Feature selection framework for functional connectome fingerprinting.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Psychiatric Research

Background:

• Functional connectome (FC) fingerprinting aims to uniquely identify individuals using brain connectivity patterns.
• Existing methods face challenges with large sample sizes and coarse parcellations, impacting accuracy.
• Understanding these limitations is vital for advancing precision psychiatry.

Purpose of the Study:

• Investigate the combined effects of sample size and parcellation granularity on FC fingerprinting accuracy.
• Explain the underlying reasons for accuracy degradation using data mining clustering metrics.
• Develop and evaluate a systematic feature selection framework to enhance subject-specific information identification in resting-state functional connectivity (RSFC).

Main Methods:

• Examined the interplay between sample size and parcellation resolution in FC fingerprinting.
• Applied clustering quality metrics to elucidate accuracy reduction mechanisms.
• Developed and tested six feature selection approaches within a general framework for RSFC data.

Main Results:

• Confirmed that fingerprinting accuracy declines with increased sample size and reduced parcellation granularity.
• Identified specific RSFC elements that contribute most to unique subject identification.
• Evaluated the performance of different feature selection strategies in mitigating accuracy loss.

Conclusions:

• The developed feature selection framework offers a systematic way to identify subject-specific FC information.
• This approach holds promise for improving the robustness and accuracy of FC fingerprinting in large-scale neuroimaging studies.
• Findings contribute to the development of more effective tools for precision psychiatry.