Related Experiment Video
Updated: Sep 13, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Multi-scale interaction and locally enhanced bridging network for medical image segmentation.
Zhiyong Huang1, Shiyao Zhou2, Zhi Yu2
1Key Laboratory of Dependable Service Computing in Cyber Physical Society (Chongqing University), Ministry of Education of China, Chongqing University, Chongqing 400044, China; School of Microelectronics and Communication Engineering, Chongqing University, Chongqing 400044, China.
MILENet improves medical image segmentation accuracy by effectively integrating multi-scale features and enhancing local details. This novel network achieves state-of-the-art results across diverse imaging modalities and anatomical structures.
Area of Science:
- Medical Image Analysis
- Computer Vision
- Artificial Intelligence
Background:
- Accurate medical image segmentation is vital for diagnosis.
- Current Convolutional Neural Networks (CNNs) and Transformers struggle with integrating multi-scale features due to simple skip connections.
- This limitation leads to misalignment between global and local features, impacting segmentation accuracy, especially for contours and details.
Purpose of the Study:
- To introduce MILENet, a novel Multi-scale Interaction and Locally Enhanced Bridging Network.
- To address the limitations of existing methods in effectively integrating multi-scale features and preserving fine-grained details.
- To improve the accuracy and generalizability of automatic medical image segmentation.
Main Methods:
- MILENet employs a context bridge with a multi-scale interaction module for feature reorganization and global correlation.
- A local enhancement module, featuring dilated coordinate attention and a cascaded convolutional FFN, improves local context modeling.
- A source-driven connection mechanism preserves detailed information across layers for enhanced decoder reconstruction.
Main Results:
- MILENet achieved state-of-the-art performance on abdominal CT, cardiac MRI, and colonoscopy datasets.
- The network demonstrated effectiveness in segmenting both large organs (CT/MRI) and small polyps (endoscopic images).
- Results indicate strong generalizability across diverse anatomical structures and imaging conditions.
Conclusions:
- MILENet effectively aligns multi-scale features and enhances local details, significantly improving segmentation accuracy.
- The proposed network offers a robust solution for various medical image segmentation tasks.
- The open-source code facilitates further research and application in medical image analysis.
