Using Machine Learning for Motion Analysis to Early Detect Autism Spectrum Disorder: A Systematic Review

Abstract

Diagnosis of autism spectrum disorder (ASD) is typically performed using traditional tools based on behavioral observations. However, these diagnosis methods are time-consuming and can be misleading. Integrating machine learning algorithms with technological screening tools within the typical behavioral observations can possibly enhance the traditional assessment and diagnostic process. In the last two decades, to improve the accuracy and reliability of autism detection, many clinicians and researchers began to develop new screening methods by means of advanced technology like machine learning (ML). These methods include artificial neural networks (ANN), support vector machines (SVM), a priori algorithms, and decision trees (DT). Mostly, these methods have been applied to pre-existing datasets, derived from the standard diagnostic and assessment tools, to implement and test predictive models. On the other hand, the detection of new objective behavioral measures such as biomarkers could lead to a significant strengthening of existing screening tools. In the present study, we carried out a critical review of the literature about the latest findings in this field. The aim was to shed light about the effectiveness of using ML systems for motion analysis to enhance both clinical assessment and diagnostic processes. Specifically, we discussed the contribution of ML systems in promoting early diagnosis of ASD. The literature review showed that motion patterns ML analysis predicts ASD classification as accurately as that of classical gold standard tools. However, the application of these methods is still challenging, as discussed in this review.