Related Experiment Video
Updated: Nov 28, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.2K
Fusion based on attention mechanism and context constraint for multi-modal brain tumor segmentation
Tongxue Zhou1, Stéphane Canu2, Su Ruan3
1Université de Rouen Normandie, LITIS - QuantIF, Rouen 76183, France; INSA de Rouen, LITIS - Apprentissage, Rouen 76800, France; Normandie Univ, INSA Rouen, UNIROUEN, UNIHAVRE, LITIS, France.
Summary
This study introduces a novel deep learning network for 3D brain tumor segmentation using multi-sequence MRI. The efficient three-stage approach improves accuracy while reducing computational parameters for better tumor region identification.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Accurate brain tumor segmentation from multi-sequence MRI is crucial for diagnosis and treatment planning.
- Existing deep learning methods often require significant computational resources and parameter tuning.
Purpose of the Study:
- To develop an efficient and accurate 3D deep learning network for brain tumor segmentation using multi-sequence MRI.
- To reduce the number of parameters and computational cost in deep learning models for medical image analysis.
Main Methods:
- A three-stage deep learning network combining 3D U-Net, attention mechanisms, and a novel loss function.
- Stage 1: Initial segmentation and context constraint generation using 3D U-Net.
- Stage 2: Multi-sequence MRI fusion under constraints with attention, followed by a specialized loss function for multi-class segmentation.
- Stage 3: Refinement of segmentation results using a second 3D U-Net with reduced initial filters.
Main Results:
- The proposed network achieved promising results on the BraTS 2017 dataset.
- Evaluated metrics include Dice score and Hausdorff distance, indicating high segmentation accuracy.
- The method demonstrated a significant reduction in the number of trainable parameters and memory requirements.
Conclusions:
- The developed three-stage deep learning network offers an efficient and effective solution for 3D brain tumor segmentation from multi-sequence MRI.
- The approach shows potential for clinical application by balancing accuracy with reduced computational demands.
- Further validation on diverse datasets is warranted to confirm generalizability.
