Integrating Convolutional Neural Networks and Multi-Task Dictionary Learning for Cognitive Decline Prediction with...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Data Analysis

Background:

• Predicting Alzheimer's disease (AD) progression using neuroimaging biomarkers is crucial.
• Convolutional Neural Networks (CNNs) excel at feature extraction from images but face challenges with limited labeled neuroimaging data and integrating multiple data sources.

Purpose of the Study:

• To propose a novel multi-task learning framework based on CNN to address challenges in AD progression prediction.
• To improve prediction accuracy by jointly analyzing multiple data sources and handling limited labeled samples.

Main Methods:

• Utilized a pre-trained CNN on ImageNet for knowledge transfer to neuroimaging, serving as a deep feature extractor.
• Introduced Multi-task Stochastic Coordinate Coding (MSCC) for unsupervised learning of sparse features from multi-task feature maps.
• Applied Lasso regression on multi-task sparse features to predict Alzheimer's Disease Assessment Scale cognitive subscale (ADAS-Cog) and Mini-Mental State Examination (MMSE) scores.

Main Results:

• The novel CNN-MSCC system was applied to the Alzheimer's Disease Neuroimaging Initiative dataset.
• The proposed method demonstrated superior performance in predicting future MMSE and ADAS-Cog scores compared to seven other methods.

Conclusions:

• The developed CNN-MSCC framework offers new insights into data augmentation and multi-task deep learning for neuroimaging.
• This research facilitates the broader adoption of deep learning models in neuroimaging studies for conditions like Alzheimer's disease.