Drawing reveals hallmarks of children with autism

Highlights

• A Paintings of Autism Spectrum Disorder (PASD) dataset is established.

• Subjective and objective analysis show significant hallmarks of painting of autistic children.

• The hallmarks have potential usages in large scale screening of ASD.

Abstract

Autism spectrum disorder (ASD) as a kind of mental disorder, has become an internationally recognized serious public health problem. Paintings of autistic children have not been compared systematically to those from Typically Developed (TD) children. In this work, we construct an ASD painting database which contains 478 paintings drawn by ASD individuals and 490 drawn by the TD group. Through subjective and objective analysis, some significant hallmarks, such as structuring logic, face, repetitive structure, composition location, edge completeness, etc. are found within the ASD paintings. We further train a classifier of ASD and TD painters using those extracted features, which shows encouraging accuracy as a potential screen tool for ASD. This work sheds light on understanding the uniqueness of autistic children through their paintings. The database will be released to the public.

Introduction

Autism spectrum disorder (ASD) is a neurodevelopmental disorder that may cause communication impairments, as well as restricted and repetitive behaviors [2]. As a kind of mental disorder, it has become an internationally recognized serious public health problem [6]. The prevalence of ASD has become an economic burden [7] and resulted in parental stress [17]and other social problems [18].

Craig j et al. [8] explore the cause of a specific impairment in 21 children with autism in terms of the ability to process non-veridical representations through four different painting tasks. Jolley et al. [19] evaluated 15 autistic children's emotional and social abilities, imagination, and intellect by asking them to draw happy and sad pictures. To explore whether the relationship between cognitive processes and imagination differs between neurotypically developing children and children with autism, Kala et al. [21] administered a cognitive task battery and Karmiloff-Smith's drawing task. Findings suggest that cognitive processes involved in imagination change in the early development stage. Children with autism employ a unique cognitive strategy in the imaginative drawing. Therefore, autistic paintings are like treasure trove that has not been fully exploited. They contain a lot of information that may reflect certain characteristics or defects of autism, which can contribute to future screening and rehabilitation of autistic patients.

Jill Mullin has assembled an array of works from both renowned artists like Gregory Blackstock and Jessica Park and those who are less well known but no less talented [16]. Their creations, coupled with artist interviews, comprise a fascinating and compelling book that serves to educate and inspire anyone who knows someone diagnosed with ASD. It advocates nurturing the talents, artistic and otherwise, of autistic individuals [16]. Using artwork created by individuals diagnosed with ASD, Drawing Autism celebrates their artistry and self-expression and serves as an accessible point of entry into understanding how ASD manifests in individuals. This book is not only an art collection as the contributors show their unique perspective on how they see the world and their place in it. But in fact, we can collect a wider range of paintings of ordinary individuals with autism and conduct systematic analysis at the image level, which may lead to new directions for assisting the diagnosis of autism in the future.

Recently, some studies have used the eye movement of autistic individuals to characterize their traits, which are associated with the diagnosis of ASD. Wang et al. [36]quantified atypical visual attention of ASD across multiple levels of image features. Duan et al. established a Saliency4ASD dataset [38] and fine-tuned four deep-learning-based saliency prediction methods to characterize the visual attention traits of individuals with ASD [11], [12]. Besides, they also examined the visual attention of individuals with ASD on human faces [10]. Liu et al. proposed a machine learning method to classify children with ASD and control groups based on their gaze patterns in a face recognition task. Jiang et al. [27] fine-tuned one saliency prediction model and used fixation information to classify individuals with ASD and their controls. However, these methods all need an eye tracker to collect eye movement data [37], [3], [5], which is hard to accessible to parents.

Longard compared the scores of autistic children and normal children on understanding emotional [25]) and found that autistic children may have potential advantages in paying attention to artistic expression and visual processing. Drawing is a non-verbal way of communication that may help children with autism speak up [32], [29], [13]. Having an outlet for self-expression may reduce the challenges autistic people face [29]. Besides, these findings suggest that ASD may be discovered in time by analyzing their works of artistic expression. However, as far as we know, existing studies have not paid attention to the difference in the characteristic of paintings between autistic individuals and healthy controls before, let alone using painting features to classify these two groups. Thus, in this paper, we explored the possibility of using the features of children’s paintings to identify ASD. Since it is easy for parents to get paintings from their children, this method may provide a new perspective for large-scale screening of ASD, which is easy to implement and less costly.

In this paper, we propose a semi-subjective method to classify individuals with ASD and their typically controls (TD) group using their paintings. To analyze the differences in painting features between autistic children and normal children, we first establish a Paintings of Autism Spectrum Disorder (PASD) dataset, including 478 paintings from individuals of ASD and 490 paintings from the TD group. All these paintings are collected from 15 ASD children and 20 TD children. Based on these paintings, we manually extract and analyze 7 types of features for each group, as shown in the Painting Analysis section. Then, we conduct an SVM classification experiment based on extracted features and the experimental results show that the proposed method has high accuracy for the classification of autistic children and normal children. This paper is the first to explore the potential of using the painting to screen individuals with autism. The advantages of this method are listed as follows. First of all, we believe that painting can reveal the communication awareness and behavior within ASD individuals, as well as show the characteristics of ASD. Second, children's paintings are easy to obtain by parents and do not require additional instruments, making objective large-scale screening and evaluation possible. Our PASD dataset including drawing and analysis results will be released to the public to facilitate future research.