Diagnosing Parkinson’s disease in Early Stages using Image Enhancement, ROI Extraction and Deep Learning Algorithms

Parkinson’s disease or PD is among the most common neurodegenerative disorders which, due to loss of neurons, leads to the deterioration of motor and non-motor functions like tremors, shaking, slow movement, impaired balance and depression. It is a slow developing disorder and there is no definitive treatment for it; but its symptoms can be reduced with the help of specific drugs. Various methods have been introduced in order to detect PD in its early stages. Magnetic resonance imaging (MRI) is an effective and efficient method since it is an inexpensive and non-invasive tool which produces high-resolution images. In this paper, early diagnosis of PD has been carried out using different CNN models. Image enhancement has been performed on the magnetic resonance (MR) images in the dataset. After extracting the region of interest (ROI), in this case the substantia nigra (SN) region, a custom CNN model and VGG-16 are implemented for automatically differentiating SN in image and no SN in the image. We then performed binary classification to accurately discriminate between PD and control subjects by implementing deep learning models ResNet-34, VGG-19, ResNet-50 and improved AlexNet. At the end, after comparing the accuracies and losses of different models, it can be concluded that the ResNet-50 model depicted the best performance for this purpose.