NeuroMimicNet: A Multimodal Neuromorphic Framework for Early Screening and Cognitive Pattern Recognition in Children with Autism.

Background: Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition which is composed of social, behavioral, and communication challenges that generally require early detection. Purpose: This research proposes NeuroMimicNet, a brain-like event-driven neuromorphic computing framework designed for early screening and cognitive pattern recognition in children with autism. The framework integrates audio signal and facial expression images as multimodal inputs to capture both neural and behavioral patterns. Methods: The audio modality involves pre-processing steps including noise elimination and normalization, while the neuromorphic processing of audio features is done with Spike-Timing-Dependent Plasticity (STDP), Hebbian learning and a Loihi-inspired spiking neural processing model to capture temporal auditory patterns efficiently. The facial expression images are pre-processed through face alignment, resizing and normalization, and high-level visual features are extracted using a pretrained ResNet50 convolutional neural network. The extracted audio and image features are fused at the feature level to form a unified multimodal representation. The fused features are then classified into four ASD severity level such as Typical, Mild, Moderate, and Severe using a supervised classification model. Results: Performance is evaluated using accuracy, F1-score, precision, recall, and computational efficiency across benchmark EEG-autism datasets. Experimental results demonstrate that NeuroMimicNet achieves higher accuracy and faster response compared to conventional deep learning models. Conclusion: NeuroMimicNet highlights the potential of biologically inspired neuromorphic computing for pediatric ASD screening. By combining multimodal behavioral and neural cues with event-driven processing, the framework offers improved interpretability, scalability and clinical relevance.