Glioma Grading by Integrating Radiomic Features from Peritumoral Edema in Fused MRI Images and Automated Machine...

Area of Science:

• Neuroimaging
• Radiomics
• Machine Learning

Background:

• Glioma grading is crucial for treatment planning.
• Accurate and non-invasive grading methods are needed.
• Radiomics offers potential for quantitative image analysis.

Purpose of the Study:

• To assess the utility of peritumoral edema radiomic features from fused MRI sequences for glioma grading.
• To enhance machine learning-based glioma classification performance.
• To investigate Laplacian Re-decomposition (LRD) for fusing multimodal MRI sequences.

Main Methods:

• Utilized the BraTS 2023 dataset.
• Fused multimodal MRI sequences using Laplacian Re-decomposition (LRD).
• Extracted radiomic features from peritumoral edema, selected using Boruta, and optimized with TPOT for classification.

Main Results:

• LRD produced high-quality fused MRI images.
• Boruta algorithm identified informative radiomic features from edema regions.
• A Stochastic Gradient Descent (SGD) classifier trained on fused T1Gd+FLAIR images achieved high performance (Accuracy=0.96, AUC=1.0).

Conclusions:

• Peritumoral edema radiomic features from fused MRI images show distinct, grade-specific patterns.
• This approach provides a non-invasive, accurate, and rapid method for glioma grade classification.
• Fused MRI sequences enhance the diagnostic value of radiomics in neuro-oncology.