| Summary: | The common learning disability known as dyslexia can have a negative impact on reading and writing abilities, so early diagnosis is essential for providing proper intervention and support. Furthermore, the traditional approach to dyslexia screening has addressed several drawbacks, including subjectivity, consumption of time, and a propensity for bias since it depends on the evaluator's observation and rating. Technology has demonstrated that computer-aided methods for dyslexia screening are effective and reliable assessments. Directed Acyclic Graph (DAG) networks outperform sequential networks in Convolutional Neural Networks (CNNs) due to their capacity to capture both local and global features of the input data. This study investigates the classification performance of DAG Networks using dyslexic handwriting images. The study further explores the impact of skip connections in DAG networks on dyslexia screening by comparing four networks: DAG-I, DAG-2, DAG-3, and DAG4. The findings revealed a high accuracy of 99.01% in training and 89.68% in testing accuracy, indicating the expectations of DAG Networks as a dependable and helpful method for dyslexia screening. The results also demonstrated that skip connections could enhance the performance of DAG networks in identifying complex features in images of dyslexia handwriting. The proposed method can be incorporated into a computer-assisted dyslexia screening system to provide a rapid, accurate, and unbiased dyslexia assessment. © 2023 IEEE.
|
|---|