Related Experiment Video
Updated: Jan 12, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.3K
Efficient frequency-decomposed transformer via large vision model guidance for surgical image desmoking
Jiaao Li1, Diandian Guo2, Youyu Wang1
1Thoracic Surgery, The First Affiliated Hospital of Shenzhen University, Guangdong, China.
Summary
This study introduces SmoRestor, a novel framework for surgical image restoration. It effectively removes surgical smoke, enhancing visualization during minimally invasive procedures.
Area of Science:
- Medical Imaging
- Computer Vision
- Surgical Technology
Background:
- Surgical image restoration is crucial for minimally invasive surgery, often hindered by surgical smoke.
- Current desmoking algorithms and learning strategies show limited progress.
Purpose of the Study:
- To develop an efficient framework for surgical image desmoking.
- To analyze the characteristics of surgical smoke degradation for improved restoration.
Main Methods:
- Proposed SmoRestor, a frequency-aware Transformer framework.
- Introduced a Fourier-embedded neighborhood attention transformer for spatial and frequency domain analysis.
- Utilized semantic priors from large vision models and a transfer learning paradigm for content-degradation separation.
Main Results:
- SmoRestor demonstrated substantial improvements in quantitative performance and visual quality.
- The framework effectively separates anatomical structures from complex degradations.
- Experimental results validated on public and in-house datasets.
Conclusions:
- The proposed SmoRestor framework offers an effective solution for surgical image desmoking.
- The frequency-aware approach and transfer learning enhance the ability to restore clear surgical visuals.
- This work advances the field of surgical image restoration for improved surgical outcomes.
