Detection of Aspartylglucosaminuria Patients from Magnetic Resonance Images by a Machine-Learning-Based Approach

Area of Science:

• Neuroimaging
• Radiology
• Biomedical Engineering

Background:

• Neurodegenerative diseases, including lysosomal storage disorders like aspartylglucosaminuria (AGU), present diagnostic challenges.
• Magnetic resonance (MR) imaging offers rich data for developing computer-assisted diagnostic tools.
• Identifying reliable imaging biomarkers is crucial for early and accurate diagnosis.

Purpose of the Study:

• To compare the efficacy of different MR imaging features in classifying AGU patients from healthy controls.
• To evaluate volumetric, textural, and intensity ratio features for diagnostic potential.
• To determine the optimal MR imaging-based features for differentiating AGU.

Main Methods:

• Retrospective analysis of MR imaging data from 22 AGU patients and 24 healthy controls.
• Feature extraction included volumetric data from T1-weighted images, gray level size zone matrix (GLSZM) variance from susceptibility-weighted images, and caudate-thalamus intensity ratio from T2-weighted images.
• Random forest classifiers were trained and evaluated using leave-one-out cross-validation and receiver operating characteristic (ROC) curve analysis.

Main Results:

• Thalamic volumes (left-right-averaged, normalized volumes of 25 nuclei) and thalamic zone variance from susceptibility-weighted images showed excellent and equal performance in binary classification.
• Texture-based features from susceptibility-weighted images and thalamic volumes demonstrated high accuracy in differentiating AGU patients.
• The developed models achieved a very low error rate in classifying AGU patients versus healthy controls.

Conclusions:

• Thalamic volumes and texture features derived from susceptibility-weighted MR images are highly effective biomarkers for AGU.
• Computer-assisted diagnostic tools utilizing these MR imaging features can accurately differentiate AGU patients from healthy individuals.
• This study highlights the potential of advanced MR imaging analysis in diagnosing neurodegenerative lysosomal storage disorders.