Classification of High-Grade Glioma into Tumor and Nontumor Components Using Support Vector Machine.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Current Response Assessment in Neuro-Oncology (RANO) criteria for high-grade brain tumors use conventional MRI, which may not reliably distinguish tumor from non-tumor tissue.
• This limitation can impact accurate assessment of treatment response and disease progression.

Purpose of the Study:

• To develop and validate an automated classification method to segment enhancing and non-enhancing lesion areas into tumor and non-tumor components.
• To improve the accuracy of therapy response assessment in high-grade gliomas using advanced imaging techniques.

Main Methods:

• A support vector machine classifier was trained on 140 MRI scans from patients with high-grade gliomas and brain metastases.
• The classifier utilized T1-weighted, FLAIR, and dynamic-contrast-enhancing MRI parameters, including plasma volume and bolus-arrival-time.
• Classification accuracy was validated using 2-fold cross-validation and MR spectroscopy, with longitudinal component volume changes compared to RANO criteria.

Main Results:

• Advanced imaging parameters effectively differentiated tumor and non-tumor components with 100% sensitivity and specificity.
• Longitudinal changes in segmented component volumes correlated with RANO criteria, identifying tumor progression in 16% of patients earlier than conventional assessments.
• Bevacizumab treatment in seven patients revealed a shift towards an infiltrative progression pattern.

Conclusions:

• An automated classification method based on advanced imaging, termed segmented RANO criteria, reliably distinguishes tumor from non-tumor components in high-grade gliomas.
• This novel approach offers improved therapy-response assessment and earlier detection of tumor progression, potentially leading to more timely treatment adjustments.