Related Experiment Video
Updated: Oct 9, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
548
MSRF-Net: A Multi-Scale Residual Fusion Network for Biomedical Image Segmentation
IEEE Journal of Biomedical and Health Informatics
|December 23, 2021
Summary
This study introduces the Multi-Scale Residual Fusion Network (MSRF-Net) for improved medical image segmentation. The novel architecture effectively handles variable object sizes and small datasets, outperforming existing methods.
Area of Science:
- Biomedical image analysis
- Deep learning for medical imaging
- Computer vision in healthcare
Background:
- Convolutional neural networks (CNNs) have advanced biomedical image segmentation but struggle with variable object sizes and limited datasets.
- Existing multi-scale fusion methods often employ complex models unsuitable for specific medical imaging challenges.
- Efficient and accurate segmentation of medical images remains a critical need, especially for small or biased datasets.
Purpose of the Study:
- To propose a novel deep learning architecture, the Multi-Scale Residual Fusion Network (MSRF-Net), specifically designed for medical image segmentation.
- To address the limitations of existing methods in handling variable object sizes and small, biased datasets.
- To improve the accuracy and efficiency of biomedical image segmentation.
Main Methods:
- Developed a novel Multi-Scale Residual Fusion Network (MSRF-Net) architecture.
- Introduced a Dual-Scale Dense Fusion (DSDF) block to exchange multi-scale features with varying receptive fields.
- Employed sequential DSDF blocks within the MSRF sub-network for effective multi-scale fusion, preserving resolution and enhancing feature propagation.
Main Results:
- MSRF-Net demonstrated superior performance on four public biomedical datasets, outperforming state-of-the-art methods.
- Achieved high Dice Coefficients (DSC): 0.9217 (Kvasir-SEG), 0.9420 (CVC-ClinicDB), 0.9224 (2018 Data Science Bowl), and 0.8824 (ISIC-2018).
- Generalizability tests yielded DSC of 0.7921 (CVC-ClinicDB) and 0.7575 (Kvasir-SEG).
Conclusions:
- MSRF-Net effectively segments medical images, accurately capturing object variabilities even with small or biased datasets.
- The proposed DSDF block and MSRF architecture significantly improve information flow and feature propagation for enhanced segmentation accuracy.
- MSRF-Net represents a significant advancement in medical image segmentation, offering improved performance and generalizability across diverse datasets.
