Feature selection from magnetic resonance imaging data in ALS: a systematic review

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Advances in neuroimaging suggest multiparametric MRI can aid early amyotrophic lateral sclerosis (ALS) diagnosis.
• Machine learning (ML) is crucial for integrating multiparametric MRI data into diagnostic classifiers.
• Limited data and feature selection challenges exist in developing ML classifiers for ALS.

Purpose of the Study:

• To conduct a systematic review of MRI biomarkers for ALS.
• To identify potential features for ML-based diagnostic classifiers.
• To synthesize findings regarding neuroimaging patterns in ALS.

Main Methods:

• Systematic literature search on PubMed for MRI studies in ALS within the last 4 years.
• Inclusion of studies with cross-sectional neuroimaging data and control groups.
• Qualitative and semi-quantitative synthesis of MRI modalities and affected brain regions.

Main Results:

• 151 studies and 221 datasets were identified.
• Findings align with known ALS neuropathology, including motor cortex and corticospinal tract degeneration.
• Degeneration was also observed in frontal, temporal, and subcortical structures, consistent with TDP-43 propagation models.

Conclusions:

• Insights support MRI feature selection for future ML-based ALS neuroimaging classifiers.
• Integrating multiparametric MRI (DTI, volumetric, texture) with ML offers a promising diagnostic approach for ALS.