Predicting Progression From Mild Cognitive Impairment to Alzheimer's Disease Using Autoregressive Modelling of Longitudinal and Multimodal Biomarkers

• Published in: IEEE journal of biomedical and health informatics 2018-05, Vol.22 (3), p.818-825
• Main Authors: Minhas, Sidra, Khanum, Aasia, Riaz, Farhan, Khan, Shoab A., Alvi, Atif
• Format: Article
• Language: English
• Subjects: Accuracy; Aged; Aged, 80 and over; Alzheimer Disease - diagnosis; Alzheimer Disease - diagnostic imaging; Alzheimer Disease - physiopathology; Alzheimer's disease; Area Under Curve; Autoregressive; Autoregressive models; Biological system modeling; Biomarkers; classification; Cognitive ability; Cognitive Dysfunction - diagnosis; Cognitive Dysfunction - diagnostic imaging; Cognitive Dysfunction - physiopathology; Conversion; correlation; Diagnosis, Computer-Assisted - methods; Disease Progression; Female; Humans; Impairment; linear; longitudinal; Magnetic Resonance Imaging; Male; Medical imaging; Medical treatment; Mild cognitive impairment; Modelling; Neurodegenerative diseases; Neuroimaging; Neurology; Neuropsychological Tests; Patients; prediction; Predictions; Predictive models; Support Vector Machine; Support vector machines
• ISSN: 2168-2194; 2168-2208
• DOI: 10.1109/JBHI.2017.2703918

Mild cognitive impairment is a preclinical stage of Alzheimer's disease (AD). For effective treatment of AD, it is important to identify mild cognitive impairment (MCI) patients who are at a high risk of developing AD over the course of time. In this study, autoregressive modelling of multiple heterogeneous predictors of Alzheimer's disease is performed to capture their evolution over time. The models are trained using three different arrangements of longitudinal data. These models are then used to estimate future biomarker readings of individual test subjects. Finally, standard support vector machine classifier is employed for detecting MCI patients at risk of developing AD over the coming years. The proposed models are thoroughly evaluated for their predictive capability using both cognitive scores and MRI-derived measures. In a stratified five-fold cross validation setup, our proposed methodology delivered highest AUC of 88.93% (Accuracy = 84.29%) and 88.13% (Accuracy = 83.26%) for 1 year and 2 year ahead AD conversion prediction, respectively, on the most widely used Alzheimer's disease neuroimaging initiative data. The notable conclusions of this study are: 1) Clinical changes in MRI-derived measures can be better forecasted than cognitive scores, 2) Multiple predictor models deliver better conversion prediction than single biomarker models, 3) Cognitive score boosted by MRI-derived measures delivers better short-term ahead conversion prediction, and 4) Neuropsychological scores alone can deliver good accuracy for long-term conversion prediction.