Handwriting and Speech Prototypes of Parkinson Patients: Belief Network Approach

Articulator phonetics and handwriting dysfunctions are frequent observations in Parkinson’s disease (PD). In this paper we make an inductive study of speech and handwriting skills of PD patients by proposing ways for discovering prototypes of PD patients. Each discovered prototype consists of labeled cluster that combines a similar handwriting and speech skills. For this approach, a mixed acquisition system of electronic pen and speech signals have been performed through voice and handwriting experiments on ten PD patients that share the same experimental conditions. The acquired signals were preprocessed and subjected to feature extractor. Our modeling approach is based on unsupervised learning of a probabilistic graphical model, i.e. a Bayesian Belief Network (BBN) based on Expectation Maximization (EM) algorithm. The structure components of BBN consist of layered architecture and hidden variables hierarchy. Each written and spoken test is represented by its own local hidden pattern; we considered that there exists a global hidden pattern dealing with each local pattern. The discovered patterns have been labeled and then conceptualized as a prototype to serve as a helpful assistant to a motor diagnostic tool based on articulator and handwriting diagnosis, more specifically for PD.