Related Experiment Video
Updated: Apr 26, 2026
09:21
A Novel Dual-Modal Deep Learning Approach for Real-Time Removal of Hepatic Fluorescence in Indocyanine Green-Guided Laparoscopic Cholecystectomy
Published on: April 17, 2026
30
Few-shot learning for surgical phase recognition: Performance and generalization in cholecystectomy
Flakë Bajraktari1, Robert Asmußen1, Giuliano A Giacoppo1
1Institute of Medical Device Technology, University of Stuttgart, Pfaffenwaldring 9, 70569, Stuttgart, Germany.
Computer Methods and Programs in Biomedicine
|April 24, 2026
Summary
Few-shot learning (FSL) shows promise for automated surgical phase recognition, achieving high accuracy with minimal data. Domain-specific training is key for optimal performance in surgical workflow analysis.
Area of Science:
- Artificial Intelligence in Medicine
- Computer Vision for Surgery
- Machine Learning for Healthcare
Background:
- Automated surgical phase recognition is vital for advancing digital surgery and enhancing surgical assistance systems.
- Traditional deep learning methods struggle with limited annotated surgical data due to privacy and scarcity.
- Few-shot learning (FSL) offers a potential solution to overcome data limitations in surgical phase recognition.
Purpose of the Study:
- To explore the efficacy of few-shot learning (FSL) for automated surgical phase recognition.
- To adapt a transformer-based FSL model for recognizing phases in cholecystectomy videos.
- To evaluate the model's performance in domain-specific and cross-domain settings with minimal labeled data.
Main Methods:
- Adapted a transformer-based few-shot learning (FSL) model for action recognition to surgical phase recognition.
- Utilized the Cholec80 dataset comprising cholecystectomy surgery videos.
- Evaluated performance across three experimental splits, including domain-specific and cross-domain scenarios.
Main Results:
- Achieved test accuracies of 89.0% (Split 1), 75.4% (Split 2), and 49.1% (Split 3).
- Demonstrated strong applicability of FSL to surgical data, even with few labeled examples per phase.
- Highlighted the critical importance of domain-specific training for optimal performance.
Conclusions:
- Few-shot learning (FSL) is a feasible approach for surgical phase recognition in low-label and transfer learning settings.
- FSL provides a promising direction for surgical workflow analysis when annotation resources are constrained.
- Further research can build upon this foundation to develop more robust AI-driven surgical tools.
