Previous studies have shown that late-life depression (LLD) may be a precursor of neurodegenerative diseases and may increase the risk of dementia. At present, the pathological relationship between LLD and dementia, in particularly Alzheimer's disease (AD) is unclear. Structural MRI (sMRI) can provide objective biomarkers for the computer-aided diagnosis of LLD and AD, providing a promising solution to understand the clinical progression of brain disorders. But few studies have focused on sMRI-based predictive analysis of clinical progression from LLD to AD. In this paper, we develop a deep learning method to predict the clinical progression of LLD to AD up to 5 years after baseline time using T1-weighted structural MRIs. We also analyze several important factors that limit the diagnostic performance of learning-based methods, including data imbalance, small-sample-size, and multi-site data heterogeneity, by leveraging a relatively large-scale database to aid model training. Experimental results on 308 subjects with sMRIs acquired from 2 imaging sites and the publicly available ADNI database demonstrate the potential of deep learning in predicting the clinical progression of LLD to AD. To the best of our knowledge, this is among the first attempts to explore the complex pathophysiological relationship between LLD and AD based on structural MRI using a deep learning method.
Keywords: Alzheimer’s disease; Cognitive impairment; Late-life depression; Structural MRI.