Enhancing Functional Brain Network Construction for Neurological Disorder Diagnosis through Modularity-Driven Regularization

The burgeoning field of Functional Brain Network (FBN) analysis has emerged as a promising avenue for diagnosing neurological disorders. As innovative biomarkers, FBNs offer a novel perspective on probing the intricacies of brain connectivity. However, the fidelity of FBN construction depends on the quality of data acquisition, which is often a formidable challenge due to inherent limitations. Inspired by the no free lunch theorem, a straightforward ambition is to integrate the exogenous knowledge to enhance the fidelity of FBNs. Motivated by this ambition, this paper provides a Modularity-Driven Regularization (MDR) for the FBN construction. Specifically, we design and incorporate a dyadic L<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">21</inf> regulariser into the FBN model to formulate the modularity prior. Furthermore, we have developed a generalized alternating multiplier method algorithm for the estimation of the FBN. To demonstrate the effectiveness of our MDR, we perform rigorous experiments on a Mild Cognitive Impairment (MCI) identification task using the NITRC dataset. The validity of our model has been confirmed by the experimental outcomes. Additionally, it enhances biomarker precision for diagnosing neurological disorders like MCI, supporting potential early intervention strategies.