http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
- 中文 을 입력하시려면 zhongwen을 입력하시고 space를누르시면됩니다.
- 北京 을 입력하시려면 beijing을 입력하시고 space를 누르시면 됩니다.
RISS 인기검색어
- 검색키워드
- 저자 : 엄희상 (검색결과 2 건)
-
어텐션 메커니즘 기반 딥러닝 알고리즘을 이용한 연속적 혈압 추정 연구
(This paper is based on the previous work. ‘H. Eom, D. Lee, S. Han, Y. S. Hariyani, Y. Lim, I. Sohn, K. Park and C. Park, “End-to-End Deep Learning Architecture for Continuous Blood Pressure Estimation Using Attention Mechanism.” Sensors, Volume 20, no. 8, pp. 2338, 2020.’) Blood pressure (BP) is one of the vital signs that provides fundamental human health information. Continuous BP monitoring is important for people who have problems such as hypertension or hypotension. Various studies have proposed cuff-less BP monitoring methods using pulse transit time, a feature extracted from electrocardiogram (ECG) and photoplethysmogram (PPG) signals. In this paper, we proposed an attention-based deep learning architecture using only raw signals of ECG, PPG, and ballistocardiogram (BCG) without the process of extracting features, which improves the performance of continuous BP estimation. The proposed model consists of a convolutional neural network (CNN), a bidirectional gated recurrent unit (Bi-GRU), and an attention mechanism. The model was trained by a calibration-based method, using all combinations of ECG, PPG, and BCG of each subject, and the performance was evaluated for each combination. A total of 15 subjects were recruited, and ECG, PPG, BCG, and BP levels were measured. The 95% confidence interval of the reference BP was [86.34, 143.74] and [51.28, 88.74] for systolic BP (SBP) and diastolic BP (DBP), respectively. The coefficients of determination between the estimated blood pressure of the proposed model and the reference blood pressure (SBP and DBP) were 0.52, 0.49, and their mean absolute error values were 4.06±4.04, 3.33±3.42 for SBP and DBP. The results complied with global standards and showed better performance than other state-of-the-art methods, including conventional linear regression method using pulse transit time (PTT). In this paper, we confirmed that the continuous blood pressure estimation is possible using deep learning and attention mechanisms with only raw signals from ECG, PPG, and BCG without feature extraction process. (본 논문은 ‘H. Eom, D. Lee, S. Han, Y. S. Hariyani, Y. Lim, I. Sohn, K. Park and C. Park, “End-to-End Deep Learning Architecture for Continuous Blood Pressure Estimation Using Attention Mechanism.” Sensors, Volume 20, no. 8, pp. 2338, 2020.’ 을 기반으로 작성됨.) 혈압은 사람의 기본적인 건강정보를 제공하는 핵심적인 신호 중 하나이다. 연속적으로 혈압을 모니터링 하는 것은 고혈압이나 저혈압과 같은 문제를 겪고 있는 사람에게 있 어서 중요하다. 이를 위하여 다양한 연구에서 심전도와 맥파 신호를 이용하여 추출할 수 있는 특징인 맥파전달시간(pulse transit time, PTT)을 활용한 커프리스(cuff-less) 방식의 혈압 추정 방법이 제안되었다. 본 논문에서는 연속적인 혈압 추정 성능을 향상 시키고 심전도, 맥파 및 심탄도의 원시신호(raw signal)만을 사용하여 특징 추출 과정 이 필요 없는 어텐션(attention) 메커니즘 기반의 딥러닝 모델을 제안하였다. 본 논문에 서 제안된 모델은 합성곱 신경망(convolutional neural network, CNN)과 양방향 순환 신경망(bidirectional gated recurrent Unit, Bi-GRU) 및 어텐션 메커니즘(attention mechanism)으로 구성되었다. 모델은 각 피실험자의 심전도, 맥파 및 심탄도 신호 데이 터의 모든 조합을 입력으로 하여 보정 기반(calibration-based)으로 학습되었고 각 조합 에 대한 성능을 평가하였다. 총 15명의 피실험자의 심전도, 맥파 및 심탄도 신호와 혈 압을 측정하였고 수집된 혈압분포에서 95% 신뢰구간은 수축기혈압(systolic blood pressure, SBP)과 이완기혈압(diastolic blood pressure, DBP)에 대하여 각각 [86.34, 143.74], [51.28, 88.74]였다. 제안된 모델의 예측 혈압과 실제혈압 사이의 결정계수 (coefficient of determination)는 SBP와 DBP에 대하여 각각 0.52, 0.49였고, 평균절대오 차(mean absolute error, MAE)는 4.06±4.04, 3.33±3.42였다. 이 결과는 혈압계 세계표준 허용오차를 준수하였으며, 이전 연구에서 제시한 PTT를 사용한 선형회귀모델을 포함 한 관련 연구에서 제안한 모델의 성능보다 우수하였다. 이를 통하여 심전도, 맥파 및 심탄도 신호에서 별도의 특징 추출과정 없이 원시신호(raw signal)만으로도 딥러닝 및 attention mechanism을 활용하여 연속적인 혈압 예측이 가능함을 확인하였다.
