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

Abstract

Parkinson's disease (PD) is a progressive neurodegenerative disorder characterized by a wide spectrum of motor and non-motor symptoms, often leading to delayed or inaccurate diagnosis. Conventional diagnostic methods frequently suffer from limited sensitivity, scalability, and interpretability, thereby restricting their utility in clinical settings. To address these limitations, this study presents a novel AI-driven diagnostic framework that integrates multimodal data fusion, deep learning-based classification, and generative language modeling to improve diagnostic accuracy and enable personalized reporting. The proposed framework leverages the Parkinson's Progression Marker Initiative (PPMI) dataset, incorporating structural Magnetic resonance imaging (MRI), Single-Photon Emission Computed Tomography (SPECT) imaging, cerebrospinal fluid (CSF) biomarkers, and clinical assessments. Statistical analysis was employed to select 14 key biomarkers-including dopamine transporter SBR values and CSF protein levels-from a total of 21 features identified as clinically relevant. A 1D Convolutional Neural Network (1D-CNN) was developed and trained using 121 engineered features, comprising radiomic descriptors and biologically derived metrics. Preprocessing and extensive feature engineering were conducted prior to a 70:30 train-test split, with data augmentation applied to the training set to enhance model generalization. The classifier achieved an accuracy of 93.7%, surpassing baseline approaches and emphasizing the value of domain-informed feature design. To improve interpretability and clinician usability, a Mini ChatGPT-4.0 Large Language Model (LLM) was fine-tuned using approximately 1,000 domain-specific prompt-response pairs generated from literature, classifier-derived eXplainable AI (XAI) feature scores, and expert annotations. The generated responses were evaluated using a custom scoring metric (0.0-5.0) based on their semantic alignment with ground truth completions. This LLM module produces patient-specific diagnostic summaries and treatment suggestions. Additionally, a cloud-based interface was developed to facilitate real-time MRI uploads, automated inference, and chatbot-driven consultations. Overall, the framework demonstrates high diagnostic performance, transparency, and user accessibility, offering significant potential for real-world clinical deployment in PD diagnosis and decision support.

Keywords: Clinical decision support systems; Cloud-Based Diagnostic Platform; Fine-tuned LLM; Multimodal data analysis; PD diagnosis; Personalized medical report.

Fig. 1

Schematic representation of the 1D-CNN architecture for multimodal classification of PD stages.

Fig. 2

Integration framework of fine-tuned LLM for PD diagnosis and clinical decision support.

Fig. 3

MRI Preprocessing and Subcortical Structure Segmentation Pipeline for PD Analysis.

Fig. 4

Overview workflow of the 1D-CNN classifier model for PD classification using multimodal data.

Fig. 5

Interactive Interface and Case-Based Outputs from the Fine-Tuned PD Language Model.

Fig. 6

Text-Based User Interactions and Corresponding Responses from the Fine-Tuned PD LLM.

Fig. 7

AI-Generated PD Report Using the Fine-Tuned LLM in a Standardized Clinical Template.

Fig. 8

Graphical User Interface (GUI) of the Proposed Cloud-Based AI Platform for PD Management. The platform integrates multiple modules to streamline clinical and diagnostic workflows: (a) protein level prediction using patient-specific clinical inputs, (b) neuroimaging analysis through NiFTI file uploads for automated radiomic evaluation, (c) an AI-powered chatbot trained for PD-specific interactions and guidance, and (d) a centralized navigation panel enabling access to all analytical and support tools within the system.

Fig. 9

MRI processing and radiomics-based prediction workflow in the cloud-based PD platform. The interface demonstrates the end-to-end processing pipeline beginning with (i) NiFTI file upload, followed by (ii) image visualization, (iii) segmentation, (iv) brain mask generation, (v) image correction, and (vi) MRI registration. Subsequently, (vii) radiomic features are extracted and exported as a CSV file, which is then used by the integrated machine learning model to predict the presence of PD based on learned feature patterns.