Parkinson's Disease Diagnosis with Gait Characteristics Extracted Using Wavelet Transforms

• Published in: IEEE journal of translational engineering in health and medicine 2023-01, Vol.11, p.1-1
• Main Authors: Vimalajeewa, Dixon, McDonald, Ethan, Tung, Megan, Vidakovic, Brani
• Format: Article
• Language: English
• Subjects: Classification; Classifiers; Correlation; Diagnosis; Diagnostic systems; Discrete wavelet transforms; Entropy; Feature extraction; Force; Gait; Level-wise cross correlation; Parkinson's disease; Performance evaluation; Self-similarity; Support vector machines; Time domain analysis; Vertical forces; Wavelet domain; Wavelet Transform; Wavelet transforms
• ISSN: 2168-2372; 2168-2372
• DOI: 10.1109/JTEHM.2023.3272796

Objective: Parkinson's disease (PD) is a common neurodegenerative disorder among adult men and women. The analysis of abnormal gait patterns is among the most important techniques used in the early diagnosis of PD. The overall aim of this study is to identify PD patients using vertical ground reaction force (VGRF) data produced from subjects while walking at a normal pace. Methods and procedures: The current study proposes a novel set of features extracted on the basis of self-similar, correlation, and entropy properties that are characterized by multiscale features of VGRF data in the wavelet-domain. Five discriminatory features have been proposed. PD diagnosis performance of those features are investigated by using a publicly available VGRF dataset (93 controls and 73 cases) and standard classifiers. Logistic regression (LR), support vector machine (SVM) and k-nearest neighbor (KNN) are used for the performance evaluation. Results: The SVM classifier outperformed the LR and KNN classifiers with an average accuracy of 88.89%, sensitivity of 89%, and specificity of 88%. The integration of these five features from the wavelet domain of data, with three time domain features, stance time, swing time and maximum force strike at toe improved the PD diagnosis performance (approximately by 10%), which outperforms existing studies that are based on the same data set. Conclusion: with the previously published approaches, the proposed prediction methodology consisting of the multiscale features in combination with the time domain features shows better performance with fewer features, compared to the existing PD diagnostic techniques. Clinical impact: The findings suggest that the proposed diagnostic method involving multiscale (wavelet) features can improve the efficacy of PD diagnosis.