Normalizing flow based neural processes for Alzheimer's disease progression prediction.

As one of the most common neurodegenerative diseases, Alzheimer's accounts for serious health problems worldwide. Accurate detection and prediction of this disease contribute to the health system for better prevention and interventions in the treatment plans. However, traditional models designed for prediction and classification face several challenges, including handling complex data, which neglects many data points for the diagnosis. To overcome this challenge, we propose a novel model based on the integration of Neural Processes (NPs) and Normalizing Flows (NFs). The dataset used for this study is the Alzheimer's Disease Prediction of Longitudinal Evolution (TADPOLE). We selected various features to build an efficient model, including cognitive, neuroimaging, genetic, and demographic data. which contains three classes: Cognitively Normal (CN), Mild Cognitive Impairment (MCI), and AD. The proposed model is able to capture the temporal dependencies present in the complex distribution. The stochastic processes were modeled by NPs, while NF was able to transform the Gaussian distributions from simple to complex distributions, allowing them to model a wide range of data distributions. The prediction performance and robustness have been enhanced since this framework is able the adapt to every patient trajectory and generalizing across different populations. The model was compared with other models, such as SNP, deep geometric learning, Manifold DCNN, and other models. Our model (SNP-NF) made an improvement regarding mAUC, Precision, and Recall, approximately 3%,1%, and 0.7%, respectively from our previous model, which utilized only NP. These results demonstrate the capability of the proposed approach to provide early detection and personal treatment plans for patients suffering from this disease.