Global Advanced Research Journal of Medicine and Medical Sciences (GARJMMS) ISSN: 2315-5159
November 2014 Vol. 3(11), pp. 362-366
Copyright © 2014 Global Advanced Research Journals

Full Length Research Paper

Classification of handwriting patterns in patients with Parkinson´s disease, using a biometric sensor

Silke Anna Theresa Weber^1,2*, Carlos Alberto dos Santos Filho³, Arthur Oscar Schelp¹, Luiz Antonio Lima Resende¹, João Paulo Papa⁴, and Christian Hook⁵

¹Movement Disorder Group, Service of Neurology, Botucatu Medical School, State University São Paulo-UNESP, Brazil
²Department of Ophtalmology, Otolaryngology and Head and Neck Surgery, Botucatu Medical School - State University São Paulo, UNESP, Distrito de Rubião Júnior, S/N, 18618-970 - Botucatu - SP, Brazil
³Botucatu Medical School, State University São Paulo-UNESP, Brazil
⁴Computer Engeneering FCC-UNESP, Bauru-SP, Brazil
⁵Mathematics, Ostbayrische Technische Hochschule Regensburg, Germany

*Corresponding Author E-mail: silke@fmb.unesp.br; Tel: +55 14 3811-6256 or +5514 -3880-1524

Accepted 11 November, 2014

Abstract

Parkinson disease (PD) is characterized by typical movement disorders, important for clinical diagnosis and management. Objective assessment may be possible by mathematic classification of characteristics extracted by a sensor BiSP (Biosensor smart pen). The study aim to analyze handwriting characteristics of PD patients using a biosensor, and to classify the results by SVM-Support Vector Machines. 36 PD patients (group I) and 48 healthy adults (control group) with similar demographic characteristics were included. All realized drawing of patterned figures (spirals and meander) and tested diadochokinesia (pronation-supination test), using the BiSP pen. Biometric data were obtained from pen pressure, finger pressure on pen tip, acceleration of the movement, dislocation, tremor and instability. For each sensor were extracted characteristic features. Classification was tested using 70% of the data for learning and 30% for testing for each group, using the mathematic model of support vector machines. Accuracy of correct classification for each group and figure was described. For each figure, 8 to 12 features were extracted and submitted to SVM classification.  Correct classification of PD patients and controls showed an accuracy of 96.7% for spirals, 95.4% for meander, 92.5%  for diadochokinesia of the dominant hand and 93.6% diadochokinesia of the non-dominant hand. Combination of three figures, meander, spirales and diadochokinesia resulted in 99.6% of correct classification. The biometric features obtained by the BiSP permitted a correct classification of PD patients and control, using SMV as the mathematic tool. Biometrics and applied mathematics may help in PD characterization and follow- up.  

Keywords: Parkinson´s disease, biosensor, mathematic classification, SV