Exécution de notre outil de synthèse sur un nouvel article

Résumé : Artificial intelligence (AI) technology has recently shown promising results in detecting Autism Spectrum Disorder (ASD), but it faces significant challenges because it is still in early stages of development. This study proposes to use linguistic features in the detection of ASD using artificial intelligence. In order to systematically understand different AI model performances, two types of AI models have been explored and evaluated. Five classical ML models were used: logistic regression, Gaussian naive Bayes, random forests, k-nearest neighbors (K-NN), and support vector machines (SVMs), and two deep learning models were used: multilayer perceptron (MLP) and a convolutional neural network (CNN). All model development work is based on analyzing speech transcripts of 64 children total aged 3 to 6 from two data banks, CHILDES and ASDBank, including 30 children with ASD and 34 typically developing (TD) controls. The first step was to examine the annotations of the transcriptions, then extract various linguistic features from the texts. AI models were trained to determine whether a child has ASD based on the characteristics of the transcripts analyzed. There are three major findings of this study. First, neural network models will outperform classical models in ASD detection by around 5%-10%. The models multilayer perceptron and convolutional neural networks achieved 83 % and 84 % accuracy, respectively. Second, among the classical machine learning models, logistic regression, SVMs, K-NN, and random forests all achieved accuracies in the range 77%-80%, while Gaussian naive Bayes performed the worst by around 10%. Third, the linguistic feature with the highest importance in ASD detection is MLU, and the features MLU, MLT, rUtts, and rTNW show significant distinctions between age and gender demographics in addition to having high importance in ASD detection. Overall, this research recommends using deep learning for clinical applications of ASD detection.