Human-Guided Functional Connectivity Network Estimation for Chronic Tinnitus Identification: A Modularity View.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Functional connectivity networks (FCNs) are crucial for diagnosing neuro-degenerative disorders.
• Accurate estimation of biologically meaningful FCNs is challenging due to brain complexity.
• Existing methods often lack integration of domain expert knowledge, limiting performance.

Purpose of the Study:

• To incorporate domain expert knowledge into FCN estimation using a modularity perspective.
• To develop a human-guided modular representation (MR) framework for FCN estimation.
• To improve the accuracy and biological interpretability of FCNs for disease diagnosis.

Main Methods:

• Proposed a human-guided modular representation (MR) framework for FCN estimation.
• Designed an adversarial low-rank constraint guided by domain expert knowledge (participant index).
• Evaluated the MR method on a chronic tinnitus (TIN) identification task.

Main Results:

• The MR method achieved a 92.11% accuracy in identifying chronic tinnitus.
• MR significantly outperformed baseline and state-of-the-art (SOTA) methods.
• Post-hoc analysis confirmed that MR-estimated FCNs highlight more biologically meaningful connections.

Conclusions:

• The proposed MR framework effectively integrates domain expert knowledge for improved FCN estimation.
• MR enhances the diagnostic accuracy for neuro-degenerative disorders like chronic tinnitus.
• MR facilitates exploration of disease mechanisms and early diagnosis by revealing biologically relevant brain connections.