Mild cognitive impairment classification using combined structural and diffusion imaging biomarkers.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Alzheimer's disease is a neurodegenerative disorder with a prodromal stage, mild cognitive impairment (MCI).
• Early diagnosis of MCI is critical for timely intervention and treatment optimization.

Purpose of the Study:

• To develop and validate a random forest (RF) classifier for distinguishing between MCI and healthy controls (HC).
• To identify key imaging and neuro-psychological features for accurate MCI classification.

Main Methods:

• Utilized structural MRI (sMRI) and diffusion-weighted imaging (DWI) features from hippocampus, thalamus, and amygdala.
• Employed random forest classifier with feature selection using extra trees classifier and mean decrease in impurity.
• Trained on Alzheimer's Disease Neuroimaging Initiative data and tested on a Mexican cohort, incorporating neuro-psychological scores.

Main Results:

• sMRI features were most influential in distinguishing MCI from HC.
• Combined sMRI and DWI significantly improved classification performance (AUROC ~93.5-93.79%, accuracy ~88.8-91.3%).
• Inclusion of neuro-psychological scores further enhanced the classifier's predictive power.

Conclusions:

• Multimodal analysis (sMRI+DWI+neuro-psychological scores) outperforms unimodal approaches for MCI diagnosis.
• The proposed RF classifier demonstrates high accuracy in discriminating MCI from HC.
• This approach shows promise as a valuable tool for early MCI detection and management.