Comparing Cyclicity Analysis With Pre-established Functional Connectivity Methods to Identify Individuals and Subject...

Area of Science:

• Neuroscience
• Medical Imaging
• Signal Processing

Background:

• Resting-state functional magnetic resonance imaging (fMRI) reveals cyclic patterns in brain activity, suggesting complex interactions.
• Traditional functional connectivity methods (e.g., correlation, dynamic time warping) struggle to capture temporal ordering and cyclicity in brain region interactions.
• Cyclicity analysis offers a novel approach to assess pairwise interactions by considering temporal dynamics.

Purpose of the Study:

• To compare the efficacy of cyclicity analysis against traditional methods for capturing individual and group-level information from resting-state fMRI.
• To investigate the influence of preprocessing steps, specifically filtering and global signal regression, on the performance of these analytical techniques.
• To evaluate the potential of these methods in differentiating tinnitus patients from neurotypical controls.

Main Methods:

• Analyzed resting-state fMRI data from tinnitus patients and controls across two sessions.
• Compared cyclicity analysis with zero-lag correlation, lagged correlation, and dynamic time warping distance.
• Assessed the impact of global signal regression and high-pass filtering (>0.1 Hz) on feature reliability.
• Utilized machine learning models (SVM, discriminant analysis, CNN) to evaluate group-level information representation.

Main Results:

• Cyclicity and correlation-based features demonstrated higher reliability in identifying individuals across visits compared to dynamic time warping.
• Global signal regression generally improved the reliability of most features, while high-pass filtering (>0.1 Hz) diminished it.
• Machine learning models could not sufficiently differentiate tinnitus patients from controls based on the analyzed features, indicating limited group-level information representation.

Conclusions:

• Cyclicity analysis shows promise for capturing individual-level functional brain interactions, but preprocessing choices are critical.
• Current functional connectivity features derived from resting-state fMRI may not adequately capture group-level differences for tinnitus diagnosis.
• Further research is needed to enhance feature representation for identifying neurological disorders like tinnitus using fMRI data.