Related Experiment Video
Updated: Sep 11, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.8K
An Adaptive Multi-Stage and Adjacent-Level Feature Integration Network for Brain Tumor Image Segmentation
Jiwen Zhou1, Yulun Wu2, Yue Xu1
1School of Mechatronic Engineering and Automation, Shanghai University, Shanghai, 200444, China.
Interdisciplinary Sciences, Computational Life Sciences
|August 14, 2025
Summary
A novel network, MAI-Net, enhances brain tumor segmentation in MRI by effectively handling blurred boundaries and small lesions. This method significantly improves accuracy, outperforming existing techniques on benchmark datasets.
Area of Science:
- Medical image analysis
- Artificial intelligence in healthcare
- Neuroimaging
Background:
- Brain tumor segmentation from MRI is vital for clinical decision-making.
- Current Convolutional Neural Networks (CNNs) and transformer methods face challenges with blurred boundaries, small lesions, and interwoven regions.
- Existing approaches often struggle with complex segmentation scenarios in medical imaging.
Purpose of the Study:
- To introduce a new network, MAI-Net, designed to overcome limitations in brain tumor MRI segmentation.
- To improve segmentation accuracy by effectively addressing issues like blurred boundaries and small lesion volumes.
- To enhance the performance of medical image segmentation tasks through advanced feature integration.
Main Methods:
- Developed MAI-Net, a dual-branch, multi-level network incorporating three novel modules: Stage-Level Multi-scale Feature Extraction (SMFE), Adjacent-Level Feature Fusion (AFF), and Multi-Stage Feature Fusion (MFF).
- The SMFE module captures multi-scale details for improved edge and small lesion detection.
- The AFF and MFF modules facilitate cross-level information exchange and integration for enhanced accuracy in complex and small-volume regions.
Main Results:
- MAI-Net demonstrated superior performance on the BraTS2020 and BraTS2021 datasets, achieving better Dice and HD95 metrics compared to existing methods.
- Generalization experiments on an ischemic stroke dataset confirmed MAI-Net's robustness across different medical image segmentation tasks.
- The proposed network effectively handles challenges such as blurred boundaries, small lesion volumes, and interwoven regions.
Conclusions:
- MAI-Net offers significant advantages for brain tumor segmentation, effectively addressing domain-specific challenges.
- The network's architecture provides superior accuracy and robustness in medical image segmentation.
- MAI-Net represents a promising advancement for clinical applications requiring precise tumor delineation.
