DefinePK

Abstract

The lifelong condition named Autism Spectrum Disorder (ASD) prevents people from communicating, interacting socially, and behaving in ways normal people would. Early diagnosis is very important because it means that individuals who are affected by ASD get the right help and support. Traditional diagnosis methods are however, are time consuming as well as inaccurate. In order to solve these problems this research brings a deep deep learning framework which enables ASD to be detected faster and more reliably. As it is, the proposed framework is composed of two extremely strong deep learning systems, namely ResNet50 and a Bottleneck Residual Deep (BRD5). ResNet50 aids in extracting high level spatial as well as hierarchical features from images, while BRD5 brings in efficiency and accuracy in the process of classification. Each of these models uses a depth concatenation layer to combine feature important features and permits overall best performance. All of the autism image data loaded will be fed to the model using the autism-image-data. The data is processed before training by many steps such as normalizing and augmenting it, to allow the model to learn effectively and give the right result. It is shown through extensive testing on the dataset that this framework outperforms classical deep learning methods. Mainly precision, recall and computational efficiency improvements, it achieves an accuracy of 98.8%. Since the system can properly differentiate autistic from non-autistic individuals, it is a fast and reliable tool for ASD diagnosis. However, this study shows the possibility of artificial intelligence in medical imaging and ASD detection. In future research the model can be extended by including more diverse data sources and validated in real clinical environments to further make it effective in practical medical applications.