REPRESENTATIVE FUNCTIONAL CONNECTIVITY LEARNING FOR MULTIPLE CLINICAL GROUPS IN ALZHEIMER'S DISEASE.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Mild cognitive impairment (MCI) is a precursor to Alzheimer's disease (AD), with 10-15% progressing annually.
• Differentiating stable MCI (sMCI) from progressive MCI (pMCI) is vital for understanding disease mechanisms and intervention.
• Functional connectivity (FC) alterations may indicate early-stage neurodegeneration relevant to MCI progression.

Purpose of the Study:

• To characterize group-level differences in functional connectivity (FC) between MCI subtypes (sMCI and pMCI).
• To identify representative FC patterns associated with different clinical groups, including normal cognition (NC), sMCI, pMCI, and AD.
• To develop a deep learning model for classifying these clinical groups based on FC features.

Main Methods:

• Integration of an autoencoder and multi-class classification into a single deep learning model.
• Training non-linear mappings for mutual transformations between the original FC space and a learned feature space.
• Construction of representative FCs for each clinical group by mapping learned feature vectors back to the FC space.

Main Results:

• Successfully learned clinical group-related feature vectors from functional connectivity data.
• Generated representative FC patterns specific to each clinical group (NC, SMC, sMCI, pMCI, AD).
• Achieved a multi-class classification accuracy of 68% for differentiating between NC, SMC, sMCI, pMCI, and AD.

Conclusions:

• The developed deep learning model effectively captures group-level FC differences relevant to MCI progression.
• Representative FC patterns can be constructed to visualize and understand neurobiological changes across clinical stages.
• This approach offers a promising avenue for early detection and differentiation of MCI subtypes, potentially aiding in Alzheimer's disease research.