Machine learning with multitype functional connectivity uncovers whole-brain network disruption in primary...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Primary angle-closure glaucoma (PACG) is an irreversible blinding condition causing retinal ganglion cell damage and optic nerve atrophy.
• PACG significantly impacts brain functional networks, necessitating advanced diagnostic tools.

Purpose of the Study:

• To investigate brain functional network alterations in PACG patients using resting-state functional magnetic resonance imaging (rs-fMRI).
• To evaluate the efficacy of machine learning models in detecting PACG-specific neuroimaging biomarkers.

Main Methods:

• rs-fMRI data from 34 PACG patients and 34 healthy controls (HCs) were analyzed.
• Four connectivity features (static FC, dynamic FC, EC, dEC) were extracted using the AAL90 atlas.
• Elastic net feature selection and ten machine learning models were employed, with logistic regression (LR) showing optimal performance.

Main Results:

• The FC-based LR model achieved high diagnostic accuracy (0.92) and AUC (0.96).
• SHapley Additive exPlanations (SHAP) identified 20 critical connections with abnormal patterns in visual, attention, and sensorimotor networks.
• Reduced VSN-SMN and VSN-DAN connectivity, alongside enhanced DAN-thalamus connectivity, were observed in PACG patients.

Conclusions:

• Machine learning effectively detects PACG-related brain network disruptions.
• Specific connectivity alterations serve as potential neuroimaging biomarkers for PACG detection.