Computer Aided Diagnosis of Autism Spectrum Disorder Based on Thermal Imaging

Abstract

AbstractAutism Spectrum Disorder is a fast-growing area of study in the field of neurodevelopmental sciences. It is widely recognized that individuals with ASD have emotional processing impairments which ultimately leads to the inability to recognize facial expressions. The aim of the study was as follows: (1) To compare the facial skin temperature for different emotions namely happiness, sadness and anger using thermal imaging by doing a comparison between autistic and non-autistic children; (2) To develop a CAD tool which comprises of segmentation using K-means algorithm, GLCM feature extraction and classification using SVM. A number of 30 autistic and 30 non-autistic subjects were considered for the study. The thermography approach was used to acquire the temperature of the many facial regions such as eyes, cheek, forehead, and nose while the subjects were asked to react to the projected formats. The mean temperature difference between the autistic and non-autistic individuals in the nose region for the emotion happy, anger and sad is 2.77%, 12.7%, and 13% respectively. The accuracy obtained by classifying the thermal images of the autistic and non-autistic children using SVM classifier was found to be 88% respectively compared to random forest (73%) and Naïve bayes classifier (66%). The Dense-net 121provided better accuracy of 89.47% compared to the other machine learning classifiers. Thermography is a diagnostic approach used to acquire temperature differences with high resolution. The computer-aided diagnostic tool can be a reliable and an enduring method in the diagnosis of the autistic individuals. KeywordsAutism spectrum disorderThermal imagingMachine learning