Deep Learning and Multi-Sensor Fusion for Glioma Classification Using Multistream 2D Convolutional Networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Glioma grading is crucial for treatment planning, but single-sensor MRI analysis has limitations.
• Multi-sensor MRI (T1-MRI, T2-MRI, FLAIR) offers complementary information for brain tumor characterization.
• Existing methods often rely on single-sensor data, limiting diagnostic performance.

Purpose of the Study:

• To develop and evaluate a novel multi-stream deep Convolutional Neural Network (CNN) for enhanced glioma grading.
• To leverage sensor fusion techniques by integrating data from multiple MRI modalities (T1-MRI, T2-MRI, FLAIR).
• To improve the accuracy of classifying low/high grade gliomas and gliomas with/without 1p19q codeletion.

Main Methods:

• Proposed a novel multi-stream deep CNN architecture for feature extraction and fusion.
• Implemented sensor fusion by aggregating features from T1-MRI, T2-MRI, and FLAIR images.
• Utilized 2D brain image slices with 2D augmentation to mitigate overfitting.

Main Results:

• Achieved high test accuracy: 90.87% for low/high grade glioma classification.
• Attained 89.39% accuracy for classifying glioma with/without 1p19q codeletion.
• Demonstrated superior performance compared to several existing methods through comparative experiments.

Conclusions:

• The proposed multi-stream CNN with sensor fusion significantly enhances glioma grading accuracy.
• Multi-modal MRI data integration via deep learning offers a promising approach for brain tumor subclassification.
• The method provides a robust and accurate tool for clinical decision support in neuro-oncology.