應用U-NET和Mask R-CNN於鼓膜影像切割及破洞特徵提取實踐居家聽力功能預測及中耳炎自動輔助診斷系統;Home-Based Hearing Loss Prediction and Otitis Media Diagnosis Using U-Net and Mask R-CNN for Tympanic Membrane Image Analysis and Perforation Detection

NCU Institutional Repository > 生醫理工學院 > 生物醫學工程研究所 > 博碩士論文 > Item 987654321/94807

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/94807

题名:	應用U-NET和Mask R-CNN於鼓膜影像切割及破洞特徵提取實踐居家聽力功能預測及中耳炎自動輔助診斷系統;Home-Based Hearing Loss Prediction and Otitis Media Diagnosis Using U-Net and Mask R-CNN for Tympanic Membrane Image Analysis and Perforation Detection
作者:	郭又榕;Kuo, Yu-Jung
贡献者:	生物醫學工程研究所
关键词:	慢性中耳炎;耳鏡;影像分割;Chronic otitis media(COM);Otoscope;Segmentation
日期:	2024-07-29
上传时间:	2024-10-09 15:31:20 (UTC+8)
出版者:	國立中央大學
摘要:	世界衛生組織於2024年2月發表的耳聾和聽力損失報告書中指出，估計於2050年將有超過7億人次發生殘疾性聽力損失。其中，中耳炎是造成殘疾性聽損的原因之一。中耳炎是常見的感染性疾病，根據病程和症狀，可以分為急性、積液性和慢性三類。急性中耳炎通常好發於兒童，但偶而也會發生於成人，特別在季節交替時期，尤為嚴重。慢性化膿性中耳炎則是感染後，持續數週甚至更久的發炎，可能導致鼓膜穿孔。由於患病嚴重者將可能造成永久聽力損害或其他併發症狀，對病患來日生活影響甚深。因此，迫切需要能即時輔助檢測中耳炎病灶之方法。隨著遠程醫療的快速發展，全球許多團隊紛紛開發出各式新型以深度學習為基礎的鼓膜影像輔助診斷策略。本研究的目的亦利用深度學習和影像處理的技術，建立鼓膜分割與電腦輔助診斷系統。此系統技術的核心為建立中耳炎資料庫及即時影像分割系統，實現輔助診斷系統分割鼓膜構造。影像分割所使用的資料來自經國泰醫療財團法人國泰綜合醫院人體試驗審查委員會申請核可計畫編號批准 (IRB編號CGH-CS109004)，共1014張鼓膜影像，研究中比較了U-Net及Mask R-CNN於鼓膜及鼓膜穿孔上的分割結果，Mask R-CN於鼓膜及穿孔上之MIoU分別為0.88及0.87。此外，本研究還初步探討了慢性中耳炎影像與傳導性聽力損失之間的關聯性研究，以拓展系統的應用範圍。使用經國泰醫療財團法人國泰綜合醫院人體試驗審查委員會申請核可計畫批准 (IRB編號CGH-P113003)，研究於2024年4月迄今，研究旨於估計慢性中耳炎之鼓膜破損造成的傳導性聽力受損，比較了基於聲壓差之估計方法與線性回歸方法於聽力損失之表現。 ;The World Health Organization′s report on deafness and hearing loss published in February 2024 stated that it is estimated that more than 700 million people will suffer from disabling hearing loss in 2050. One of the causes of disabling hearing loss is otitis media. Otitis media is a common infectious disease that can be classified into three types based on its course and symptoms: acute, otitis media with effusion, and chronic. Acute otitis media commonly occurs in children but can occasionally affect adults, especially during seasonal transitions, where it can be particularly severe. Chronic suppurative otitis media involves persistent inflammation lasting for weeks or longer after an infection, potentially leading to tympanic membrane perforation. Severe cases can result in permanent hearing damage or other complications, significantly affecting the patient′s future life. Therefore, there is an urgent need for methods that can assist in the timely detection of otitis media lesions. With the rapid development of telemedicine, many teams worldwide are developing various new tympanic membrane imaging diagnostic strategies based on deep learning. The purpose of this study is to use deep learning and image processing technologies to establish a tympanic membrane segmentation and computer-aided diagnosis system. The core of this system′s technology is the creation of an otitis media database and a real-time image segmentation system to enable the computer-aided diagnosis system to segment tympanic membrane structures. The image segmentation data comes from a project approved by the Institutional Review Board of Cathay General Hospital (IRB number CGH-CS109004), consisting of 1,014 tympanic membrane images. We compared the segmentation results of U-Net and Mask R-CNN on tympanic membranes and perforations, with Mask R-CNN achieving MIoU scores of 0.88 and 0.87 on the tympanic membrane and perforations, respectively. Additionally, this study preliminarily explored the correlation between chronic otitis media images and conductive hearing loss to expand the system′s application range. Using data from another project approved by the Institutional Review Board of Cathay General Hospital (IRB number CGH-P113003), conducted from April 2024 to the present, the study aims to estimate the conductive hearing impairment caused by tympanic membrane damage in chronic otitis media. It compares the performance of air pressure difference-based estimation methods and linear regression methods in evaluating hearing loss.
显示于类别:	[生物醫學工程研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	41	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....