Radiomics and Machine Learning in Diagnostics of Glial Brain Tumors: a Systematic Review and Meta-Analysis

Area of Science:

• Neuro-oncology
• Medical imaging
• Artificial intelligence

Background:

• Glial tumors are the most common primary brain neoplasms.
• Accurate subtyping is crucial for effective treatment strategies.
• Non-invasive diagnostic methods like neuroimaging and radiomics are increasingly important.

Purpose of the Study:

• To review and analyze the challenges and quality of radiomics and machine learning in diagnosing glial tumors using MRI data.
• To assess the accuracy of these techniques in predicting molecular biomarker status.

Main Methods:

• Systematic literature review and meta-analysis of 42 publications.
• Analysis of studies using radiomics and machine learning on MRI data for glial tumor subtyping.
• Inclusion of various molecular biomarkers: IDH, ATRX, BRAF, H3K27M mutations, TERT promoter mutations, 1p/19q codeletion, MGMT methylation, and Ki-67 index.

Main Results:

• High overall accuracy of 0.86 [0.83; 0.89] for radiomics and machine learning in predicting molecular biomarker status.
• Significant methodological heterogeneity observed across studies.
• Lack of uniform standards for selecting regions of interest (ROI) for feature extraction was a major challenge.

Conclusions:

• Radiomics and machine learning show promise for non-invasive glial tumor diagnostics.
• Methodological heterogeneity, particularly in ROI selection, impedes clinical reproducibility.
• Standardization of radiomics procedures is essential for future research and clinical translation.