Predicting conversion to Alzheimer’s disease from Mild Cognitive Impairment using longitudinal MRI data and a Convolutional Neural Network

Mild cognitive impairment (MCI) is considered a substantial risk factor for Alzheimer’s disease (AD), as it is an intermediate condition between normal ageing and dementia. Even though some never develop AD, up to 50% of MCI subjects do progress to AD over three years. The purpose of this study was to predict conversion to AD within six months using a convolutional neural network (CNN). The model was trained with differential images obtained from 1.5T T1-weighted ADNI database magnetic resonance images (MRI) of 100 converters (MCIc) and 68 nonconverters (MCInc) obtained at five six-months intervals over a period of two years. The CNN conversion model obtained a predictive classification accuracy of 79%, 82% sensitivity, 67% specificity, 90% precision, and a F1-score of 86%. These results indicate that the methodology presented in this work using a CNN model provides high classification results, which are highly competitive with the state-of-the-art techniques using structural MRI as input. The presented work is thus a single modality approach which does not require additional clinical or neuropsychological data.