Multimodal dynamic hierarchical clustering model for post-stroke cognitive impairment prediction.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Post-stroke cognitive impairment (PSCI) is a frequent and severe outcome of stroke.
• Early prediction of PSCI is crucial for personalized rehabilitation strategies.
• Current prediction methods face challenges in integrating multimodal data and capturing complex brain alterations.

Purpose of the Study:

• To develop an accurate and interpretable AI model for early prediction of PSCI.
• To leverage multimodal neuroimaging and clinical data for improved prediction.
• To introduce the multimodal dynamic hierarchical clustering network (MDHCNet).

Main Methods:

• Constructing brain graphs from diffusion-weighted imaging, magnetic resonance angiography, and T1/T2-weighted MRI.
• Integrating multimodal brain graphs with clinical features using a hierarchical cross-modal fusion module.
• Utilizing a graph neural network architecture (MDHCNet) for PSCI prediction.

Main Results:

• MDHCNet demonstrated superior performance compared to existing deep learning models in a real-world stroke cohort.
• Ablation studies confirmed the effectiveness of multimodal data fusion.
• Saliency-based interpretation identified key brain regions associated with cognitive impairment.

Conclusions:

• MDHCNet offers an effective and explainable approach for early PSCI prediction.
• The model has the potential to guide individualized clinical decision-making in stroke rehabilitation.
• This work advances the application of AI in understanding and managing neurological disorders.