Web based AI-driven framework combining multi-modal data with CNN and LLM for Parkinson's disease diagnosis

Area of Science:

• Neuroscience and Artificial Intelligence
• Medical Imaging and Diagnostics
• Biomarker Discovery

Background:

• Parkinson's disease (PD) diagnosis is challenging due to its progressive nature and varied symptoms.
• Current diagnostic methods lack sensitivity, scalability, and interpretability.
• Delayed diagnosis hinders effective treatment and patient management.

Purpose of the Study:

• To develop a novel AI-driven framework for accurate and early Parkinson's disease diagnosis.
• To integrate multimodal data including MRI, SPECT, CSF biomarkers, and clinical assessments.
• To enhance diagnostic transparency and clinician usability through personalized reporting.

Main Methods:

• Utilized the Parkinson's Progression Marker Initiative (PPMI) dataset.
• Integrated structural MRI, SPECT, CSF biomarkers, and clinical data.
• Employed a 1D Convolutional Neural Network (1D-CNN) trained on 121 engineered features.
• Fine-tuned a Large Language Model (LLM) for personalized diagnostic summaries and treatment suggestions.
• Developed a cloud-based interface for real-time analysis and consultation.

Main Results:

• Selected 14 key biomarkers from 21 clinically relevant features.
• Achieved a diagnostic accuracy of 93.7% with the 1D-CNN classifier.
• The fine-tuned LLM generated semantically aligned patient-specific reports.
• The cloud interface enabled automated inference and chatbot consultations.

Conclusions:

• The AI framework demonstrates high diagnostic performance for Parkinson's disease.
• Multimodal data fusion and deep learning significantly improve diagnostic accuracy.
• LLM integration enhances interpretability and clinical utility for personalized patient care.
• The system offers potential for real-world clinical deployment in PD diagnosis and decision support.