Journal of the Formosan Medical Association = Taiwan yi zhi · 2026
Improving OSA diagnosis: An AI-powered approach for analyzing DISE videos to assess obstruction site and severity.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Chang Ting-So, Liu Chen-You, Lai Ying-Hui, Hsueh Ming Kuei, Lee Ren-Guey
Аннотация
BACKGROUND: Obstructive sleep apnea (OSA) is a prevalent sleep disorder, Drug-induced sleep endoscopy (DISE) offers direct visualization of upper airway obstruction and collapse. Unlike awake fiberoptic examinations, DISE allows for direct visualization of airway collapse during sleep, simulating natural sleep conditions under intravenous anesthesia. This is crucial for identifying obstruction sites and patterns, especially before surgical interventions or after failed CPAP treatment. While the VOTE classification system is commonly used to assess collapse severity at different anatomical sites. It relies on manual, time-consuming and subjective scoring. This study aims to develop an automated learning-based analysis system to identify the velum and epiglottis and their obstruction ratios from continuous DISE images in the supine position. This approach could provide a more efficient and objective assessment of upper airway collapse during DISE.
METHODS: We developed an AI-assisted system that performs anatomical region classification, continuous obstruction rate estimation, and velum collapse pattern recognition across all four VOTE regions. A total of 165 DISE videos were collected, with 157 used for model training and 8 for clinical validation. From these videos, 2295 frames were extracted and annotated by physicians with region labels and obstruction grades.To enhance performance in anatomically ambiguous regions, such as the tongue base and oropharynx, the system incorporates depth-aware compensation and distance-based frame selection.
RESULTS: The model achieved high prediction accuracy in anatomical classification (Velum 82.47 %, Epiglottis: 85.48 %) significantly improved obstruction grading in complex regions, increasing accuracy from 51.78 % to 78.32 % in the tongue base and from 48.38 % to 77.66 % in the oropharynx. Velum collapse patterns were identified with 81.07 % accuracy. The use of continuous obstruction rates provided more detailed and consistent assessments than conventional grading.
METHODS: We developed an AI-assisted system that performs anatomical region classification, continuous obstruction rate estimation, and velum collapse pattern recognition across all four VOTE regions. A total of 165 DISE videos were collected, with 157 used for model training and 8 for clinical validation. From these videos, 2295 frames were extracted and annotated by physicians with region labels and obstruction grades.To enhance performance in anatomically ambiguous regions, such as the tongue base and oropharynx, the system incorporates depth-aware compensation and distance-based frame selection.
RESULTS: The model achieved high prediction accuracy in anatomical classification (Velum 82.47 %, Epiglottis: 85.48 %) significantly improved obstruction grading in complex regions, increasing accuracy from 51.78 % to 78.32 % in the tongue base and from 48.38 % to 77.66 % in the oropharynx. Velum collapse patterns were identified with 81.07 % accuracy. The use of continuous obstruction rates provided more detailed and consistent assessments than conventional grading.