Integrating Deep Learning and Radiogenomics: A Novel Approach to Glioblastoma Segmentation and MGMT Methylation...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Radiogenomics integrates imaging and genomic data for glioblastoma (GBM) management.
• Current methods for assessing MGMT promoter methylation require invasive biopsies, causing delays and risks.
• A non-invasive method to predict MGMT methylation status is crucial for personalized glioblastoma treatment.

Purpose of the Study:

• To establish a correlation between radiomic features and MGMT promoter methylation status in glioblastoma.
• To develop a non-invasive, integrated diagnostic paradigm for glioblastoma using radiogenomics.
• To predict MGMT promoter methylation status accurately using advanced AI models.

Main Methods:

• An enhanced U-Net model was utilized for precise brain tumor segmentation (Dice coefficient: 0.889).
• A hybrid classifier combining EfficientNetB0 and ResNet50 was developed to predict MGMT promoter methylation status.
• The model processed segmented tumor volumes to derive radiomic features for classification.

Main Results:

• The proposed framework achieved 95% classification accuracy and an AUC of 0.96 in predicting MGMT promoter methylation status.
• The AI-driven radiomic approach demonstrated superior performance compared to conventional methods.
• High precision in tumor segmentation was achieved, forming a reliable basis for subsequent analysis.

Conclusions:

• Non-invasive radiomic analysis can accurately predict MGMT promoter methylation status in glioblastoma.
• This approach facilitates enhanced patient stratification and personalized treatment selection.
• The findings support the use of radiogenomics to optimize therapeutic strategies and improve glioblastoma patient outcomes.