指導教授：傅立成臺灣大學：資訊工程學研究所黃冠翎Huang, Kuan-LingKuan-LingHuang2014-11-262018-07-052014-11-262018-07-052014http://ntur.lib.ntu.edu.tw//handle/246246/261481由於醫療技術的發展與進步，患者罹患癌症接受化療或手術治療等的存活率大幅地提高，伴隨著人口高齡化，從癌症治療後存活的長者亦成了不可忽視的一個族群。根據研究，他們十分擔心癌症的復發，也屬於罹患老年疾病的高風險族群，比方說心血管疾病等老年病。因此對於癌症存活長者(以下簡稱年長癌友)而言，家中若擁有能初步評估整體健康情況的照護系統將可帶來許多幫助。然而目前的居家生理資訊量測系統，主要利用普遍的醫學標準來給予使用者警示，而缺少了針對年長癌友族群個人化的資料分析；而針對生理資訊進行風險分析的系統，藉由醫生診斷及病歷等多方資訊的輔助，他們可預測特定疾病的發生，但無法針對年長癌友做整體的健康評估。因此我們提出一套以年長癌友為中心之的健康監測與健康初估系統，且為了減輕使用上的負擔，年長癌友可在家中與行動裝置上之應用程式配合生理資訊量測器檢測其健康狀態；從多個生理訊號分析出對年長癌友必須特別注意的特徵資訊之後，我們提出的動態貝氏網路可以學習這些特徵隨時間的變化，並進一步對健康進行初步的評估。為了驗證此系統的有效性，我們邀請兩位曾罹患乳癌的年長癌友來使用本系統，進行約三週的生理資訊量測，而資料分析結果也顯示了，提出的動態貝氏網路模型可分析針對年長癌友相關聯之生理特徵變化有效地進行健康初估。Survival rate of cancer patient has recently been greatly raised due to the advances of clinical technology. With the aging population, elder people recovering from cancer ac-count for large portion in the society. According to the research literatures, elderly can-cer survivors (ECSs) are afraid of the cancer recurrence, and suffer from geriatric syn-dromes with high risk. Thus, there is an increasing need for ECSs to be aware of their health status at home environment. However, most of current home monitoring systems use common medical standard to identify the hazardous situation without referring to ESC-centric personalized data analysis. Besides, those vital-sign-based risk analysis systems are able to estimate the occurrence of certain disease since multivariate clinical information such as medical diagnosis and disease history are available, whereas they are cannot estimate the overall health status for ECSs. Therefore, in this thesis we propose an ECS-centric health monitoring and its preliminary estimating system. The underlying analysis technique of the system is to utilize dynamic Bayesian network (DBN) to esti-mate ECS’s preliminary health condition. Specifically, after relevant features are ex-tracted from vital sign of elders, the developed DBN is able to first learn the temporal pattern of those features under various corresponded health situation, and then further estimate their current preliminary health statuses (e.g. poor or good). To verify the plau-sibility of our proposed system, two elderly survivors with breast cancer history are in-vited to apply the system for about 3 weeks, and the analysis results show that the abil-ity of estimation and friendliness of our system are quite promising.誌謝 i 中文摘要 ii Abstract iii Table of Contents iv List of Figures vi List of Tables vii Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Objectives and Contributions 3 1.3 Challenges 4 1.4 Related Work 7 1.4.1 Bio-Monitoring System 7 1.4.2 Risk Analysis for Vital Signs 9 1.5 Thesis Organization 11 Chapter 2 Backgrounds 12 2.1 Hidden Markov Model (HMM) 12 2.1.1 Basic Elements 14 2.1.2 Inference 16 2.1.1 Decoding 17 2.1.2 Learning the Model Parameters 18 2.2 Dynamic Bayesian Network 20 2.2.1 Representation 21 2.2.2 Learning 22 Chapter 3 Health Assessment Model 24 3.1 Overview 24 3.2 Physiological Factors that Cancer Survivors Should Notice 26 3.2.1 Weight 26 3.2.2 Blood Pressure 29 3.2.3 Heart Rate Variability 30 3.2.4 Fatigue 34 3.2.5 Self-rated Health (SRH) 35 3.2.6 Functional Status: Health Ground Truth 36 3.3 DBN Model for Health Assessment 38 3.3.1 Feature Extraction: Random Variable in DBN 38 3.3.2 DBN Structure 46 Chapter 4 System Evaluation 49 4.1 Vital Sign Monitoring Platform 49 4.2 Data Collection 56 4.3 Analysis 58 4.3.1 Trend of Feature Data 58 4.3.2 Performance of Health Estimation 61 Chapter 5 Conclusion 69 5.1 Summery 69 5.2 Future Work 70 REFERENCE 713306801 bytesapplication/pdf論文公開時間：2017/08/26論文使用權限：同意有償授權(權利金給回饋學校)生物資訊學癌症存活長者生理時間序列動態貝氏網路健康評估[SDGs]SDG3thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/261481/1/ntu-103-R01922068-1.pdf