Early Detection of ASD from Facial Images Using DeepDeblur and Denoising Optimized Convolutional Neural Network

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder with ineffective brain functions leading to challenges in verbal and non-verbal communication, social behavior, and learning disabilities.Early detection of ASD in children can help determine the required therapies to stabilize physical and psychological behaviors.The accurate diagnosis of ASD has been primarily obtained through analysis of behavioral changes and facial features.Estimating the presence of ASD in children from static facial images is very challenging due to many challenges in the quality and resolution of the images.This paper develops an advanced Deep Learning (DL)-based framework of DeepDeblur-Denoising Optimized Convolutional Neural Network (DeepDeblur-DOCNN) for screening and detection of ASD in children images with the presence of noise and blur.The proposed DeepDeblur-DOCNN framework utilizes the DeepDeblur method in the first stage for image sharpening.Then, it employs a novel DOCNN in the second stage for joint denoising and classification.In this proposed DeepDeblur-DOCNN framework, the DOCNN is enhanced by combining the features of denoising network and CNN with Fire Hawk Optimizer (FHO) for parameter tuning to improve the training process.By combining denoising and classification tasks in a single DOCNN, the model can leverage the denoised image and perform better feature learning for more accurate classification predictions.Experiments are conducted using the Kaggle ASD Children Facial dataset containing 2940 facial images.The results showed that the proposed model is more effective for ASD diagnosis, with 96.94% detection accuracy with reduced model complexity.