A Fine-Tuned Transfer Learning Approach for Parkinson’s Disease Detection on New Hand PD Dataset

Parkinson’s Disease (PD) is a neurological condition that affects large masses of individuals around the world, having a great impact on the quality of life and raising notable challenges to the existing healthcare systems. Detection of PD at an early stage in an individual is crucial for the on-time intervention, diagnosis and improved patient outcomes. In the current scenario, various Machine Learning (ML) techniques, specifically, transfer learning-based approaches have proved to show promising results in analysis of medical images for diagnosis of diseases. A lot of work has been done in this area of utilizing transfer learning-based models such as VGG-19, ResNet etc. for the early detection of PD. Although most of these approaches have proved to have high performance values up to 95% accuracy, but there seems to have a large scope to improve upon and achieve higher performance measures that minimizes the risk of error in PD detection. This study uses the New Hand PD Dataset to propose an improved transfer learning model using fine-tuning specifically intended for the diagnosis of Parkinson's disease. The proposed novel approach in this paper uses feature extraction from the images and feeding them to fine-tuned transfer learning model, ResNet-152 leading to an improved testing accuracy of 100% and loss of 0.0040 only. To illustrate the suggested model's detection capability, its performance is compared with current cutting-edge deep learning and transfer learning models.