# Artificial intelligence for surgical scene understanding: a systematic review and reporting quality meta-analysis

## TL;DR

This paper reviews AI research for understanding surgical scenes and finds that most studies lack diverse data and strong validation, limiting their clinical use.

## Contribution

The study systematically evaluates the quality and clinical relevance of surgical scene understanding AI research.

## Key findings

- Most SSU studies use small, single-center datasets, mainly from laparoscopic cholecystectomies.
- Only a small percentage of studies validate models with external data or involve clinical experts.
- Limited progress has been made in translating SSU research into clinical practice over the past decade.

## Abstract

Surgical scene understanding (SSU) uses artificial intelligence (AI) to interpret visual data from surgeries, such as laparoscopic videos. Despite promising foundational research on instrument and anatomy recognition, clinical adoption remains minimal. This systematic review and meta-analysis (PROSPERO: CRD420251005301) evaluates current SSU research in minimally invasive abdominal surgery, focusing on data curation, model design, validation, reporting standards, and clinical relevance. A total of 188 studies were reviewed. Most relied on small, single-center datasets (70.7%), primarily laparoscopic cholecystectomies (59.0%), reflecting an overall narrow topical breadth. Validation practices were often weak, rarely involving external datasets (10.1%) or clinical experts. Few studies addressed clinical translation (5.9%), model performance variability estimation (38.3%), or made code available (29.8%). Overall, limited progress toward clinical integration has been made over the past decade. Our findings highlight the need for diverse, multi-institutional datasets, robust validation practices, and clinically driven development to unlock the full potential of SSU in surgical practice.