Published Paper Details:

This survey paper explores the progress made in using neural networks for diagnosing Autism Spectrum Disorder (ASD), a condition marked by challenges in social interaction, communication, and behavior. Timely diagnosis is crucial for effective intervention; however, traditional diagnostic methods can be subjective, lengthy, and may not provide the accuracy needed for early detection. As machine learning and artificial intelligence (AI) advance, neural networks present promising opportunities to automate and enhance the accuracy of ASD diagnosis, thereby reducing reliance on conventional assessment techniques. The paper systematically reviews recent studies on the use of neural networks for ASD, focusing on the analysis of data from behavioral assessments, speech patterns, and imaging data, including MRI and other imaging modalities. Key research is assessed based on the neural network architectures employed, the sources of datasets, and the performance metrics used, offering insights into the strengths and weaknesses of different approaches. Our analysis highlights significant challenges in current research, such as limited data availability, model overfitting, and the difficulty of translating complex neural network models into tools that can be easily interpreted in clinical settings. The importance of developing models that generalize effectively across diverse populations and provide reliable insights in real-world scenarios is underscored. Moreover, this survey points out critical gaps in the existing literature and proposes avenues for future research. It recommends the creation of self-updating datasets as a means to ensure that models remain relevant and accurate as new data emerges. Additionally, combining various data types--such as behavioral, speech, and imaging data--could improve diagnostic accuracy and offer a more comprehensive understanding of ASD characteristics. In summarizing the current landscape of neural network applications for ASD diagnosis, this survey seeks to inform future developments aimed at creating more robust, adaptable, and accessible diagnostic solutions.