Early Detection of Parkinson's Disease Using Fusion of Discrete Wavelet Transformation and Histograms of Oriented Gradients

Parkinson's disease primarily affects people in their later years, and there is no cure for this disease; however, the proper medication of patients can lead to a healthy life. Appropriate care and treatment of Parkinson's disease can be improved if the disease is detected in its early phase. Thus, there is an urgent need to develop novel methods for early illness detection. With this aim for the early detection of Parkinson's disease, in this study, we utilized hand-drawn images by Parkinson's disease patients to effectively reduce the clinical experimental costs for poor people. Initially, discrete wavelet coefficients were extracted for each pattern of images; thereafter, on top of that, histograms of oriented gradient features were also extracted to refine the level of features. Thereafter, the fusion approach-based features were fed to various machine learning algorithms. The proposed work was validated on two different datasets, each of which consisted of various patterns, including spiral, wave, cube, and triangle images. The main contribution of this work is the fusion of two feature extraction techniques, which are histograms of oriented gradient features and discrete wavelet transform coefficients. The extracted features were then provided as input into different machine learning algorithms. In our experiment(s) on two datasets, the results achieved an accuracy of 79.7% and 97.8%, respectively, for all four discrete wavelet transform coefficients. This work demonstrates the utilities of fusion-based features for all four discrete wavelet transformation coefficients to detect Parkinson's disease, using image processing and machine learning techniques.