Related Experiment Video
Updated: Jan 12, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
MPCNet: multi-scale decomposition network with perception weaving and context-aware fusion for robust 2D medical
Xingyue Ding1, Haiyan Li1, Yiyin Tang2
1School of Information Science and Engineering, Yunnan University, Kunming 650504, People's Republic of China.
Physics in Medicine and Biology
|October 30, 2025
Summary
A new multi-scale network (MPCNet) enhances medical image classification by effectively integrating fine-grained and coarse-grained features. This robust approach achieves state-of-the-art performance across diverse datasets, improving diagnostic accuracy.
Area of Science:
- Medical image analysis
- Artificial intelligence in healthcare
- Computer vision
Background:
- Accurate medical image classification is vital for diagnostics but challenged by global dependency modeling, feature balancing, and multi-scale integration.
- Existing methods struggle with adaptability across diverse datasets and lesion types.
Purpose of the Study:
- To introduce a novel network, MPCNet, for robust 2D medical image classification.
- To address limitations in modeling global dependencies, balancing features, and integrating multi-scale information.
Main Methods:
- Proposed a multi-scale decomposition network (MPCNet) featuring perception weaving and context-aware fusion.
- Employed hierarchical convolutions for fine-grained features and dual-pooling with dynamic attention for coarse-grained features.
- Utilized spatial and channel attention for adaptive fusion of complementary features.
Main Results:
- MPCNet achieved high accuracy on diverse datasets: Ulcerative Colitis (87.54%), Kvasir (81.98%), COVID-19 (73.68%), PAD-UFES-20 (72.81%), and Fetal-Planes-DB (96.57%).
- Demonstrated strong adaptability to various disease types and data characteristics.
- Ablation and comparative studies confirmed the effectiveness of individual modules and overall performance.
Conclusions:
- MPCNet offers a robust and unified solution for medical image analysis.
- The proposed method exhibits superior generalization capability and state-of-the-art performance.
- The code is publicly available for further research and application.
