Identifying Mild Hepatic Encephalopathy Based on Multi-Layer Modular Algorithm and Machine Learning.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Hepatic encephalopathy (HE) is a severe liver disease complication with high mortality.
• Mild hepatic encephalopathy (MHE) progression to overt HE necessitates early and accurate identification.
• Current MHE biomarker research often overlooks dynamic brain network changes.

Purpose of the Study:

• To investigate dynamic brain network properties in MHE patients using a multi-layer modular algorithm.
• To develop a machine learning model for distinguishing MHE from no-HE (cirrhosis without HE) patients.
• To identify novel biomarkers for MHE detection based on brain network evolution.

Main Methods:

• Resting-state functional MRI data were collected from healthy subjects, noHE patients, and MHE patients.
• A multi-layer modular algorithm was applied to analyze dynamic functional connectivity graphs.
• Nodal disjointness, a measure of network affiliation changes, was calculated within higher cognitive networks.

Main Results:

• Significant differences in nodal disjointness were observed between MHE and noHE groups within key cognitive networks.
• These abnormalities correlated with cognitive deficits in attention and visual memory, as measured by the Digit Symbol Test.
• A support vector machine model utilizing nodal disjointness achieved 88.71% accuracy in differentiating MHE from noHE.

Conclusions:

• Abnormal nodal disjointness in dynamic brain networks serves as a potential biomarker for MHE identification.
• This finding offers a fine-scale understanding of MHE's underlying disease mechanisms.
• The developed machine learning approach shows promise for clinical application in MHE diagnosis.