Surgeons versus computer vision: a comparative analysis on surgical phase recognition capabilities

Marco Mezzina ^1,2, Pieter De Backer ^3,4, Tom Vercauteren ⁵

⁰Orsi Academy, Ghent, Belgium. marco.mezzina@orsi.be.

International Journal of Computer Assisted Radiology and Surgery +

|

May 31, 2025

View abstract on PubMed

Summary

Temporal context significantly improves surgical phase recognition (SPR) accuracy for both experts and AI in robot-assisted partial nephrectomy. Visual landmarks are key for accurate classification in complex, nonlinear procedures.

Area Of Science

Artificial Intelligence in Medicine
Surgical Workflow Analysis
Computer Vision

Background

Automated surgical phase recognition (SPR) aids surgical video review, education, and skill assessment.
Prior research on SPR primarily examined short, linear procedures.
The influence of temporal context on expert classification accuracy in nonlinear procedures remains underexplored.

Purpose Of The Study

To investigate the impact of temporal context on surgical phase recognition for robot-assisted partial nephrectomy (RAPN), a nonlinear procedure.
To compare the performance of human experts and AI models in classifying surgical phases.
To identify key visual landmarks influencing phase classification.

Main Methods

Urologists of varying expertise classified RAPN phases using single frames and video snippets.
Participants reported confidence and identified visual landmarks.
AI models, with and without temporal context, were trained on RAPN data.

Main Results

Video snippets and visual landmarks enhanced phase classification accuracy for all participants.
Expert surgeons demonstrated higher confidence and accuracy than novices.
AI model performance was comparable to surgeons, with temporal context improving both.

Conclusions

Surgical phase recognition is complex for both humans and AI.
Providing temporal information enhances SPR performance.
Surgical tools and organs are critical landmarks for automated SPR development.

Keywords:

Deep learning RAPN Surgical data science Surgical phase recognition