Molecular Subtype Classification in Lower-Grade Glioma with Accelerated DTI.

Area of Science:

• Neuroimaging
• Oncology
• Machine Learning

Background:

• Lower-grade glioma molecular subtyping is crucial for prognosis.
• Diffusion tensor imaging (DTI) aids subtyping but requires long scans.
• This study explores faster DTI methods for glioma classification.

Purpose of the Study:

• To investigate lower-grade glioma classification using machine learning.
• To estimate fractional anisotropy (FA) from accelerated 3-direction DTI scans.
• To assess the added value of FA estimates in molecular subtyping.

Main Methods:

• 41 lower-grade glioma patients underwent preoperative DTI.
• Whole-tumor volumes were autodelineated.
• Fractional anisotropy (FA) and estimated FA were computed using DiffNet from 3-direction DTI subsets.
• Multivariate classification models used histogram and texture features.

Main Results:

• FA and estimated FA significantly differed between IDH-wild-type and IDH-mutant gliomas.
• Classification performance improved with FA and estimated FA inclusion.
• ADC-only models predicted IDH mutation and codeletion status with AUCs of 0.81 and 0.83, respectively.
• ADC + FA models achieved AUCs of 0.90/0.94, and ADC + estimated FA models achieved AUCs of 0.89/0.89.

Conclusions:

• Estimated FA from accelerated 3-direction DTI scans adds value to lower-grade glioma classification.
• This approach offers a faster alternative to conventional DTI for molecular subtyping.
• Machine learning-based FA estimation shows promise for improved glioma prognostication.