A Neural Network based Approach for the Analysis of Motion Disorders of Patients with Autism Spectrum Disorders

The autistic spectrum disorders (ASD) are behaviorally-defined developmental disorders of the immature brain which affect three domains of behavior: sociability and empathy; communication, language and imagination; and mental flexibility and range of interests. Main symptoms include motion disorders and stereotyped behaviors. This paper presents a neural-network based approach for the analysis of stereotyped motion disorders of patients with ASD. Specifically, monitoring is realized by means of tri-axis accelerometers applied to the patientʼs wrists. Signals obtained by accelerometers are pre-processed to obtain features that, in turn, are passed to the neural network that classifies the current observation in order to detect stereotyped motions. Results are under validation at the Department of Child Psychiatry at Childrenʼs Hospital Santobono-Pausilipon in Naples. I . I NT RODUC T I ON Autism is a group of developmental brain disorders, collectively called autism spectrum disorder (ASD). The term spectrum refers to the wide range of symptoms, skills, and levels of impairment, or disability, that children with ASD can have. Some children are mildly impaired by their symptoms, but others are severely disabled [1]. Thirty years ago autism was considered to be a rare childhood disorder most often associated with severe intellectual disabilities, lack of social awareness and the absence of meaningful expressive language [2]. Today, the spectrum of autistic disorders (or Autism Spectrum Disorder, ASD) is now recognized as a set of common developmental disorders, with an estimated prevalence of about 1 in every 110 children in the U.S. [3]. Symptoms of autism spectrum disorder (ASD) vary from one child to the next, but in general, they fall into three areas: 1) Social impairment; 2) Communication difficulties; and, 3) Repetitive and stereotyped behaviors [1]. This paper focuses on the automatic recognition of repetitive and stereotyped behaviors, with a specific emphasis for motion disorders like hand flapping or hand hitting ears, which are The proposed approach uses tri-axis accelerometers applied to patientʼs wrists to get motion information. In case of motion disorders signals caught from accelerometers reflects specific patterns. Thus, some features extracted from signals are used to classify the motion by a neural network. The rest of the paper is organized as it follows. Section II introduces some related work. Section III describes the proposed approach and reports some results. Finally, Section IV concludes the paper. I I . RE L AT E D WOR K Different Artificial Intelligence techniques are adopted to detect human activities in smart environments [4]. Among the others, Hidden Markov Models and Neural Networks [5] are major solutions in case of gesture recognition. To the best of our knowledge, there is no published paper on the automatic recognition of motion disorders of patients with ASD. To develope our system, we chose to adopt neural networks because they are more flexible models that can dynamically adapt to changes. Neural networks have been adopted for other applications of gesture recognition by means of signals acquired by ac-celerometers. Here we report some related work. Arce and Valdez [6] used the Nintendo Wiimote controller and built an artificial neural network to recognize five gestures: Circle, Square, Trinangle and the letters S and Z. A similar approach has been adopted by Kim et al. [7] to design a dance game controller. Starting from a basic posture, the user can move her hands Up, Down, Left and Right. Ahsan et al. [8] trained an artificial neural network with data extracted from electromiograpiphy signal in order to recognize hand motions for advanced humand computer interaction mechanisms. In this paper, we focus on stereotyped motion disorders such as Repeated hand Hitting against the ear (RH). This iis a typical motion disord for a patient with autism who tries to isolate himself from the sourrounding environment. Fig. 1. TI eZ430-Chronos Fig. 3. Example of motion disorder I I I . PROP OS E D AP P ROAC H