Predicting time-to-conversion for dementia of Alzheimer's type using multi-modal deep survival analysis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Gerontology

Background:

• Predicting the trajectory of cognitive decline in individuals is crucial for early intervention in neurodegenerative diseases.
• Dementia of Alzheimer's Type (DAT) progression is complex and influenced by multiple factors, making individual prediction difficult.
• Utilizing multimodal data may improve the accuracy of predicting time-to-conversion to DAT.

Purpose of the Study:

• To develop a deep-learning model for predicting time-to-conversion to DAT using multimodal baseline data.
• To evaluate the predictive power of different data modalities (MRI, genetic, CDC) for DAT conversion.
• To identify the optimal combination of features for accurate DAT progression prediction.

Main Methods:

• A deep-learning survival analysis model was employed.
• Data from 401 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were analyzed.
• Features included MRI, genetic data, and CDC (Cognitive tests, Demographic, and CSF) measures.

Main Results:

• Cognitive, Demographic, and CSF (CDC) data demonstrated superior predictive power for time-to-conversion to DAT in Mild Cognitive Impairment (MCI) subjects compared to genetic or MRI data.
• Genetic data offered the highest predictive value for subjects with Normal Cognition (NC).
• Combining MRI and genetic features enhanced prediction accuracy over individual modalities, but adding CDC data did not further improve predictions beyond CDC alone.

Conclusions:

• The predictive utility of different data modalities for Dementia of Alzheimer's Type conversion varies depending on the subject's cognitive status (NC vs. MCI).
• CDC data are highly effective for predicting DAT conversion in MCI.
• Multimodal data integration, particularly MRI and genetic features, shows promise for improving DAT progression prediction.