Classifying irritable bowel syndrome using spatio-temporal graph convolution networks on brain functional MRI data.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Irritable bowel syndrome (IBS) is a functional gastrointestinal disorder with unclear pathophysiology.
• Current diagnostic and predictive models for IBS are limited by small sample sizes and inadequate analysis of functional MRI data.
• Functional MRI studies suggest alterations in brain networks are associated with IBS.

Purpose of the Study:

• To develop and validate a novel machine learning model for classifying IBS using rs-fMRI data.
• To identify key brain regions associated with IBS using an interpretable deep learning approach.
• To improve the diagnostic accuracy for IBS compared to existing methods.

Main Methods:

• Utilized rs-fMRI data from 79 IBS patients and 79 healthy controls.
• Applied a spatio-temporal graph convolution network (ST-GCN) for classification.
• Incorporated a novel interpretability module to identify important brain regions.

Main Results:

• The ST-GCN model achieved an average accuracy of 83.51%, outperforming other state-of-the-art methods.
• The interpretability module identified the Inferior Parietal Lobule, Inferior Frontal Orbital part, Postcentral Gyrus, Middle Frontal Orbital part, and Superior Medial Frontal Orbital part as key regions.
• External validation experiments confirmed the significant impact of these selected brain regions on IBS classification.

Conclusions:

• The developed ST-GCN model with an interpretability module shows high accuracy in classifying IBS from rs-fMRI data.
• Specific brain regions, including parts of the parietal and frontal lobes, are crucial for distinguishing IBS patients.
• This approach offers a promising avenue for developing more effective diagnostic tools for IBS.