Early Diagnosis of Autism in Children Using Convolutional Neural Networks (CNNs)

Abstract

The use of artificial intelligence (AI) alongside medical skills has witnessed significant growth in recent years, leading to impressive results in classification and processing, while facilitating the work of medical staff. Moreover, there has been an urgent need for software tools to assist in the early classification of diseases. In response to this need, this paper aims to develop a bidirectional neural network model with a web application interface for diagnosing Autism Spectrum Disorder (ASD) in children. The methodology employed relies on machine learning techniques for the early detection of this disorder, which includes collecting a comprehensive dataset containing 6,000 images, adjusting the image dimensions to focus on the facial area, and extracting morphological features associated with the disorder's symptoms. Edge detection techniques were applied, and images were segmented into recognizable parts. Subsequently, the model was trained using these features, and its performance was evaluated using metrics such as accuracy and F1 score. The proposed model achieved an accuracy ranging from 70% to 76% and an F1 score ranging from 75% to 79%, indicating its ability to classify cases of Autism Spectrum Disorder with high precision.The web application provides a user-friendly and accessible interface to utilize the diagnostic model, facilitating the screening and evaluation process. However, it is important to emphasize that the Convolutional Neural Network (CNN) model is a supportive tool for diagnosis and not a definitive solution. While it can assist in identifying potential cases of Autism Spectrum Disorder, the findings highlight the importance of expert involvement in diagnosis.  Recommendations include expanding the dataset, applying data augmentation techniques, and training specialists to use the model and interpret its results correctly