure identification. However, further research is needed to address dataset limitations, standardize annotation protocols, and minimize biases.