Parameter-efficient framework for surgical action triplet recognition.

Area of Science:

• Computer Vision
• Medical Imaging Analysis
• Surgical Technology

Background:

• Surgical action triplet recognition is crucial for clinical decision-making.
• High similarity among action triplets poses a significant challenge.
• Previous methods required large models, leading to high computational costs.

Purpose of the Study:

• To develop a novel framework for accurate and computationally efficient surgical action triplet recognition.
• To address the challenge of high similarity among action triplets.
• To reduce the computational burden associated with prior methods.

Main Methods:

• Proposed the Lite and Mega Models (LAM) framework.
• Utilized a CNN-based fully fine-tuned model (LAM-Lite) and a parameter-efficient Transformer-based foundation model (LAM-Mega).
• Introduced temporal multi-label data augmentation for robust feature extraction.

Main Results:

• LAM outperformed prior methods on the CholecT50 dataset across various parameter scales.
• Achieved a mean average precision (mAP) of 42.1%, a 3.6% improvement over the state of the art.
• Demonstrated superior performance with fewer tunable parameters.

Conclusions:

• The LAM framework effectively balances accuracy and computational efficiency through structural design and foundational model capabilities.
• The proposed approach offers a significant advancement in surgical action recognition.
• Source code is publicly available for further research and development.