A Multimodal Knowledge-Based Deep Learning Approach for MGMT Promoter Methylation Identification.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Computational Biology

Background:

• Glioblastoma Multiforme (GBM) is an aggressive brain tumor with poor prognosis.
• O(6)-methylguanine-DNA methyltransferase (MGMT) repair limits chemotherapy efficacy.
• MGMT promoter methylation is a predictive biomarker for GBM treatment response.

Purpose of the Study:

• To develop a non-invasive Deep Learning (DL) method for identifying MGMT promoter methylation in GBM.
• To leverage Magnetic Resonance Imaging (MRI) data for this detection.
• To improve upon existing methods for biomarker identification.

Main Methods:

• A Convolutional Neural Network (CNN) was designed to analyze FLAIR MRI series.
• An unsupervised Knowledge-Based filter was used to pre-select suspicious regions using FLAIR and T1-weighted images.
• The DL model was trained and validated on two public GBM datasets.

Main Results:

• The proposed DL approach achieved results comparable to, or better than, existing methods.
• The model demonstrated high efficiency, utilizing less than 0.29% of competitor parameters.
• eXplainable AI (XAI) analysis was performed to assess clinical applicability.

Conclusions:

• DL analysis of MRI data is a viable, efficient method for detecting MGMT promoter methylation in GBM.
• This approach offers a promising, less invasive alternative to current diagnostic procedures.
• Further eXplainable AI (XAI) analysis supports the clinical translation of this DL-based tool.