Estimating Local Cellular Density in Glioma Using MR Imaging Data

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Gliomas exhibit increased cellular density in both tumor bulk and infiltration zones.
• Altered cellular density affects imaging findings, but quantitative estimation from imaging remains challenging.

Purpose of the Study:

• To identify optimal MR imaging and processing techniques for quantitative, spatially specific cellular density estimation in gliomas.

Main Methods:

• Prospective collection of preoperative MR imaging (anatomic, diffusion, perfusion, permeability) and stereotactic biopsy histopathology from glioma patients.
• Application of machine learning methodologies to estimate cellular density from MR image intensity data, using biopsy measurements as the ground truth.

Main Results:

• Random forest models achieved R² = 0.59 using four imaging sequences (T2, fractional anisotropy, CBF, permeability AUC).
• Conventional MR imaging inputs (T1, T2, FLAIR) resulted in slightly lower performance (R² = 0.52).
• Spatially specific cellular density maps were generated as outputs.

Conclusions:

• Moderate-to-strong correlations exist between MR imaging inputs and cellular density.
• Random forest machine learning provides the most accurate estimates for glioma cellular density.
• Spatially specific cellular density estimations hold potential for guiding glioma diagnosis and treatment.