Autism spectrum Disorder detection Using Face Features based on Deep Neural network

Abstract

The majority of screening instruments for autism spectrum disorder (ASD) rely on subjective questions given to caregivers. Although behavioral observation is more objective, it is also more expensive, takes longer to complete, and requires a high level of competence. Therefore, there is still a dire need to create workable, scalable, and trustworthy systems that can identify ASD risk behaviors. Since there are no known causes of autism, early detection and intense therapy can significantly alter the behavior of children and people with the disorder. Artificial intelligence has made this possible, saving many lives in the process. Utilizing biological pictures, autism spectrum disorder (ASD) can be defined as a mental illness type which can be identified. The neurological condition known as ASD is linked to brain development and affects later appearance of the flask framework, a convolutional neural network (CNN) with transfer learning, and physical impression of the face. Xception, Visual Geometry Group Network (VGG16) the classification job was carried out using the previously trained models. 2,940 face photos made up the dataset utilized for the testing of those models, which was obtained via Kaggle platform. Outputs of the 3 models of deep learning have been evaluated with the use of common measures of assessment, including accuracy, sensitivity and specificity. With a 91% accuracy rate, Xception model had the greatest results. And theVGG16 models came next with (75%).