A Multi-Task EfficientNetV2S Approach with Hierarchical Hybrid Attention for MRI Enhancing Brain Tumor Segmentation...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain tumors pose significant clinical challenges due to heterogeneity and complex MRI characteristics.
• Current automated MRI analysis methods struggle with noise propagation, limited feature integration, and isolated task optimization.

Purpose of the Study:

• To develop an improved deep learning framework for enhanced brain tumor segmentation and classification from MRI data.
• To address limitations of existing automated methods in handling noise and integrating spatial-channel information.

Main Methods:

• Utilized an EfficientNetV2S backbone integrated with a novel Hierarchical Hybrid Attention (HHA) mechanism.
• Implemented HHA with coupled global-context and local-spatial pathways, using a fusion gate for interaction modeling.
• Incorporated multi-scale dilated blocks and applied shared representation learning for joint segmentation and classification on a multiclass brain tumor MRI dataset.

Main Results:

• Achieved a Dice score of 92.25% and Jaccard index of 86% for tumor segmentation.
• Obtained 99.53% accuracy for classification, with precision, recall, and F1 scores near 99%.
• Demonstrated sharper tumor boundaries, improved noise suppression in segmentation, and more robust classification discrimination.

Conclusions:

• The proposed framework effectively overcomes limitations in brain tumor MRI analysis.
• The integrated HHA mechanism and shared representation learning enhance segmentation quality and classification accuracy.
• The model shows strong clinical utility for precise brain tumor delineation and diagnosis.