Accurate low and high grade glioma classification using free water eliminated diffusion tensor metrics and ensemble machine learning

Sreejith Vidyadharan ¹, B V V S N Prabhakar Rao ¹, P Yogeeswari ²

⁰Department of Electrical and Electronics Engineering, Birla Institute of Technology and Science Pilani, Hyderabad Campus, Hyderabad, 500078, India.

Scientific Reports +

|

August 27, 2024

View abstract on PubMed

Summary

A new two-compartment Diffusion Tensor Imaging (DTI) model improves glioma diagnosis by accounting for free water contamination. This advanced DTI approach enhances classification accuracy for low-grade and high-grade glioma patients.

Area Of Science

Neuroimaging
Radiology
Oncology

Background

Glioma, a common brain tumor, requires early diagnosis and effective treatment.
Current biopsy-based diagnosis is invasive, necessitating non-invasive neuroimaging alternatives.
Diffusion Tensor Imaging (DTI) shows promise for assessing white matter (WM) changes in glioma, but single-shell DTI can be confounded by free water (FW) contamination.

Purpose Of The Study

To evaluate a two-compartment DTI model for its efficacy in accounting for FW contamination in glioma patients.
To identify DTI-based biomarkers for differentiating low-grade glioma (LGG) from high-grade glioma (HGG).

Main Methods

DTI data from 86 glioma patients (39 LGG, 47 HGG) were analyzed using a routine clinical protocol.
A two-compartment DTI model was compared against a standard single-compartment model.
Stacked-based ensemble learning was utilized to classify LGG and HGG patients based on DTI metrics from tumor and normal-appearing white matter (NAWM) regions.

Main Results

The two-compartment DTI model demonstrated superior performance over the single-compartment model in classifying LGG and HGG patients, evidenced by higher sensitivity, specificity, and AUC-ROC scores.
Four key DTI features, including fractional anisotropy (FA) in edema/core regions and FA/mean diffusivity in NAWM, achieved high classification accuracy (92% sensitivity, 90% specificity, 90% AUC-ROC).
Differential effects on both tumorous and NAWM regions were observed between LGG and HGG, highlighting the importance of considering these areas.

Conclusions

The two-compartment DTI model effectively addresses FW contamination, significantly improving diagnostic accuracy for glioma.
This enhanced DTI approach provides valuable biomarkers for distinguishing between LGG and HGG.
The findings support the clinical utility of the two-compartment DTI model for better glioma diagnosis and potential treatment planning.

Keywords:

Brain Diffusion tensor imaging Ensemble learning Free water elimination High-grade glioma Low-grade glioma MRI Machine learning Two-compartment DTI model White matter