Hierarchical non-negative matrix factorization to characterize brain tumor heterogeneity using multi-parametric MRI.

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Biophysics

Background:

• Brain tumor characterization, especially in high-grade gliomas, is complex due to heterogeneous tissue types within regions.
• Accurate differentiation of tumor substructures like viable tumor, necrosis, and edema is crucial for patient management.

Purpose of the Study:

• To develop and validate a method for detecting and characterizing intra-tumoral tissue substructures in gliomas.
• To assess the added value of multi-parametric MRI (MP-MRI) combined with hierarchical non-negative matrix factorization (hNMF) for tissue characterization.

Main Methods:

• A cohort of 24 glioma patients (10 LGG, 14 HGG) underwent a comprehensive MP-MRI protocol.
• MP-MRI parameters including conventional MRI, PWI, DKI, and MRSI were acquired and analyzed using hNMF.
• Tissue segmentation was performed voxel-by-voxel, and results were validated using Dice scores and correlation coefficients against radiologist segmentation.

Main Results:

• The hNMF method achieved high accuracy in segmenting tumor substructures in high-grade gliomas (mean Dice scores of 78-85%) and low-grade gliomas (mean Dice score of 85%).
• Strong correlations were observed between segmented tissues and MRI features (mean correlation coefficients of 0.91-0.97).
• Analysis of reduced MRI datasets demonstrated the unique contribution of individual MRI modalities.

Conclusions:

• hNMF applied to MP-MRI provides effective and accurate tissue characterization in gliomas.
• This method offers significant potential for improving diagnosis, treatment planning, and patient follow-up.
• The study highlights the importance of multi-parametric data for detailed brain tumor analysis.