Click-To-Clinic: A Deep Learning–Based Web Application for Parkinson’s Disease Detection Using MRI Images

Parkinson's disease (PD) is an increasing neurodegenerative disease complicated to detect early due to basic symptoms of disease and the complexities of evaluating neuroimaging data. The recent deep learning algorithms showed automatic PD detection from MRI scans. In previous studies, there is limited offline testing and lack of portable, usable end-to-end screening solutions makes it difficult to prevent the disease early. This proposed work represents a lightweight web-based platform combining standardized MRI preprocessing, advanced convolutional neural network (CNN) and real-time prediction using a flask implementation process. The Gradient-weighted Class Activation Mapping (Grad-CAM) provides graphical representation of brain areas affecting the classification decisions with improved model visibility. The proposed approach is evaluated using publicly available MRI dataset. The results show better performance of 97% classification accuracy, equal precision, recall and F1-score leads to accurate detection of Parkinson's disease and normal control cases under controlled dataset conditions. However, class imbalance, dataset size and a lack of imaging-source variety have an impact on performance. As a result, this research presents a reusable research model and experimental system instead of a clinically proven diagnostic tool. Future study will focus on cross-dataset validation, improved data balancing techniques and detailed comparisons to existing machine learning and transfer-learning algorithms.