Dynamic Routing Capsule Networks for Mild Cognitive Impairment Diagnosis

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder causing significant cognitive decline.
• Early diagnosis of mild cognitive impairment (MCI), a preclinical stage of AD, is crucial for intervention and slowing disease progression.
• Machine learning and deep learning offer potential for automated MCI diagnosis using functional connectivity data.

Purpose of the Study:

• To propose and evaluate novel dynamic routing capsule networks for the automated diagnosis of MCI.
• To investigate two variants of capsule networks utilizing intra-ROI and inter-ROI dynamic routing for functional representation.
• To introduce a learnable dynamic functional connectivity metric within the inter-ROI model.

Main Methods:

• Development of two distinct dynamic routing capsule network architectures for MCI diagnosis.
• Application of intra-ROI dynamic routing to capture regional functional representations.
• Implementation of inter-ROI dynamic routing with a learnable dynamic functional connectivity metric.
• Comparative analysis against traditional machine learning and other deep learning models.

Main Results:

• The proposed dynamic routing capsule networks achieved superior performance in MCI diagnosis across multiple evaluation metrics.
• The novel approach demonstrated effectiveness in extracting meaningful functional representations for disease detection.
• The learnable dynamic functional connectivity metric contributed to enhanced diagnostic accuracy.

Conclusions:

• Dynamic routing capsule networks represent a promising advancement in deep learning for MCI diagnosis.
• This novel methodology offers a powerful tool for early detection and potential intervention in Alzheimer's disease.
• The study highlights the potential of capsule networks and dynamic functional connectivity metrics in neurodegenerative disease research.