A Comprehensive Review Of Multimodal Laryngeal Disorder Detection Using Voice And Imaging Data

Abstract

Laryngeal disorders, including vocal cord nodules, polyps, paralysis, and laryngitis, significantly affect voice production and communication quality. Effective treatment and better patient outcomes depend on early and precise diagnosis. Traditional diagnostic procedures, such as laryngoscopy and clinical voice assessment, provide valuable information but often require specialized equipment and expert interpretation. Contemporary developmentsin artificial intelligence have enabled the development of automated diagnostic systems based on voice recordings, medical imaging, and multimodal data sources. This review presents a comprehensive analysis of deep learning approaches for laryngeal disorder detection using audio signals and laryngeal imaging data. The study examines commonly used acoustic features, including Mel-Frequency Cepstral Coefficients (MFCCs), chroma features, spectral features, and spectrogram representations, as well as image-based feature extraction techniques employed in Convolutional Neural Network (CNN) architectures. Furthermore, recent multimodal learning frameworks that combine voice and image information are analyzed to emphasise their potential for development diagnostic performance. The strengths, limitations, and research challenges of existing approaches are critically discussed, including issues related to dataset availability, model generalization capability, data heterogeneity, and real-world deployment. Finally, future research directions are presented, emphasizing the advancement of robust, accurate, and clinically applicable multimodal systems for reliable laryngeal disorder diagnosis.