Hippocampus Parcellation via Discriminative Embedded Clustering of fMRI Functional Connectivity

Area of Science:

• Neuroscience
• Brain Imaging
• Computational Neuroscience

Background:

• Understanding brain region function requires precise subregion parcellation.
• Traditional methods often separate dimensionality reduction and clustering, risking local optima.
• High-dimensional functional brain data presents challenges for accurate parcellation.

Purpose of the Study:

• To develop an integrated framework for brain parcellation combining subspace learning and clustering.
• To improve upon traditional stepwise parcellation methods by avoiding local optima.
• To investigate functional segregation within the hippocampus using a novel clustering approach.

Main Methods:

• Developed a novel parcellation framework based on discriminative embedded clustering (DEC).
• Combined subspace learning and clustering in a single, integrated procedure.
• Employed alternative minimization to optimize for a global optimum.

Main Results:

• Successfully parcellated the hippocampus into three spatially coherent subregions.
• Identified distinct functional connectivity patterns in these subregions in taxi drivers versus controls.
• Demonstrated superior parcellation consistency compared to traditional stepwise methods.

Conclusions:

• The proposed DEC-based framework offers a more robust approach to brain parcellation.
• Findings highlight the functional plasticity of hippocampal subregions related to navigation.
• The integrated approach advances the study of functional segregation and integration in brain networks.