-
엄희상 Seoul National University 2016 국내석사
Introduction : Several epidemiological studies have investigated fine particulate matter (≤ 2.5 μm in aerodynamic diameter, PM2.5) has a risk for adverse effects on human health. Previous studies have focused on the risk associated with the total mass of particles, without considering the chemical constituents of them. In this study, the hourly differences between PM2.5 chemical constituents and emergency visits for cardiovascular disease and respiratory disease were estimated using time-stratified case-crossover design. Methods: The study periods were from January 1 to December 31, 2013 in Seoul, Korea. Hourly health outcome data on emergency department visits for cardiovascular disease and respiratory disease were provided by National Emergency Department Information System (NEDIS). Emergency department visits data were classified according to the discharge diagnosis for cardiovascular disease and respiratory disease (ICD-10, cardiovascular, I00-I99 and respiratory, J00-J99). Hourly data for PM2.5 mass and chemical constituents were measured by real-time monitoring at one sampling site located at Bulgwang-dong, Seoul (37.36° N, 126.56° E). In this study, PM2.5 mass and only 13 chemical constituents (OC, EC, Cl-, Mg2+, Na+, NH4+, NO3-, SO42-, Ca, Fe, K, Pb, and Zn), were selected after QA/QC procedure. The meteorological data such as hourly mean temperature (℃), relative humidity (%), and air pressure (hPa) were adjusted as confounding variable. Time-stratified case-crossover analysis and conditional logistic regression analysis were used to estimate the adverse health effects of fine air particles and to estimate and adjusted odds ratio (ORs) and 95% confidence intervals (CIs), respectively. The short-term effects were estimated using moving averages in six periods (1-6(h), 7-12(h), 13-18(h), 19-24(h), 25-48(h), 49-72(h)) and adjustments of this association by age (≥ 65 years) and season. Results and Discussion : The strongest adverse effects for cardiovascular disease exacerbations were associated with PM2.5 mass, OC, EC, Cl-, Ca, Fe and Zn after 19-24h lag period and NH4+, NO3-, and SO42- after 25-48h lag period were estimated. The strongest adverse effects for respiratory disease exacerbations were associted with NO3-, K and Pb after short lag periods (0-6h and 7-12h) and PM2.5, OC, EC, Cl-, NH4+, SO42-, Ca, Fe, and Zn after longer lag periods (19-24h and 25-48h). For those older than ≥ 65 years, the strongest adverse effects for cardiovascular disease exacerbations were shown with PM2.5 mass, OC, EC, Cl-, Ca, Fe, and Zn after 19-24h lag period and NO3- after 25-48 h lag period and respiratory disease exacerbations of OC, EC, Fe and Zn after 19-24h lag period were observed. Especially, among PM2.5 chemical constituents, EC showed the strongest association with cardiovascular disease and respiratory disease exacerbations. For all of seasons, significant positive associations for PM2.5 mass and chemical constituents excluding Mg2+ were observed for cardiovascular and respiratory disease. Conclusion : This study found major differences of associations between PM2.5 constituents and emergency visits for cardiovascular and respiratory disease in Seoul. This study will provide robust evidences for the health impacts of PM2.5 chemical constituents.
연관 검색어 추천
이 검색어로 많이 본 자료
활용도 높은 자료
