Classification of Autism Spectrum Disorder using CNN & Transfer Learning

Autism spectrum disorder (ASD) is a neurodevelopment condition that impacts a person's ability to communicate and interact socially. Diagnosing autism spectrum disorder (ASD) early and correctly is critical for providing appropriate help to those who need it. Transferring information or learned representations from one task or domain to another is possible using a computerized learning technique known as transfer learning. In this study, we proposed a method for classifying ASD using an Enhanced Layerwise Sequential Convolutional Neural Network (ELS-CNN) with transfer learning. Transfer learning involves leveraging pre-trained models on large-scale datasets used for the Kaggle platform and fine-tuning them for specific tasks. A dataset that tested these models was collected and consisted of 2,928 face images. These images are preprocessed using Adaptive Median Filtering to remove noise from an image and Histogram Equalization to improve image contrast. Otsu's Binary Threshold can be used for image segmentation. Extracting the features can be Principal Component Analysis (PCA) for image compression. We suggested the ELS-CNN algorithm, which has demonstrated good performance in a number of image identification applications. The output of the deep learning neural network models was assessed using common assessment measures like accuracy, precision, recall, and fl-score. The Enhanced Layerwise Sequential CNN algorithm achieved the highest accuracy.