Related Experiment Video
Updated: Jul 2, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
An N-Shaped Lightweight Network with a Feature Pyramid and Hybrid Attention for Brain Tumor Segmentation
Mengxian Chi1, Hong An1, Xu Jin1
1School of Computer Science and Technology, University of Science and Technology of China, Hefei 230026, China.
Entropy (Basel, Switzerland)
|February 23, 2024
Summary
This study introduces a novel lightweight neural network for brain tumor segmentation, improving accuracy and efficiency. The new model effectively handles class imbalance, offering a promising solution for clinical applications.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Brain tumor segmentation using neural networks faces challenges in accuracy, real-time performance, and class imbalance.
- Accurate segmentation is critical for effective clinical diagnosis and treatment planning.
Purpose of the Study:
- To develop a novel, efficient, and accurate neural network for brain tumor segmentation.
- To address the challenges of diverse tumor shapes/sizes and class imbalance in segmentation tasks.
Main Methods:
- Proposed a novel N-shaped lightweight network integrating multiple feature pyramid paths and U-Net architectures.
- Incorporated hybrid attention mechanisms, specifically channel attention, into depth-wise separable convolutions for enhanced efficiency.
- Introduced a combination loss function (weighted cross-entropy and dice loss) to manage class imbalance.
Main Results:
- The proposed network achieved superior segmentation accuracy compared to state-of-the-art methods across four public datasets (UCSF-PDGM, BraTS 2021, BraTS 2019, MSD Task 01).
- Demonstrated favorable computational efficiency alongside improved segmentation performance.
Conclusions:
- The developed lightweight network offers a promising approach for clinical brain tumor segmentation.
- The integration of attention mechanisms and a specialized loss function effectively enhances segmentation accuracy and efficiency.
