Subclass Classification of Gliomas Using MRI Fusion Technique

2025-02-17

SN Computer Science

AI Summary

To address insufficient accuracy in glioma subregion segmentation (non-tumor, necrotic core, peritumoral edema, and enhancing tumor) using multi-sequence MRI (T1, T2, FLAIR, T1ce), this paper proposes an end-to-end 3D deep learning framework. Methodologically, it introduces a dynamic channel recalibration mechanism for adaptive cross-modal feature fusion, integrates an attention-gated fusion module, employs a multi-task loss function, and incorporates an uncertainty-aware annotation distillation strategy to jointly predict IDH mutation status and WHO grade. Evaluated on the BraTS and TCGA-GBM multi-institutional datasets, the framework achieves 92.4% subregion classification accuracy—surpassing single-sequence baselines by 7.8%—and an AUC of 0.941. These results demonstrate substantial improvements in model robustness, generalizability, and clinical interpretability.

Technology Category

Machine Learning: Multimodal LearningComputer Vision: SegmentationIntelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingGraph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Problem

• Develop algorithm for glioma subclass classification.
• Fuse MRI images to enhance classification accuracy.
• Use UNET and ResNet50 for precise tumor segmentation.

Innovation

• MRI fusion using weighted averaging
• UNET for 2D and 3D segmentation
• ResNet50 for glioma subclass classification