Classification of Facial Images to Assist in the Diagnosis of Autism Spectrum Disorder

Abstract

Autism Spectrum Disorder (ASD) is a developmental disorder that persistently impairs communication and social interaction, with a high degree of clinical heterogeneity. The fact that the diagnosis is based on clinical observation which requires therapy expertise, interviews with parents and application of questionnaires, it contributes to delaying the diagnosis until school age. The study of anthropometric measures in individuals with ASD and individuals in typical development (TD) showed there are some differences between the two groups. However, there is a lack of computational tools to assist in the acquisition and analysis process of these measures. This paper proposes a computer-aided medical decision support system that, given a child’s facial image captured by a digital camera, can discriminate between the two groups, ASD and TD, thus helping in the diagnosis. We defined a protocol for image acquisition and preprocessing, tested and compared methods for dataset balancing, dimension reduction and classification. The best results were obtained by a SVM classifier with 86.2% accuracy. Once the proposed model is based on facial images, it has the potential to facilitate early diagnosis of ASD.

High-level classification using complex networks for Autism Spectrum Disorder detection

“…Outra ferramenta para auxiliar o diagnóstico do TEA é proposta em [Michelassi et al 2023] a qual utiliza detecc ¸ão facial e classificac ¸ão de imagens. Sabendo que características faciais podem estar relacionadas ao desenvolvimento cerebral, a utilizac ¸ão de classificadores capazes de processar uma face e inferir o diagnóstico para TEA é uma ideia promissora.…”

“…Atualmente o diagnóstico é realizado por especialistas mediante observac ¸ão comportamental [Ministério da Saúde do Brasil 2021]. Existem, porém, diversos estudos que buscam maneiras mais efetivas de detectar o TEA, inclusive utilizando análises químicas e/ou biológicas [Michelassi et al 2023]. Esse trabalho se propõe a analisar um desses métodos o qual considerara a análise de dados sequenciais obtidos via espectroscopia no infravermelho por transformada de Fourier com reflexão total atenuada (ATR-FTIR) de amostras de saliva [Silva 2020].…”