Comparison of MRI Sequences to Predict ATRX Status Using Radiomics-Based Machine Learning

Area of Science:

• Neuro-oncology
• Radiology
• Machine Learning

Background:

• ATRX (alpha-thalassemia/mental retardation syndrome X-linked) is a key molecular marker in the 2021 WHO classification of adult-type diffuse glioma.
• Accurate non-invasive prediction of ATRX mutation status is crucial for glioma management.

Purpose of the Study:

• To develop and validate radiomics-based machine learning models for non-invasively predicting ATRX mutation status in diffuse gliomas.
• To identify the optimal MRI sequence for this prediction task.

Main Methods:

• Retrospective analysis of MRI data (T1w, T1w with contrast, T2w, FLAIR) from patients with histologically confirmed glioma.
• Extraction of radiomics features from hand-delineated regions of interest.
• Development of machine learning models using Lasso regression with 100-fold cross-validation.

Main Results:

• The T2w MRI sequence demonstrated the highest suitability for predicting ATRX mutations.
• A seven-feature model using T2w images achieved a mean AUC of 0.831, mean accuracy of 0.746, mean sensitivity of 0.772, and mean specificity of 0.697.
• FLAIR sequences were found to be the least suitable for ATRX mutation prediction.

Conclusions:

• T2w MRI sequences are the most effective among common MRI sequences for non-invasively predicting ATRX mutation status in diffuse gliomas using radiomics and machine learning.
• Radiomics models applied to T2w MRI offer a promising approach for assessing ATRX mutation status without invasive procedures.