Differential Role for Hippocampal Subfields in Alzheimer's Disease Progression Revealed with Deep Learning

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Mild cognitive impairment (MCI) is a potential precursor to Alzheimer's disease, but its progression varies significantly.
• Current research on MCI progression often overlooks the detailed structure and subdivisions of the hippocampus.
• Understanding hippocampal subfield contributions is crucial for predicting MCI trajectory.

Purpose of the Study:

• To investigate the role of specific hippocampal subfields in differentiating between stable and progressive mild cognitive impairment (MCI).
• To develop and validate a deep learning model for predicting MCI progression based on hippocampal morphometry.

Main Methods:

• Utilized a dense convolutional neural network (CNN) architecture for analyzing hippocampal morphometry.
• Employed a novel occlusion analysis to determine the contribution of individual hippocampal subfields to model performance.
• Assessed the model's accuracy in differentiating stable from progressive MCI.

Main Results:

• The deep learning model achieved an accuracy of 75.85% in distinguishing between stable and progressive MCI.
• Occlusion analysis identified the presubiculum, CA1, subiculum, and molecular layer as critical subfields for prediction.
• Found that 10.5% of hippocampal volume was redundant for differentiating MCI progression.

Conclusions:

• Deep learning models can effectively utilize hippocampal morphometry to predict MCI progression.
• Specific hippocampal subfields play a disproportionately important role in the transition from stable MCI to progressive MCI.
• These findings highlight the potential of detailed hippocampal subfield analysis for early detection and understanding of Alzheimer's disease pathogenesis.