在模擬隔離病房中應用深度機器學習方法做跌倒辨識

Taipei Medical University Institutional Repository > 醫學科技學院 > 醫學資訊研究所 > 博碩士論文 > Item 987654321/64981

請使用永久網址來引用或連結此文件: http://libir.tmu.edu.tw/handle/987654321/64981

題名:	在模擬隔離病房中應用深度機器學習方法做跌倒辨識 Falling posture recognition using deep machine learning in simulated isolation wards
作者:	劉湘華 LIU, HSIANG-HUA
貢獻者:	醫學資訊研究所碩士在職專班林明錦
關鍵詞:	跌倒;InceptionV3;隔離病房;照護者不足;Grad-CAM Falling;InceptionV3;Isolation ward;Lack of caregiver;Grad-CAM
日期:	2024-06-20
上傳時間:	2025-01-06 09:13:00 (UTC+8)
摘要:	病患在住院期間內發生跌倒可能會顯著影響治療策略、增加醫療量能、延長住院天數，並加重醫療人員的負擔。根據台灣病人安全資訊網（TPR）的統計資料顯示，醫療環境中的跌倒可能造成嚴重傷害，並對病患的安全與健康福祉產生重大影響。為了提升隔離病房的照護，我們的研究著重於跌倒影像識別。目的是確保長時間獨自留在隔離病房的病人安全，並預防潛在的傷害。我們利用InceptionV3搭配2D-CNN及5fold交叉驗證架構，在模擬隔離病房環境中進行跌倒姿態的識別。透過應用深度機器學習的方法，我們提出了一個能夠迅速分類跌倒的解決方案，使得在關鍵的「黃金救援時段」內能夠及時介入，提升病人安全。在我們的研究中，我們旨在利用從影像中提取特徵來構建一個預測模型，並用於模擬隔離病房中分類跌倒與非跌倒的情況。本研究共有12名受試者，我們在醫院內選擇了10個模擬隔離病房，代表多種環境。這些受試者們分別會在不同環境中錄製影片，以完成本實驗影像採集的部分。我們的數據庫包含10組樣本作為訓練集，2組樣本作為測試集，共計849張影像。通過設置自定義層的InceptionV3與2D-CNN的預測模型，我們的最佳模型能有效識別模擬隔離病房中的跌倒和非跌倒姿態，AUROC可達到0.89，此方法別具潛力提升病人安全，尤其是在缺乏照護者的醫療保健機構中。 Falling incident in patient during hospitalization can significantly impact therapeutic strategies, increase costs, extend hospital stays, and burden healthcare staff. According to statistics from the Taiwan Patient Safety Reporting System (TPR) records, falls in healthcare settings can result in serious injuries and have a substantial impact on patient safety and well-being. To enhance isolation ward care, our research focuses on fall recognition. The aim of our study is to ensure the safety of patients who stay alone for prolonged periods and prevent potential injuries. Utilizing InceptionV3 with 2D-CNN and 5-fold cross-validation architectures enables precise fall posture recognition in a simulated isolation ward environment. By leveraging deep machine learning technology, we propose a solution that can promptly classify falls, enabling timely intervention during the critical "golden hour" and reducing the risk of further complications. In our study, we aimed to utilize features derived from images to construct a predictive model for classifying falling and non-falling in simulated isolation wards. A total of 12 participants were involved. We selected 10 simulated isolation wards within the hospital to represent a variety of environments. Each of the 12 participants recorded videos in these diverse settings to gather image data. Our database includes 10 samples as the training set and 2 samples as the testing set, comprising a total of 849 images. Utilizing InceptionV3 with 2D-CNN by setting a custom layer, our best model effectively recognizes falling and non-falling postures within a simulated isolation ward, achieving an AUROC of 0.89. This approach has the potential to improve patient safety, especially in medical care institutions where there is a lack of caregivers.
描述:	碩士指導教授：林明錦口試委員：林明錦口試委員：楊軒佳口試委員：邱泓文
附註:	論文公開日期：2024-07-25
資料類型:	thesis
顯示於類別:	[醫學資訊研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	70	檢視/開啟

在TMUIR中所有的資料項目都受到原著作權保護.

TAIR相關文章

資料載入中.....