Predicting Alzheimer’s disease progression in healthy and MCI subjects using multi‐modal deep learning approach

Background Alzheimer’s disease (AD) is a complex disorder influenced by many factors, but it unclear how each factor contributes to progression. An in-depth examination of these factors may yield an accurate estimate time-to-conversion AD for patients at various stages. Recent advances in deep learning have enabled researchers predict patient’s onset time exploring the influencing Method We used 543 subjects with 63 features from 3 data modalities Disease Neuroimaging Initiative (ADNI) database. The following were used: 1) MRI, 2) genetic and 3) DTC (Demographic, cognitive Tests Cerebrospinal fluid). 21 most important automatically selected three modalities. Deep Learning-based survival analysis model that extends classic Cox regression subjects' time. Here we re-define non-AD-progression as “survivor”, AD-progression “non-survivor”. divided into two groups: progressive (non-survivor), who either healthy or diagnosed Mild Cognitive Impairment (MCI) initial clinical visit later developed AD, non-progressive ("survivor”), MCI did not develop later. 10 random sub-samples, selecting 80% training 20% testing time; was internal validation. Result Figure 1 shows estimated rates over years. Both groups had high chance start. group’s dropped much faster fell below end period. remained around 50%. Feature importance displayed 2. Eight top ten are tests, demonstrating their analysis. Amygdala Hippocampus regions, well age, also notable features. Conclusion Our study demonstrated using powerful predictive models on multi-modal can improve prediction time-to-conversion. This only leads better understanding provides essential tools practitioners wish follow patients'