Researchers from Scientific Reports, led by Mahmood A. Mahmood, present a novel hybrid model integrating ResNet152 and Vision Transformer that achieves 91.33% accuracy in diagnosing autism through facial expression analysis. By combining convolutional features with transformer attention, the study offers a promising, efficient tool for early detection in clinical settings.