Spatio-temporal Tensor Multi-Task Learning for Predicting Alzheimer's Disease in a Longitudinal study.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Predicting Alzheimer's Disease (AD) progression is crucial for developing effective treatments.
• Machine learning (ML) offers potential for accurate AD progression modeling.

Purpose of the Study:

• To present a novel Multi-Task Learning (MTL) model for predicting AD progression.
• To enhance prediction accuracy and stability using spatio-temporal biomarker data and a novel regularization technique.

Main Methods:

• Developed an MTL model utilizing tensor formation from spatio-temporal similarity of brain biomarkers.
• Incorporated a novel regularization term to ensure longitudinal stability.
• Validated the model using magnetic resonance imaging (MRI) data and cognitive scores from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Main Results:

• The proposed MTL model achieved higher accuracy and stability in predicting AD progression compared to single-task and existing multi-task regression methods.
• Demonstrated significant reductions in root mean square error for MMSE and ADAS-Cog scores.
• Achieved an average RMSE reduction of 2.60 (MMSE) and 5.08 (ADAS-Cog) over single-task methods.

Conclusions:

• The novel MTL model effectively predicts Alzheimer's Disease progression.
• The method shows promise for aiding researchers and clinicians in AD management and treatment strategies.