Alzheimer's disease prediction algorithm based on hippocampal longitudinal hybrid morphological features.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder causing cognitive decline, linked to hippocampal structural changes seen in MRI.
• Current AD prediction models often overlook the complex spatiotemporal correlations within hippocampal morphology.
• There is a need for advanced frameworks to accurately model longitudinal changes in the hippocampus for better AD prediction.

Purpose of the Study:

• To develop a novel longitudinal prediction framework for Alzheimer's disease (AD) clinical progression.
• To effectively capture both temporal evolution and spatial distribution of hippocampal morphological alterations using deep learning.
• To improve the accuracy of predicting cognitive decline in AD by analyzing detailed hippocampal features.

Main Methods:

• A deep learning framework combining a multi-view feature fusion convolutional network (M-FCN) and a bidirectional gated recurrent unit (Bi-GRU).
• M-FCN utilizes 3D topological structure features, thickness, and heat kernel signature (HKS) to encode hippocampal atrophy.
• Bi-GRU module analyzes inter-sequence patterns and temporal correlations in longitudinal hippocampal features.

Main Results:

• The model demonstrated superior performance in capturing the relationship between AD-related structural changes and clinical indicators compared to existing methods.
• Evaluated on longitudinal T1-weighted MRI data from ADNI (n=221), the model achieved high predictive accuracy for Mini-Mental State Examination (MMSE) scores.
• Specific results include RMSE of 2.34 at M18 (CC=0.72), 2.58 at M24 (CC=0.77), and 2.60 at M36 (CC=0.83).

Conclusions:

• The proposed deep learning model effectively leverages spatiotemporal correlations of hippocampal morphology for accurate AD progression prediction.
• The framework provides reliable predictions, highlighting its potential for clinical application in managing Alzheimer's disease.
• This approach advances the understanding and prediction of neurodegenerative changes in AD using advanced neuroimaging analysis.