Model Deteksi Parkinson Disease Berbasis Deep Learning Menggunakan Arsitektur VGG

Abstract

Background: Parkinsons Disease (PD) is a progressive neurodegenerative disorder characterized by the loss of dopaminergic neurons in the substantia nigra, resulting in motor and non-motor impairments. Early diagnosis remains challenging due to subtle initial symptoms and the relatively low accuracy of clinical assessment during early disease stages. Magnetic Resonance Imaging (MRI) provides high-resolution anatomical visualization and has the potential to detect early morphological changes. Advances in deep learning offer opportunities for automated PD detection through MRI analysis. This study aims to develop a PD detection model using the VGG architecture and evaluate its performance on MRI images.

Methods: This study employed a Research and Development (R&D) approach to construct a deep learning–based PD detection model. The dataset consisted of 2,000 brain MRI images (1,000 PD and 1,000 healthy controls) obtained from the open-source Kaggle platform. Preprocessing included image normalization and resizing to 256×256 pixels. The dataset was divided into 80% training data and 20% testing data. The model was developed using the VGG architecture and trained for 15 epochs with a batch size of 16. Model performance was evaluated using accuracy, precision, sensitivity, and specificity metrics.

Results: The VGG model demonstrated excellent classification performance on the test dataset. Evaluation results showed an accuracy of 0.99, precision of 0.99, sensitivity of 0.98, and specificity of 0.99. The confusion matrix indicated that the model correctly classified 198 healthy control images and 196 PD images, with minimal misclassification. Visualization of MRI comparisons showed that the model was able to detect morphological changes in the substantia nigra, including loss of the normal curvature of the crus cerebri, as an early indicator of PD.

Conclusions: The VGG-based PD detection model achieved very high performance in distinguishing PD from healthy controls using MRI images. These findings highlight the potential of deep learning as a tool for early PD detection. However, the use of Kaggle data as the primary dataset represents a limitation due to unverified acquisition standards and clinical quality. Therefore, further validation using multicenter clinical datasets is required to ensure the model’s generalizability to broader patient populations.