Online Arabic and French handwriting of Parkinson's disease: The impact of segmentation techniques on the classification results. (April 2021)

• Record Type: Journal Article
• Authors: Ammour, Alae; Aouraghe, Ibtissame; Khaissidi, Ghizlane; Mrabti, Mostafa; Aboulem, Ghita; Belahsen, Faouzi
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: Parkinson's disease -- Arabic online handwriting -- Online segmentation techniques -- Feature selection -- Classification; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102429
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An original approach to classify PD patients and HCs is presented. Comparison between for segmentation strategies. Comparison between Arabic and French languages. Comparison between the segmented et non segmented texts. The classification results depend on the graphic characteristics of the used writing system. Abstract: Background and objectives: Handwriting (HW) is a task that requires fine motor control and specific neuromuscular coordination. An alteration of the HW faculties represents an early motor symptom of PD which can be exploited to develop an intelligent diagnostic system of this pathology. This article aims to identify the nature of the graphic elements in Arabic and French languages that could better reveal HW disorders related to PD, and therefore contribute to improving the separability between PD patients and healthy controls (HCs). Methods: This work includes the Arabic and French manuscripts of 56 bilingual participants of which 28 PD patients and 28 HCs. Four categories of segments were generated from each manuscript using online segmentation strategies with different cutoff criteria. Hence, kinematic, mechanic, and size features are calculated on each segment category as well as on the non-segmented texts, and were used to train the prediction models. Results: The segment category with a continuous sign in both the horizontal and vertical speed resulted in the best accuracy of 87.5%±8.3 in the case of Arabic text. Concerning the French text, the highest accuracy of 84.7%±6.6 was obtained for the segment category with a continuous sign in vertical speed. Conclusions: The contribution of HW features in the distinction of PD patients and HCs depends on how the segments are generated, and more generally on the graphic peculiarities of the writing system considered in this problem. The use of Arabic text provides more important information for the discrimination of PD patients and HCs.