Deep Learning for Early Autism Screening: Analysis of Hand-Flapping Motions in Infants

Autism Spectrum Disorder (ASD) is a highly complex neurological developmental disorder characterized by difficulties in social interactions, significant problems with communication, and repetitive motor behaviors, such as hand-flapping. These distinctive hand-flapping motions are often among the earliest detectable indicators of ASD in infants, making them critical markers for early diagnosis. This research focuses on developing a robust system for detecting and analyzing these repetitive motor behaviors using a 1D Convolutional Neural Network (1D CNN) model. The proposed system processes video data of infants, utilizing MediaPipe for precise hand motion tracking and a 1D CNN model for classifying hand-flapping motions as potential markers of ASD. The model was trained and validated on video datasets with varying lengths, demonstrating the effectiveness of capturing both spatial and temporal features of infant motor behaviors. By continuously monitoring and analyzing these specific behavioral patterns, the system aims to provide early predictions regarding the likelihood of ASD, thereby enabling timely alerts for caregivers and healthcare professionals. This approach enhances the accuracy of early autism detection while reducing the reliance on resource-intensive clinical observations, offering a more accessible and efficient solution for early intervention. The study underscores the potential of leveraging advanced deep learning techniques, specifically 1D CNNs, to improve the early detection of ASD and support caregivers in making informed decisions about their child's developmental health.