Multiclass Radiomics-Based Prediction of BRAF Mutation Status in Pediatric Low-Grade Gliomas Using Multisequence MRI.

Area of Science:

• Oncology
• Radiology
• Machine Learning
• Pediatric Neuro-oncology

Background:

• Pediatric low-grade gliomas (pLGGs) are the most common pediatric brain tumors.
• BRAF alterations, including KIAA1549-BRAF fusions and BRAF V600E mutations, are frequent in pLGGs and influence treatment decisions.
• Accurate prediction of BRAF mutation status is crucial for effective pLGG management.

Purpose of the Study:

• To develop and evaluate multiclass radiomics-based machine learning models for predicting BRAF mutation status (fusion, V600E, non-BRAF) in pediatric low-grade gliomas.
• To assess the performance of models using single and multi-sequence MRI data.
• To compare the predictive power of clinical-only, radiomics-only, and combined models.

Main Methods:

• Retrospective analysis of pre-surgical MRI scans from 511 pediatric patients with pLGG.
• Manual tumor segmentation and radiomics feature extraction using PyRadiomics.
• Training Random Forest classifiers for three-class BRAF status prediction.
• Evaluation using leave-one-out cross-validation and comparison of single-sequence vs. multi-sequence approaches.

Main Results:

• FLAIR sequences demonstrated the highest performance (AUC 0.82) among single sequences, followed by T2WI (0.80), ADC (0.77), and CE-T1WI (0.75).
• Combined clinical-radiomics models consistently outperformed single-source models.
• In a cohort of 180 patients with all four sequences, multi-sequence radiomics (feature concatenation and ensemble modeling) achieved macro-AUC of 0.79, outperforming single-sequence approaches.
• FLAIR-derived features were dominant, but integrating T2, ADC, and CE-T1WI improved classification balance.

Conclusions:

• Machine learning models leveraging multi-sequence MRI show promise for noninvasive prediction of BRAF mutation status in pLGG.
• While FLAIR is the best single sequence, integrating multiple sequences enhances prediction performance and balance.
• Multi-sequence radiomics offers a valuable tool for precision treatment guidance in pLGG, especially when tissue biopsy is not feasible.