期刊: [Institute of Electrical and Electronics Engineers] 日期:2026-07-01卷期号:: 1-5
标识
DOI:10.1109/icnpcv69434.2026.11708824
摘要
Autism Spectrum Disorder (ASD) is a prevalent neurodevelopmental disorder that has impairments in communication, as well as behavior and social relational mechanisms. Accurate early diagnosis is critical to allow access to early therapeutic measures and better quality of life. This work presents a novel unified diagnostic scheme that utilizes machine learning, deep learning and Internet of Things (IoT) based technology for multimodal ASD screening. Traditional classifiers such as Decision Tree, Random Forest, AdaBoost and K-Nearest Neighbor were applied to structured numerical data and showed well prediction performance, among which Random Forest was the best. On an image-based evaluation, deep convolutional neural networks (ResNet-18 and ResNet-50) got the best discriminative power with the highest accuracies of over 99%. Towards enhancing its applicability in real-life, IoT-enabled sensors such as pulse oximeter, gyroscope and camera module were integrated for capturing physiological, behavioral and speech-based features that are typically indicative of ASD. In addition, the system features a user-friendly interface for on-line processing and result interpretation. The experimental results thus confirm that the multimodal scheme will not only make the reliability of diagnosis more robust, but also provide a scalable, noninvasive tool for the support of clinical and caregiving usage. The results demonstrate how AI-driven, sensor assisted systems have the potential to transform digital health solutions to screen for ASD.