Related Experiment Video
Updated: Jun 3, 2026
14:15
Multicolor 3D Printing of Complex Intracranial Tumors in Neurosurgery
Published on: January 11, 2020
CMHF-3DNet: A Transformer-Based Framework for Improved Brain Tumor Segmentation Across Modalities
Abid Hussain1, Wu Jungshen1, Ali Turab1
1School of Software, Northwestern Polytechnical University, Xi'an 710000, China (A.H., W.J., A.T., Y.G., I.A., A.W.).
Academic Radiology
|June 1, 2026
Summary
We developed a novel 3D deep learning model, CMHF-3DNet, for precise brain tumor segmentation in multi-modal MRI. This method improves delineation of critical subregions like the enhancing tumor and tumor core, aiding automated tumor assessment.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuro-oncology
Background:
- Accurate brain tumor subregion segmentation in multi-modal MRI is crucial for clinical applications.
- Challenges include weak boundaries and appearance variability in enhancing tumor (ET) and tumor core segmentation.
Purpose of the Study:
- To introduce Cross-Modal Hierarchical Fusion U-Net 3D (CMHF-3DNet) for improved brain tumor segmentation.
- To address challenges in delineating heterogeneous tumor subregions.
Main Methods:
- A 3D encoder-decoder framework utilizing voxel-wise cross-modal transformer fusion.
- Hierarchy-aware multi-task learning to enforce anatomical consistency (ET ⊂ TC ⊂ WT).
Main Results:
- Evaluated on BraTS 2023, 2024, and 2025 validation datasets.
- Achieved a mean Dice Similarity Coefficient (DSC) of 0.8527 on BraTS 2025 and 0.8526 on BraTS 2023.
- Obtained a mean 95th percentile Hausdorff Distance (HD95) of 6.47 mm on BraTS 2025.
Conclusions:
- CMHF-3DNet demonstrates improved boundary delineation for tumor subregions, especially ET and tumor core.
- The model shows robust performance across multiple BraTS benchmarks.
- Suggests potential for automated brain tumor assessment.
