Related Experiment Video
Updated: Aug 3, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
470
The Lighter the Better: Rethinking Transformers in Medical Image Segmentation Through Adaptive Pruning
IEEE Transactions on Medical Imaging
|April 7, 2023
Summary
This study introduces APFormer, a novel lightweight network that uses adaptive pruning to enhance transformer models for medical image segmentation. APFormer reduces computational complexity and improves performance, offering a more efficient approach to medical imaging analysis.
Area of Science:
- Medical Image Analysis
- Computer Vision
- Artificial Intelligence
Background:
- Vision transformers show promise in medical image analysis but suffer from high computational costs and redundancy.
- Existing hybrid and transformer-based methods often overlook efficiency challenges.
Purpose of the Study:
- To propose APFormer, a lightweight hybrid network employing adaptive pruning for medical image segmentation.
- To address the computational complexity and redundancy issues in transformer models for medical imaging.
Main Methods:
- Introduced self-regularized self-attention (SSA) for improved dependency establishment.
- Incorporated Gaussian-prior relative position embedding (GRPE) for better position learning.
- Implemented adaptive pruning (query-wise and dependency-wise) to reduce computations and enhance perception.
Main Results:
- APFormer achieved prominent segmentation performance on benchmark datasets.
- The proposed method significantly reduced parameters and GFLOPs compared to state-of-the-art methods.
- Ablation studies confirmed adaptive pruning as a versatile, plug-and-play module.
Conclusions:
- APFormer offers an effective and efficient solution for medical image segmentation using pruned transformers.
- Adaptive pruning demonstrates potential for improving other transformer-based medical imaging models.
