Presenter of 1 Presentation
A MACHINE LEARNING-BASED HOLISTIC AND AGE-DEPENDENT APPROACH FOR THE DIAGNOSIS WITHIN THE ALZHEIMER'S DISEASE SPECTRUM
Abstract
Aims
Alzheimer's disease (AD) is a neurodegenerative condition driven by a multifactorial etiology. We employed a machine learning (ML)-based algorithm and data of the Alzheimer's Disease Neuroimaging Initiative (ADNI) to investigate the relative contribution of clinically relevant factors in the prediction of AD conversion in subjects affected by Mild Cognitive Impairment (MCI), a transitional condition between healthy aging and dementia.
Methods
587 MCI subjects from ADNI-1, ADNI-GO, ADNI-2, and ADNI-3 databases were included. The inclusion criteria for these patients were the ones provided by the ADNI protocols, the completion of neuropsychological assessments at baseline, and at least 36 months of follow-up. Four classes of variables were considered: neuropsychological tests, AD-related biomarkers, peripheral biomarkers, and structural MRI variables. Only baseline data were analyzed. We implemented an ML-based Random Forest (RF) algorithm to classify, in a supervised manner, converting to AD (cMCI) or stable, non-converting (ncMCI), MCI subjects. Data related to the study population were analyzed by RF as separate features or combined and assessed in terms of classification power.
Results
Our ML-based algorithm showed good accuracy in predicting AD conversion of MCI subjects. Baseline neuropsychological tests combined with CSF biomarkers were the most accurate classifiers (ACC=0.86). The combination of peripheral biomarkers and psychometric variables also exhibited good accuracy (0.80). The classification accuracy was better in younger and female subjects.
Conclusions
Our results support the notion that AD is not an organ-specific “unitary” condition and results from pathological processes inside and outside the brain, which are dynamically affected by age and sex-related factors.