基於骨架步態藉由機器學習進行臨床老化衰落分類;Skelton-Gait based Clinical Frailty Assessment using Hybrid Ensemble ML Model

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/93264

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/987654321/93264

题名:	基於骨架步態藉由機器學習進行臨床老化衰落分類;Skelton-Gait based Clinical Frailty Assessment using Hybrid Ensemble ML Model
作者:	康致瑋;Kang, Chih-Wei
贡献者:	資訊工程學系
关键词:	步態;臨床衰落;醫療物聯網;圖像辨識;Gait;Clinical Frailty;AIoMT;image recognition
日期:	2023-07-25
上传时间:	2024-09-19 16:51:08 (UTC+8)
出版者:	國立中央大學
摘要:	臨床衰弱，也稱為羸弱綜合症，是一種常見於老年人的醫學狀況，表現為身體、心理和社交功能的下降。它是由於年齡相關的生理變化、慢性疾病和環境因素等多種因素相互作用的結果。臨床上的衰弱個體對壓力因素更加脆弱，並且有更高的跌倒、住院、殘疾和死亡風險。因此要是可以提早發現衰弱趨勢，即可提早應對，減少未來負擔。隨著圖像辨識與骨架追蹤演算法的發展，許多基於骨架的病理步態分類方法近年來已被提出。然而，這些方法少有用來對人體衰弱進行分類，沒有辦法取代複雜的臨床衰落量表。本文通過基於LSTM分類器和全身骨架數據找到的時序性動作特徵，對Clinical Frailty Scale中的四個衰弱等級進行4分類，f1-score為73％，另外加上圖像辨識針對環境進行背景物品特徵提供，可以使分類準確度增加到93％。本研究表明，所提出的方法可以用於支持醫療和臨床決策。且符合AIoMT的需求，可以更簡易的推廣於各處。;Clinical frailty, also known as frailty syndrome, is a common medical condition among the elderly, characterized by a decline in physical, psychological, and social functioning. It is the result of multiple factors such as age-related physiological changes, chronic diseases, and environmental factors. Clinically frail individuals are more vulnerable to stressors and have a higher risk of falls, hospitalization, disability, and mortality. Early detection of frailty trends can enable proactive interventions and reduce future burdens. With the advancement of image recognition and skeleton tracking algorithms, several skeleton-based pathological gait classification methods have been proposed in recent years. However, these methods are rarely applied to classify human frailty and cannot replace complex clinical frailty scales. In this paper, by utilizing an LSTM classifier and temporal motion features extracted from full-body skeleton data, we perform a 4-class classification of the four frailty levels in the Clinical Frailty Scale, achieving an f1-score of 73%. By incorporating image recognition to provide background object features related to the environment, the classification accuracy is increased to 93%. This study demonstrates that the proposed method can support medical and clinical decision-making and meets the requirements of AIoMT, making it easier to generalize across various settings.
显示于类别:	[資訊工程研究所] 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	10	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.....