Scientific Reports (2025) 15:38556
Web based Al-driven framework combining multi-modal data with CNN and LLM for Parkinson's disease diagnosis
By Priyadharshini S, Ramkumar K, Narasimhan K, V. B. Surya Prasath & Venkatesh S
Executive Impact: Revolutionizing PD Diagnosis
This groundbreaking research introduces an innovative AI-driven framework for Parkinson's disease diagnosis, achieving a remarkable 93.7% accuracy. By seamlessly integrating multimodal data – including MRI, SPECT, CSF biomarkers, and clinical assessments – with advanced 1D Convolutional Neural Networks (CNNs) and a fine-tuned Large Language Model (LLM), the system significantly enhances diagnostic precision and interpretability. The framework’s ability to generate personalized reports and provide real-time, explainable insights positions it as a critical tool for early detection and personalized patient management, promising transformative improvements in neurological care.
0
Diagnostic Accuracy
0
Prediction Precision
0
Disease Recall Rate
0
F1-Score
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Multimodal AI Diagnostic Architecture
The framework leverages comprehensive multimodal data inputs including structural MRI, SPECT imaging (DaTscan SBR values), CSF biomarkers (α-synuclein, Amyloid-β1-42, total Tau, Phosphorylated Tau), and detailed clinical assessments (UPDRS, MoCA). Statistical analysis was used to select 14 key biomarkers from 21 clinically relevant features. A 1D Convolutional Neural Network (1D-CNN) processes 121 engineered features, combining radiomic descriptors and biologically derived metrics. The classification decisions are then enhanced by a fine-tuned Mini ChatGPT-4.0 Large Language Model (LLM), which generates patient-specific diagnostic summaries and treatment suggestions, offering both high accuracy and interpretability.
Breakthrough Diagnostic Performance
Our AI-driven framework achieved an impressive 93.7% accuracy in distinguishing between PD, prodromal, and control groups. Key performance metrics include 97.2% precision, 94.4% recall, and a 96.5% F1-score, surpassing baseline approaches. The integration of domain-informed feature engineering, particularly five novel ratio-based features (e.g., P-tau181/Total-tau, Right caudate/Left caudate), proved crucial for enhancing the model's ability to identify subtle disease patterns. Explainable AI (XAI) techniques like SHAP and LIME were used to quantify feature importance, providing transparent insights into classification decisions.
Interpretable & Accessible Clinical Support
To improve interpretability and clinician usability, the fine-tuned Mini ChatGPT-4.0 LLM was trained on approximately 1,000 domain-specific prompt-response pairs, enriched with XAI feature scores and expert annotations. This allows for patient-specific diagnostic summaries and personalized treatment suggestions. Furthermore, a cloud-based interface facilitates real-time MRI uploads, automated inference, and interactive chatbot consultations, making the framework scalable, accessible, and suitable for diverse healthcare settings. This ensures transparent decision support and enhances user accessibility for both clinicians and patients.
93.7%
Accuracy in Multiclass PD Diagnosis, outperforming traditional methods.
Parkinson's Disease AI Diagnosis Workflow
Multimodal Data Input (MRI, SPECT, CSF, Clinical)
→
Preprocessing & Feature Selection
→
Feature Engineering (Ratio-based)
→
Data Augmentation
→
1D-CNN Classification
→
XAI Feature Score Generation
→
Fine-tuned LLM (ChatGPT-4.0 Mini)
→
Personalized Report & Suggestions
→
Cloud-Based Interactive Platform
| Feature | Our AI Framework | Traditional/Previous AI Methods |
|---|---|---|
| Data Modalities |
|
|
| Diagnostic Accuracy |
|
|
| Interpretability & Explainability |
|
|
| Personalized Reporting |
|
|
| Accessibility & Scalability |
|
|
| Early-Stage Detection |
|
|
Real-World Impact: AI-Driven Patient Consultations
Our framework transforms patient care through its interactive, AI-driven consultation system. As depicted in the research (Fig. 5 & Fig. 7), a clinician or patient can upload an MRI scan or input clinical queries. The fine-tuned LLM instantly analyzes the multimodal data, leveraging radiomic features, SBR values, CSF biomarkers, and clinical scores. It then generates patient-specific diagnostic narratives, identifies critical findings like reduced signal intensity in the substantia nigra, and provides personalized treatment suggestions. This seamless integration of deep learning and generative AI offers real-time decision support, enhancing diagnostic accuracy and fostering a more engaging, transparent patient-clinician dialogue.
Estimate Your Practice's AI Efficiency Gains
Discover how an integrated AI diagnostic platform can save your healthcare facility significant hours and costs annually.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Phased Deployment Roadmap for AI Integration
A clear path to integrating advanced AI for Parkinson's diagnosis into your clinical workflow.
Phase 1: Data Integration & Preprocessing
Establish secure data pipelines for MRI, SPECT, CSF, and clinical data. Implement robust preprocessing, feature engineering, and data augmentation protocols to ensure high-quality inputs for the AI model.
Phase 2: Model Training & Validation
Train and fine-tune the 1D-CNN classifier and the Mini ChatGPT-4.0 LLM using curated multimodal datasets. Conduct rigorous cross-validation and XAI analysis to ensure accuracy, interpretability, and robustness.
Phase 3: Cloud Platform & User Interface Development
Build the secure cloud-based platform for real-time data uploads, automated inference, and interactive chatbot consultations. Design a user-friendly interface for clinicians and patients, ensuring seamless accessibility.
Phase 4: Clinical Pilot & Feedback Loop
Deploy the framework in a pilot clinical setting to gather real-world feedback. Iteratively refine the models and platform based on clinician insights and patient outcomes to optimize performance and usability.
Phase 5: Scalable Deployment & Continuous Improvement
Roll out the AI diagnostic framework across broader clinical populations. Implement a continuous learning mechanism to update models with new data, ensuring long-term relevance and sustained diagnostic excellence.
Ready to Transform Parkinson's Diagnosis in Your Practice?
Connect with our AI specialists to explore how this advanced framework can be seamlessly integrated into your clinical operations, enhancing patient outcomes and operational efficiency.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat