Author ORCID Identifier
0000-0001-6689-6506
Biosketch
Dr. Priyadharshini S is an Assistant Professor, Project Associate, and Ph.D. Research Scholar specializing in Artificial Intelligence, Computer Vision, and Medical Imaging. She holds a Ph.D. in Artificial Intelligence from SASTRA Deemed to be University, where her research focused on deep learning–based medical image analysis, predictive modeling, and optimization techniques aimed at improving diagnostic accuracy and interpretability in healthcare systems. She completed her M.Tech in Artificial Intelligence and Robotics and B.Tech in Electrical and Electronics Engineering from the same institution.
She has over three years of experience in teaching, research, and academic mentorship, delivering courses and supervising undergraduate and postgraduate projects in Machine Learning, Deep Learning, Computer Vision, and Image Processing. Her research contributions span AI-driven healthcare applications, including MRI-based disease diagnosis using 3D-CNN and 3D-ResNet architectures, radiomics feature engineering, multimodal diagnostic frameworks, and Explainable AI (XAI) techniques such as SHAP and LIME.
Dr. Priyadharshini has published in reputed international journals and conferences, including Scientific Reports, Alexandria Engineering Journal, Elsevier book chapters, and IEEE conferences. She also has strong expertise in MLOps, RAG-based LLM applications, and cloud-based AI deployments, with hands-on experience in Python, PyTorch, TensorFlow, MLflow, DVC, and Streamlit. She is committed to advancing AI research and education through innovative teaching, impactful research, and student-centered mentorship.
Date of Award
17-7-2025
Document Type
Thesis
School
School of Electrical & Electroncis Engineering
Programme
Ph.D.-Doctoral of Philosophy
First Advisor
Dr.K.Ramkumar
Second Advisor
Dr.S.Venkatesh
Keywords
Parkinson’s disease, Artificial Intelligence, Radiomics & Deep Learning, Multimodal Data Integration, Large Language Models
Abstract
Parkinson’s Disease (PD) is a multifaceted and progressive neurodegenerative disorder that presents a spectrum of motor and non-motor symptoms. Early and accurate diagnosis is essential for effective disease management and improved patient outcomes, yet remains clinically challenging due to symptom overlap and diagnostic limitations. This thesis proposes a comprehensive and interpretable artificial intelligence (AI)-driven diagnostic framework that aims to transform the early detection, personalised monitoring, and treatment recommendation process for PD. The proposed solution integrates deep learning, radiomics, evolutionary optimisation, and large language models (LLMs), ensuring a highly accurate and clinically adaptable system.
The research begins by analysing T2-weighted 3D Magnetic Resonance Imaging (MRI) scans sourced from the Parkinson’s Progression Marker Initiative (PPMI) database. A robust preprocessing pipeline comprising brain extraction, registration, bias correction, normalization, and segmentation is applied. From the segmented subcortical brain regions, 107 radiomics features are extracted, of which the top 20 most predictive are selected using Pearson correlation, recursive feature elimination, and ranking techniques. Statistical validation is conducted using ANOVA, pairwise t-tests, and Kruskal-Wallis H-tests.
Multiple machine learning algorithms are evaluated, and the Gradient Boosting (GB) model, enhanced by the Synthetic Minority Oversampling Technique (SMOTE), attains an improved diagnostic accuracy of 96.8 % up from 86 %. To enhance transparency and clinical trust, Explainable AI (XAI) methods such as, SHAP and LIME are implemented, offering interpretable visual insights into model predictions. For advanced volumetric analysis, a custom 3D Convolutional Neural Network (3D-CNN) is designed and optimised through architectural refinement and hyperparameter tuning, achieving an accuracy of 93.4%. This model outperforms the baseline and complements an existing 3D-ResNet, which independently achieves 90% accuracy. Canonical Correlation Analysis (CCA) is then employed to fuse high-level features from both networks, yielding a combined accuracy of 95%. Further enhancement is achieved through the application of the Whale Optimisation Algorithm (WOA), a biologically inspired evolutionary technique, which boosts the final classification accuracy to 97%.
Recognising the multi-dimensional nature of PD, the thesis expands into multimodal data integration, encompassing MRI, SPECT scans, cerebrospinal fluid (CSF) protein biomarkers, and clinical scores. A 1D-CNN model is developed using 121 multimodal features and initially achieves an accuracy of 94.9%. With the inclusion of biologically derived ratio-based biomarkers, this accuracy increases to 96.9%. The integration of a fine-tuned ChatGPT-4.0 Mini model bridges AI-driven insights with clinical narratives, enabling personalised report generation, improved patient engagement, and real-time clinical decision support. A cloud-based platform is developed to enable scalable deployment with features like real-time inference, chatbot-assisted communication, and automated medical summaries.
Overall, this thesis presents a unified, explainable, and clinically deployable AI framework that significantly enhances the capabilities of PD diagnosis and personalized care. By integrating deep learning, radiomics, evolutionary optimization, and Large Language Models within a cloud-enabled platform, the proposed system establishes a novel benchmark for future clinical AI applications in the management of neurodegenerative disease.
Recommended Citation
S, Priyadharshini Ms, "A Predictive Framework for Early Detection and Personalised Monitoring of Parkinson’s Disease Using Artificial Intelligence and Large Language Models" (2025). Theses and Dissertations. 166.
https://knowledgeconnect.sastra.edu/theses/166