LLM-guided multimodal attention network for robust multiclass Parkinson’s disease diagnosis

Key Points

• The aim is to enhance Parkinson's disease staging and identification using a multimodal approach.
• Utilized LLM-guided multimodal fusion techniques
• Integrated semantic clinical knowledge with neuroimaging data
• Employed MGCA for analysis
• Improved accuracy in identifying prodromal Parkinson's disease
• Provided reliable and interpretable staging of Parkinson's disease