Lecture Description
Obstructive sleep apnea syndrome (OSAS), the number of which has increased, is diagnosed by several methods including polysomnography. To address time and cost issues from the existing tools, computational fluid dynamics (CFD) has been used for the upper airway geometry obtained from computed tomography (CT) data. A patient with sleep apnea suffers from considerable pressure drop owing to the narrow shape of the airway, which is the main indicator of OSAS. To address the time and cost issues exhibited by the existing diagnosis methods, this study presents computational fluid dynamics (CFD) and machine learning approaches that are derived from the upper airway morphology with automatic segmentation by deep learning.
Our team develops CFD models of the upper airway to increase the quantity of airway model data and simulates them to obtain the aerodynamic features to predict the severity of sleep apnea. In order to overcome the high computational time costs, this study uses a machine-learning algorithm. We use multivariate Gaussian process regression to enable fast predictions of aerodynamic features of unknown patients. Unlike existing regression methods, multivariate Gaussian regression considers the correlation between outputs, and the algorithm can reduce the computational costs of CFD. Additionally, we use the support vector machine algorithm to classify the patients as normal or moderate OSAS.
Recently we developed an advanced OSAS diagnosis method, which was based on upper airway morphology with automatic segmentation using deep learning was developed using CFD and machine learning approaches. By auto segmentation algorithm, we can remove excessive labor and spent time to extract the morphological factors of the upper airway. Furthermore, using regression and classification models, we could obtain immediate flow characteristics and subsequent patient diagnosis results. These processes are fully automatic. Therefore, this will help clinicians by making real-time diagnosis convenient and possible.
October 28-30, 2022 กค COEX, Seoul, Korea