Image-based Parkinson disease detection using deep transfer learning and optimization algorithm

Parkinson’s disease (PD) is typically a neurodegenerative disorder that slowly affects the brain, causes muscle stiffness, limb tremor and impaired balance that tends to get worsen over the time. The early detection of PD is necessary for proper treatment of patients and for providing them better health care services. Computer-aided diagnosis (CAD) system is a non-invasive and low-cost tool which have the potential to help in the diagnosis and monitoring of various diseases. Handwriting is important in the context of PD assessment. For the early detection of this disease, a number of machine learning techniques have been researched. Yet the main problem with the majority of these manual feature extraction methods is their poor performance and accuracy. To deal with this chronic condition, we need a Deep Learning (DL) model that can help in early diagnosis. In order to accomplish this, we propose a hybrid method that incorporates technique for data augmentation, feature extraction with pretrained Convolutional Neural Network (CNN), feature selection using optimization and classification with the help of Machine Learning to enhance PD identification. In this paper firstly, all types of handwriting images (circle, spiral and meander) are fed into six different pretrained models of CNN and are fine-tuned for classification among which VGG16 framework provides the better performance among the others. In the second stage, Binary grey wolf optimization (BGWO) is used for the selection of optimal subset of features extracted from VGG16 network by freezing the layers. The proposed method achieves classification accuracy of 99.8% using Support Vector Machine (SVM). The performance of our approach has been measured over the benchmark NewHandPD dataset. The experimental result shows that the proposed approach detects Parkinson's disease better than state-of-the-art methods by minimizing the feature subsets and thereby maximizing the accuracy. © The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management 2023. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.