PPAD: A deep learning architecture to predict progression of Alzheimer's disease

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with millions affected globally.
• Mild cognitive impairment (MCI) is a precursor stage to AD, but not all individuals with MCI convert to AD.
• Current AD diagnosis occurs late, after significant dementia symptoms manifest, posing a burden due to the disease's irreversible nature.

Approach:

• This study introduces two novel Recurrent Neural Network (RNN)-based deep learning architectures: Predicting Progression of Alzheimer's Disease (PPAD) and PPAD-Autoencoder (PPAD-AE).
• These models are designed for early prediction of MCI to AD conversion, targeting both the next clinical visit and multiple visits ahead.
• To address the challenge of irregular time intervals in Electronic Health Records (EHR), the models incorporate patient age at each visit as a temporal indicator.

Key Points:

• Experimental validation on the Alzheimer's Disease Neuroimaging Initiative (ADNI) and National Alzheimer's Coordinating Center (NACC) datasets demonstrated superior performance of PPAD and PPAD-AE.
• The proposed models outperformed baseline methods in most prediction scenarios, particularly in terms of F2 score and sensitivity.
• Analysis revealed that patient age was a significant feature, effectively mitigating the issue of irregular time intervals in EHR data.

Conclusions:

• The developed PPAD and PPAD-AE models offer a promising approach for the early prediction of Alzheimer's disease conversion from MCI.
• Utilizing age as a temporal feature in deep learning models is effective for handling irregular time intervals in EHR data.
• These advancements hold potential for earlier intervention and improved patient management in Alzheimer's disease.