SOFIA: SOFTWARE GUIDED BY ARTIFICIAL INTELLIGENCE TO HELP IN THE PRE-DIAGNOSIS OF CHILDHOOD ASD

Abstract

The Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder whose signs can be identified from 18 months of age. However, early diagnosis is challenged by the lack of specialized professionals, especially in rural areas. This study aims to develop a mobile application, based on artificial intelligence (AI), to assist healthcare professionals in the pre-diagnosis of ASD in children aged 0 to 2 years in the Vale do Ribeira region. A MultiLayer Perceptron (MLP) neural network model was implemented to analyze screening data collected through the Q-CHAT-10 protocol. The application was developed with Kotlin for the interface, Java with Spring Boot for the API, and Python (FastAPI) for interaction with the AI model using Keras and TensorFlow. The database comprised 1,054 instances, and the model was trained using 10-fold cross-validation, achieving an accuracy of over 90%. Additional tests, with 54 new instances, resulted in 90.7% accuracy and 92.6% sensitivity. It is concluded that the developed application has the potential to improve ASD pre-diagnosis, especially in hard-to-reach regions, thus enhancing accessibility in the early identification process.