Integrated brain tumor segmentation and MGMT promoter methylation status classification from multimodal MRI data...

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Artificial intelligence in medicine

Background:

• Glioblastoma multiforme (GBM) is an aggressive brain tumor with limited survival rates.
• O(6)-methylguanine-DNA-methyltransferase (MGMT) promoter methylation status is critical for GBM treatment planning and predicting chemosensitivity.
• Current MGMT status determination requires invasive tissue sampling and genetic testing, which are time-consuming.

Purpose of the Study:

• To develop and evaluate a non-invasive pipeline for assessing MGMT promoter status in GBM patients.
• To leverage magnetic resonance imaging (MRI) data for predicting MGMT status, thereby informing treatment and surgical planning.
• To explore the potential of AI-driven methods in analyzing tumor sub-regions and molecular data non-invasively.

Main Methods:

• A two-phase pipeline was proposed, utilizing the BraTS2021 and MGMT promoter status classification datasets.
• Phase 1: A 3D Residual U-Net (3D ResU-Net) segmented brain tumors into sub-regions using stacked MRI modalities.
• Phase 2: A 3D ResNet10 model classified MGMT promoter status based on segmented tumor voxels.

Main Results:

• The segmentation phase achieved promising Dice scores: 0.81 for tumor core (TC), 0.84 for whole tumor (WT), and 0.80 for enhancing tumor (ET).
• The classification phase yielded a ROC-AUC score of 0.66 on the internal validation set.
• The pipeline demonstrated effective segmentation of tumor sub-regions and prediction of MGMT promoter status.

Conclusions:

• The developed pipeline shows potential as a non-invasive tool to assist neuro-oncologists in GBM diagnosis and treatment planning.
• AI-driven analysis of MRI data can provide insights into tumor characteristics and molecular status.
• Further clinical validation is necessary to establish the real-world applicability of this research-stage approach.