Predicting MGMT Promoter Methylation in Diffuse Gliomas Using Deep Learning with Radiomics.

Area of Science:

• Neuro-oncology
• Medical imaging
• Artificial intelligence

Background:

• 6-methylguanine-DNA methyltransferase (MGMT) promoter methylation is a key prognostic marker in diffuse gliomas.
• Accurate prediction of MGMT methylation status is crucial for guiding treatment decisions.

Purpose of the Study:

• To develop and evaluate a deep learning approach using MRI radiomics for predicting MGMT promoter methylation in diffuse gliomas.
• To assess the diagnostic performance of different MRI sequences and regions of interest for this prediction task.

Main Methods:

• A retrospective study involving 111 diffuse glioma patients.
• Extraction of radiomics features from T1WI, T2WI, ADC, and T1CE MR images across whole tumor and tumor core regions.
• Development of a deep learning model (residual network) trained and validated using five-fold cross-validation.

Main Results:

• The combined T1CE and ADC model using the tumor core ROI achieved the highest performance.
• This best model demonstrated an average accuracy of 0.91 and an Area Under the Curve (AUC) of 0.90.
• The study identified key radiomics features contributing to the predictive model's accuracy.

Conclusions:

• Deep learning applied to MRI radiomics offers a highly accurate and feasible method for predicting MGMT promoter methylation in diffuse gliomas.
• The tumor core region and specific MRI sequences (T1CE, ADC) are particularly valuable for this non-invasive prediction.