Autism Spectrum Disorder Detection using Autistic Image Dataset

Autism spectrum disorders (ASD) are neurodevelopmental disorders that are marked by enduring difficulties with speech, nonverbal communication, and restricted or repetitive behaviors. Early detection and intervention can greatly improve outcomes for people with ASD. Recently, deep learning algorithms have been applied to aid in the early detection of ASD using facial images. In this work, modifications of the commonly used VGG16 and VGG19 models for image recognition tasks are proposed to improve the performance of detecting ASD from a child’s frontal face image. The proposed model is unique, as it alters the architecture of existing models, adds an attentional mechanism, and applys transfer learning. These changes are intended to decrease the chance of overfitting and enhance the model’s capacity to capture subtle face characteristics. The performance of the updated model is assessed through accuracy, which is 82.55% for VGG19 and 80% for VGG16 model, and contrasted the outcomes of the original model. Performance of the modified model is also compared with that of the original model. The obtained results show that the modified model outperforms in detecting ASD from facial images, suggesting that the proposed modification is non-invasive for early detection of ASD and has the potential to contribute to the development of efficient tools.