Artificial intelligence and advanced MRI techniques: A comprehensive analysis of diffuse gliomas

Area of Science:

• Neuroradiology
• Medical Imaging
• Artificial Intelligence

Background:

• Diffuse gliomas present complex heterogeneity requiring advanced imaging like MRI.
• The UCSF-PDGM dataset, comprising 501 subjects, was utilized for comprehensive analysis.
• Understanding tumor characteristics is crucial for optimizing patient outcomes.

Purpose of the Study:

• To analyze diffuse gliomas using MRI techniques, radiomics, and artificial intelligence (AI).
• To predict patient outcomes and tumor grades.
• To identify influential radiomic features for improved tumor characterization.

Main Methods:

• A dataset of 501 diffuse glioma patients was analyzed using a comprehensive MRI protocol.
• Over 82,800 radiomic features were extracted from nine segmentations across eight MRI sequences.
• Neural network and XGBoost models were trained for prediction, with SHAP analysis for feature importance.

Main Results:

• High accuracy was achieved in automated tumor segmentation compared to manual methods.
• The neural network model demonstrated 0.9500 accuracy in predicting WHO tumor grades (necrotic label).
• Key radiomic features identified include 3D First Order mean, Original Shape Sphericity, and Original Shape Elongation.

Conclusions:

• AI and radiomics significantly impact neuroradiology, enabling reliable tumor segmentation and identification of critical radiomic features.
• While predicting patient survival remains challenging, the study highlights AI's potential in enhancing diagnostic accuracy.
• Broader datasets with diverse MRI sequences are needed to further improve patient outcomes.