Developing an Image-Based Dyslexia Detection Model Using the Deep Learning Technique

Abstract

Dyslexia is a neurological disorder. Across the globe, children are primarily affected by dyslexia. Deep learning (DL) approaches have been applied in dyslexia detection (DD). However, these approaches demand substantial computational resources to generate a meaningful outcome. In addition, healthcare centers face challenges in interpreting the DL-based DD models. Thus, this study aimed to build an effective DD model to support physicians in detecting dyslexic individuals using functional magnetic resonance imaging (FMRI). The authors applied extensive image preprocessing techniques to overcome the FMRI image complexities. They built a convolutional neural network model for extracting the key features from the FMRI images using the weights of the ShuffleNet V2 model. Random forest is ensembled to classify the extracted features. The authors evaluated the proposed model using a real-time dataset comprising 606 multidimensional FMRI images. The findings revealed that the recommended DD model outperformed the existing DD models. The proposed DD model achieved an accuracy of 98.9 and an F1-Score of 99.0. In addition, the proposed model generated an outcome with a minimum loss of 1.2, a standard deviation of 0.0002, and a confidence interval range between 98.2 and 98.7. The experimental outcome supported the effectiveness of the proposed model in detecting dyslexic individuals with few computational resources. The proposed model can be extended using graph convolutional networks for classifying complex images with optimal prediction accuracy.