Exploring Survival Models Associated with MCI to AD Conversion: A Machine Learning Approach

• Published in: bioRxiv
• Main Authors: Orozco-Sanchez, Jorge, Trevino, Victor, Martinez-Ledesma, Emmanuel, Farber, Joshua, Tamez-Pena, Jose
• Format: Paper
• Language: English
• Published: Cold Spring Harbor Cold Spring Harbor Laboratory Press 11.11.2019; Cold Spring Harbor Laboratory
• Edition: 1.1
• Subjects: Alzheimer's disease; Artificial intelligence; Bioinformatics; Cognitive ability; Learning algorithms; Machine learning; Magnetic resonance imaging; Neurodegenerative diseases; Neuroimaging; Regression analysis; Survival; Survival; Machine Learning; Cox Model; Cross-validation; Radiomics
• ISSN: 2692-8205; 2692-8205
• DOI: 10.1101/836510

Several studies have documented that structural MRI findings are associated with the presence of early-stage Alzheimer Disease (AD). However, the association of each MRI feature with the rate of conversion from mild cognitive impairment (MCI) to AD in a multivariate setting has not been studied fully. The objective of this work is the comprehensive exploration of four different machine learning (ML) strategies to build MRI-based multivariate Cox regression models. These models evaluated the association of MRI features with the time of MCI to clinical AD conversion. We used 442 MCI subjects from the Alzheimer's disease Neuroimaging Initiative (ADNI) study. Each subject was described by 346 MRI features and time to AD conversion. Cox regression models then estimated the rate of conversion. Models were built using four ML methodologies in a cross-validation (CV) setting. All the ML methods returned successful Cox models with different CV performances. The best model exhibited a concordance index of 0.84 (95% CI: 0.82-0.86). The final analysis described the hazard ratios (HR) of the top ten MRI features associated with MCI to AD conversion. Our results suggest ML exploration is a viable strategy for building and analyzing survival models that predict subjects at risk of AD conversion.