Diagnosis of Parkinson's disease using deep CNN with transfer learning and data augmentation

Parkinson's disease (PD) is one of the main types of neurological disorders affected by progressive brain degeneration. Early detection and prior care may help patients to improve their quality of life, although this neurodegenerative disease has no known cure. Magnetic Resonance (MR) Imaging is capable of detecting the structural changes in the brain due to dopamine deficiency in Parkinson's disease subjects. Deep learning algorithms provide cutting-edge results for various machine learning and computer vision tasks. We have proposed an approach to classify MR images of healthy and Parkinson's disease patients using deep convolution neural network. However, these algorithms require a large training dataset to perform well on a particular task. To this effect, we have applied a deep convolution neural network classifier that incorporates transfer learning and data augmentation techniques to improve the classification. To increase the size of training data, GAN-based data augmentation is used. A total of 504 images are collected, and 360 images are used to augment data. The increased data set of this model is as many as 4200 images, and the produced images are of good quality by using this data set for the detection of peak signal-to-noise ratio (PSNR) having an innovative value in the norm of real images. The pre-trained Alex-Net architecture helps in refining the diagnosis process. The MR images are trained and tested to provide accuracy measures through the transfer learned Alex-Net Model. The results are addressed to demonstrate that the fine-tuning of the final layers corresponds to an average classification accuracy of 89.23%. The experimental findings show that the proposed method offers an improved diagnosis of Parkinson's disease compared to state-of-the-art research.