Quality Assessment of MRI-Radiomics-Based Machine Learning Methods in Classification of Brain Tumors: Systematic...

Area of Science:

• Neuro-oncology
• Medical Imaging Analysis
• Machine Learning in Medicine

Background:

• Brain tumors pose diagnostic challenges in oncology.
• Radiomics extracts quantitative imaging features for analysis.
• Methodological quality of radiomic studies requires assessment.

Purpose of the Study:

• Systematically review radiomic studies on brain tumors.
• Assess the quality of radiomic methodologies using the Radiomics Quality Score (RQS).
• Evaluate the potential of radiomics for glioma classification.

Main Methods:

• Systematic literature search of PubMed for radiomics studies on brain tumors (2015 onwards).
• Inclusion/exclusion criteria applied to 300 identified articles, with 18 selected for synthesis.
• Radiomic features used to train and validate machine learning models for glioma classification.

Main Results:

• 18 studies met inclusion criteria, demonstrating radiomics' potential in glioma classification.
• Various imaging modalities (MRI, PET/CT, ASL, DTI) were used.
• Machine learning algorithms (deep learning, SVM, random forests) achieved high classification accuracies.

Conclusions:

• Radiomics-based machine learning shows high accuracy in glioma classification, surpassing traditional methods.
• Further validation and standardization are crucial for clinical integration.
• Open science practices in radiomics enhance transparency and collaboration.