Systematic Literature Review of Machine Learning Algorithms Using Pretherapy Radiologic Imaging for Glioma Molecular...

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Glioma molecular subtypes are crucial prognostic indicators for patient survival and treatment decisions.
• Current pathology-based molecular diagnosis is invasive and limits neoadjuvant therapy options due to tumor heterogeneity.

Approach:

• A systematic review was conducted to identify and evaluate algorithms predicting glioma molecular subtypes using Magnetic Resonance (MR) imaging.
• 12,318 abstracts were screened, and 85 articles met inclusion criteria following PRISMA guidelines.

Key Points:

• Machine learning models demonstrated strong performance in predicting isocitrate dehydrogenase (IDH) mutation status (AUC 0.88 internal, 0.86 external) and O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation (AUC 0.79 internal, 0.89 external).
• Despite high accuracy in internal and limited external validation, all reviewed studies exhibited high bias according to the Prediction model Risk Of Bias Assessment Tool (PROBAST).

Conclusions:

• While AI-driven MR imaging shows potential for non-invasive glioma subtyping, widespread clinical adoption is challenged by insufficient external validation and inherent biases in current algorithms.
• Further research focusing on robust external validation and bias mitigation is essential for translating these predictive techniques into routine clinical practice.