Classifying disorders of consciousness using a novel dual-level and dual-modal graph learning model

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Disorders of Consciousness (DoC) present challenges in assessing awareness and communication.
• Current neuroimaging methods often use single modalities and overlook brain injury impacts.
• There is a need for advanced analytical techniques to improve DoC classification.

Purpose of the Study:

• To develop and validate a novel dual-modal neuroimaging analysis method for enhanced DoC classification.
• To integrate functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) data for improved diagnostic accuracy.
• To investigate the role of specific brain networks in consciousness and language processing within DoC patients.

Main Methods:

• Constructed a dual-model individual graph using fMRI and DTI data.
• Implemented a brain injury mask mechanism to consolidate damaged brain regions.
• Developed a dual-level graph to dynamically integrate individual and population-level data.
• Employed a subgraph exploration model with task-fMRI for interpretability validation.

Main Results:

• The proposed method achieved high accuracy in classifying patients into unresponsive wakefulness syndrome (UWS), minimally conscious state (MCS), and normal conscious state.
• Experimental results on 204 DoC patients and 89 healthy controls outperformed existing state-of-the-art methods.
• Identified key brain regions (e.g., default mode network, salience network) and their relevance to consciousness.
• Demonstrated that language-related subgraphs can distinguish MCS from UWS patients.

Conclusions:

• A novel graph learning method effectively classifies DoC using integrated fMRI and DTI data, incorporating a brain injury mask.
• The method's classification performance is superior to current approaches, offering improved diagnostic capabilities.
• Explainability analysis highlights crucial brain networks and their relation to language processing in consciousness, aiding diagnosis and prognosis.