PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
September 30, 2021Open Access

Comparative Validation of Machine Learning Algorithms for Surgical Workflow and Skill Analysis with the HeiChole Benchmark

View Full Paper
Ask AI
Bookmark
Share

Why the study?

Although surgical workflow analysis achieved high precision in single-center data, the generalizability of phase recognition algorithms in a multi-center setting including more difficult tasks like surgical action and skill remained uninvestigated.

Population

33 laparoscopic cholecystectomy videos from three surgical centers

Comparison

12 teams machine learning algorithms for recognition of phase, action, instrument, or skill assessment

Design

Comparative validation benchmark challenge

Key result

Machine learning algorithms achieved F1-scores between 23.9% and 67.7% for surgical phase recognition, demonstrating that surgical workflow analysis is a promising but unsolved challenge.

Authors

MWMartin WagnerBMBeat P. Müller‐StichAKAnna Kisilenko

Discussion

Loading...

Member takes

Overview

Single-center surgical phase recognition may aid workflow analysis; leaves open multi-center generalizability for clinical systems.

Structured PICO

P
Population
33 laparoscopic cholecystectomy videos from three surgical centers in Germany, totaling 22 hours of operation time, annotated for surgical phases, actions, instruments, and skills.
I
Intervention
Machine learning algorithms for recognition of phase, action, instrument and/or skill assessment (submitted by 12 teams)
O
Outcome
F1-scores for phase recognition, instrument presence detection, action recognition, and average absolute error for skill assessmentsurrogate

Machine learning algorithms for surgical workflow and skill analysis show promise but require further development, as demonstrated by variable performance in a multi-center benchmark challenge.

Limitations

  • Small dataset size of 33 videos
  • Lack of standardized phase definitions in existing datasets
  • Autonomy could not be assessed for surgical skill due to lack of audio and senior surgeon assistance information

Cite This Study

Wagner et al. (2021) studied Laparoscopic cholecystectomy (n=33). Machine learning algorithms was evaluated on F1-score for phase recognition. Machine learning algorithms achieved F1-scores between 23.9% and 67.7% for surgical phase recognition, demonstrating that surgical workflow analysis is a promising but unsolved challenge.

synapsesocial.com/papers/6a93ca5a00ab855628c097b6https://doi.org/10.48550/arxiv.2109.14956
View Full Paper
Ask AI
Bookmark
Share