A deep learning-based tool for the diagnostic decision support of peripheral vestibular disorders

• Autor(a) principal: Castro, Antônia de Maria Rodrigues de Sousa
• Data de Publicação: 2022
• Outros Autores: Teles, Ariel Soares, Lima, Lucas Daniel Batista, Fontenele, José Everton da Silva, Bastos, Victor Hugo do Vale, Teixeira, Silmar Silva
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/27753
• topic: Nystagmus; Videonystagmography; Peripheral vestibulopathy; Diagnosis; Artificial intelligence; Computer vision; Deep learning.; Nistagmo; Videonistagmografía; Vestibulopatía periférica; Diagnóstico; Inteligencia artificial; Visión por computador; Aprendizaje profundo.; Nistagmo; Videonistagmografia; Vestibulopatia periférica; Diagnóstico; Inteligência Artificial; Visão Computacional; Aprendizado profundo.

Nystagmus is involuntary eye movement characterized by smooth movement, called the slow phase of nystagmus, interrupted by rapid fixation in the opposite direction. It is one of the preponderant factors in the diagnosis of vestibular disorders. This study presents Smart Nystagmography, a proposal for a computer vision-based tool to support the diagnosis of peripheral vestibular disorders, which encompasses the entire process, from the eye movement collection device to the disorder classifier. The proposed solution is based on feature vectors that present eye movement patterns, which are identified using machine learning, in particular, Deep Learning (DL). The videonystagmography technique and its different tests were performed by the subjects in order to generate a representative dataset for both healthy subjects and those with vestibular dysfunction. Data pre-processing methods, as well as a hyperparameter optimization technique of the DL algorithms were employed with the purpose of improving the performance of state-of-the-art models. The performance results for identifying the presence of peripheral vestibular dysfunction reached an accuracy of 96.7% for the best model, after going through the optimization process. The results show the efficiency of Smart Nystagmography, which has a solution that involves from the video collection device to the system with data preparation techniques and the DL model deployed. Additional clinical studies are needed to validate the solution