Prediction of cognitive conversion within the Alzheimer's disease continuum using deep learning

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Early diagnosis and prognosis of cognitive decline in Alzheimer's disease (AD) are crucial for effective treatment.
• Predicting cognitive conversion aids in re-assigning patients to optimal, potentially more intensive, therapies.

Purpose of the Study:

• To develop a deep learning model for predicting cognitive conversion in Alzheimer's disease.
• To identify parsimonious prediction models with high accuracy for forecasting cognitive changes over time.

Main Methods:

• Longitudinal data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort were analyzed, including demographics, medical history, neuropsychological outcomes, laboratory, and neuroimaging results.
• A deep learning model was developed and refined by gradually removing variable sets to create parsimonious models for 1-4 year forecasts.
• Model performance was evaluated based on Area Under the Curve (AUC), with a threshold set for acceptable reduction in fit.

Main Results:

• The comprehensive deep learning model achieved excellent predictive performance (AUC 0.87-0.92).
• Parsimonious models, including only two variable sets, maintained good performance (AUC 0.80-0.84) and identified key predictors like neuropsychological outcomes, biomarkers, imaging data, and demographics.
• Predicted cognitive conversion led to a higher rate of recommended treatment upgrades compared to actual conversion, reducing the risk of missed interventions.

Conclusions:

• Neuropsychological tests combined with other indicators that change along the Alzheimer's disease continuum can significantly aid clinical treatment decisions.
• These predictive models can lead to improved disease management by facilitating timely and appropriate therapeutic interventions.