Radiogenomics of lower-grade gliomas: machine learning-based MRI texture analysis for predicting 1p/19q codeletion...

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• 1p/19q codeletion is a critical biomarker for lower-grade gliomas (LGG).
• Accurate prediction of this status is vital for treatment planning.
• Current methods may be invasive or time-consuming.

Purpose of the Study:

• To evaluate machine learning (ML)-based MRI texture analysis for predicting 1p/19q codeletion in LGGs.
• To compare the performance of various state-of-the-art ML algorithms for this prediction task.

Main Methods:

• Retrospective analysis of 107 LGG patients from a public database.
• Texture features extracted from T2-weighted and contrast-enhanced T1-weighted MRI using LIFEx software.
• Stratified 10-fold cross-validation with minority over-sampling, dimension reduction, and feature selection (ReliefF) were employed. Classification performed using multiple ML algorithms.

Main Results:

• ML algorithms showed statistically significant differences in predictive performance (p < 0.001).
• Neural network, naive Bayes, support vector machine, random forest, and stochastic gradient descent achieved comparable performance (AUC 0.769-0.869, accuracy 80.1-84%).
• The neural network model demonstrated the highest mean AUC (0.869) and accuracy (83.8%).

Conclusions:

• ML-based MRI texture analysis is a promising non-invasive method for predicting LGG 1p/19q codeletion status.
• Multiple ML algorithms can achieve high classification accuracy, with over four-fifths of LGGs correctly classified.
• Volumetric segmentation techniques and careful consideration of feature selection are important for future applications.