Diagnosis of Autism in Children Based on their Gait Pattern and Movement Signs Using the Kinect Sensor.

Abstract

Background: Autism spectrum disorders are a type of developmental disorder that primarily disrupt social interactions and communications. Autism has no treatment, but early diagnosis of it is crucial to reduce these effects. The incidence of autism is represented in repetitive patterns of children's motion. When walking, these children tighten their muscles and cannot control and maintain their body position. Autism is not only a mental health disorder but also a movement disorder.

Method: This study aims to identify autistic children based on data recorded from their gait patterns using a Kinect sensor. The database used in this study comprises walking information, such as joint positions and angles between joints, of 50 autistic and 50 healthy children. Two groups of features were extracted from the Kinect data in this study. The first one was statistical features of joints' position and angles between joints. The second group was the features based on medical knowledge about autistic children's behaviors. Then, extracted features were evaluated through statistical tests, and optimal features were selected. Finally, these selected features were classified by naïve Bayes, support vector machine, k-nearest neighbors, and ensemble classifier.

Results: The highest classification accuracy for medical knowledge-based features was 87% with 86% sensitivity and 88% specificity using an ensemble classifier; for statistical features, 84% of accuracy was obtained with 86% sensitivity and 82% specificity using naïve Bayes.

Conclusion: The dimension of the resulted feature vector based on autistic children's medical knowledge was 16, with an accuracy of 87%, showing the superiority of these features compared to 42 statistical features.