Related Experiment Video
Updated: Aug 6, 2025
07:46
Author Spotlight: Revolutionizing Remote Surgery with Augmented Reality and Robotics for Enhanced Precision and Accessibility
Published on: August 9, 2024
799
TRandAugment: temporal random augmentation strategy for surgical activity recognition from videos.
Sanat Ramesh1,2, Diego Dall'Alba3, Cristians Gonzalez4,5
1Altair Robotics Lab, University of Verona, 37134, Verona, Italy. sanat.ramesh@univr.it.
Summary
A new method, TRandAugment, enhances surgical video analysis by applying consistent random transformations to temporal segments. This automated approach improves surgical phase and step recognition accuracy by 1-6% over existing methods.
Area of Science:
- Medical image analysis
- Computer-assisted interventions
- Surgical robotics
Background:
- Automatic recognition of surgical activities from intraoperative videos is vital for intelligent support systems.
- Deep learning models for surgical video analysis benefit from data augmentation to improve generalization.
- Augmenting still images is established, but extending to videos requires addressing the temporal dimension and challenges of long, complex surgical activities.
Purpose of the Study:
- To introduce TRandAugment, a simplified, automated data augmentation technique specifically for long surgical videos.
- To address the complexities of temporal data augmentation in surgical video analysis.
- To enhance the generalization and performance of deep learning models in surgical activity recognition.
Main Methods:
- TRandAugment treats surgical videos as sequences of temporal segments, applying consistent random transformations within each segment.
- An end-to-end spatiotemporal model, integrating a Convolutional Neural Network (CNN - ResNet50) with a Temporal Convolutional Network (TCN), was trained using the proposed augmentation method.
- The method was evaluated on the Bypass40 and CATARACTS surgical video datasets for surgical phase and step recognition tasks.
Main Results:
- TRandAugment demonstrated effectiveness on two distinct surgical video datasets and two recognition tasks.
- The proposed method achieved a performance improvement of 1-6% compared to state-of-the-art methods utilizing manual augmentations.
- This highlights the advantage of automated temporal augmentation over manual strategies for surgical videos.
Conclusions:
- A simplified and automated augmentation method, TRandAugment, has been developed for long surgical videos.
- Validation across different datasets and tasks confirms the significance of temporal augmentation strategies for surgical video analysis.
- The findings underscore the need for specialized temporal augmentation techniques to advance intelligent surgical support systems.
