Abstract
PROBLEM: Dyslexia is a learning disorder affecting an individual's ability to recognize words and understand concepts. It remains underdiagnosed due to its complexity and heterogeneity. The use of traditional assessment techniques, including subjective evaluation and standardized tests, increases the likelihood of delayed or incorrect diagnosis. MOTIVATION: Timely identification is essential to provide personalized treatment and improve the individual's quality of life. The development of artificial intelligence techniques offers a platform to identify dyslexia using behavior and neuroimaging data. However, the limited datasets and black-box nature of ML models reduce the generalizability and interpretability of dyslexia detection (DD) models. The dimensionality reduction technique (DRT) plays a significant role in providing dyslexia features to enhance the performance of machine learning (ML)- and deep learning (DL)-based DD techniques. AIM: In this review, the authors intend to investigate the role of DRTs in enhancing the performance of ML- and DL-based DD models. METHODOLOGY: The authors conducted a comprehensive search across multiple digital libraries, including Scopus, Web of Science, PubMed, and IEEEXplore, to identify articles associated with DRTs in identifying dyslexia. They extracted 479 articles using these digital libraries. After an extensive screening procedure, a total of 39 articles were included in this review. RESULTS: The review findings revealed various DRTs for identifying critical dyslexia patterns from multiple modalities. A significant number of studies employed principal component analysis (PCA) for feature extraction and selection. The authors presented the essential features associated with DD. In addition, they outlined the challenges and limitations of existing DRTs. IMPACT: The authors emphasized the need for the development of novel DRTs and their seamless integration with advanced DL techniques for robust and interpretable DD models.