Advancing Parkinson's detection from MRI: a deep learning comparison of classical and quantum architectures.
Authors
Affiliations (6)
Affiliations (6)
- Department of Biomedical Engineering, School of Engineering and Technology, Karunya Institute of Technology and Sciences, Karunya Nagar, Coimbatore, Tamil Nadu, 641114, India.
- School of Computer Science and Artificial Intelligence, SR University, Ananthasagar, Hasanparthy, Warangal, Telangana, 506371, India.
- Department of Information Systems, College of Computer and Information Sciences, Majmaah University, Al Majmaah, 11952, Saudi Arabia.
- Department of Applied Computer Sciences, Applied Computer Science College, King Saud University, Riyadh, 11451, Saudi Arabia. [email protected].
- Department of Biomedical Engineering, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chengalpattu, Tamil Nadu, 603203, India. [email protected].
- College of Engineering, Architecture and Fine Arts, Batangas State University, Batangas, Philippines. [email protected].
Abstract
The precise diagnosis of Parkinson's Disease (PD) continues to pose a considerable barrier in clinical neurology. This work investigates the capability of sophisticated deep learning architectures to improve Parkinson's disease identification via Magnetic Resonance Imaging (MRI) scans, aiming to improve diagnostic sensitivity and specificity. This research implements an automated PD classification framework utilizing deep learning models to analyze real-time MRI scans. CNN models and Transformers that leverages the inherent spatial information within MRI scans for automated feature extraction and classification of PD was implemented. The comparative analysis of diverse architectures aimed to discern the optimal feature extraction and classification strategies for distinguishing PD-affected brains from healthy controls was performed. K-fold cross-validation was conducted to evaluate the reliability of dataset partitioning and to ensure model generalizability. The QNN demonstrated superior performance during training and validation with accuracy of 0.9851 and 0.9731 respectively. Among the Transformer-based architectures, the Swin Transformer achieved a training and validation accuracy of 0.9712 & 0.9673. In the category of CNN-based models, ResNet50 exhibited the highest accuracy, attaining 0.9322 during training and 0.9192 during validation. The K-fold cross-validation results corroborated the stability and consistency of the model performances with the 80:20 data split, reinforcing the robustness of the proposed approach. Furthermore, to assess the generalizability of the best-performing model, an external validation was conducted using the Parkinson's Progression Markers Initiative (PPMI) dataset. The results demonstrated comparable performance trends, thereby confirming the adaptability of the DL model to independent datasets. This study's findings highlight the capability of deep learning-driven neuroimaging analysis in facilitating automated and precise PD diagnosis. This investigation provides a scalable and clinically viable AI framework that enhances diagnostic accuracy, thereby supporting clinicians in disease detection and improving patient management strategies.