Authors: Lokesh, Saurav Ingale, Ayush Kapse, Om Solanke, Milind Ankleshwa, Professor Kirti Randhe
Abstract: This survey paper investigates advancements in applying neural networks to Autism Spectrum Disorder (ASD) diagnosis, a condition characterized by challenges in communication, social interaction, and behavioral patterns. With early intervention critical for positive outcomes, traditional diagnostic methods are often time-consuming, subjective, and prone to limitations in accuracy. Emerging technologies like neural networks offer promising solutions for automating and improving ASD diagnostics. Our study systematically reviews current applications of neural networks, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), in analyzing behavioral patterns and facial image data. Key findings underscore the strengths of these models in capturing distinct ASD traits while also addressing challenges such as overfitting, data scarcity, and model generalizability. The integration of multi-modal data—such as combining behavioral cues with facial analysis—is explored as a pathway for enhancing diagnostic precision. While demonstrating the potential of these techniques, this paper highlights ethical considerations, including data privacy and the interpretability of neural network-based decisions in clinical settings. Future directions focus on developing self-updating datasets, promoting explainable AI, and fostering global collaborations to ensure diverse and representative data pools. This comprehensive review aims to guide the development of innovative, scalable, and ethically compliant diagnostic tools that make early ASD diagnosis more accessible and reliable.
