Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005700
Ibrahim Hammoud, I. Ramakrishnan, M. Henry
In this work, we describe our early sepsis prediction model for the PhysioNet/Computing in Cardiology Challenge 2019. We prove that maximizing a general family of utility functions (of which the challenge utility function is a special case) is equivalent to minimizing a weighted 0-1 loss. We then utilize this fact to train an ensemble of gradient boosting decision trees using a weighted binary cross-entropy loss.Our model takes the time-series nature of the data into account by using a fixed size window of all measurements within the last 20 hours as a feature vector. Data were imputed in a way that gives the same information to the model as present to healthcare professionals in real-time. We tune the model hyper-parameters using 5-fold cross-validation. The model performance was measured on each evaluation set using the threshold that gives the maximum utility on the training set. Our best model achieves an official normalized utility score of 0.332 on the final full test set of the challenge (Team name: SBU, rank: 6th/78).
在这项工作中，我们描述了2019年PhysioNet/Computing In Cardiology Challenge的早期败血症预测模型。我们证明了最大化一般效用函数族(其中挑战效用函数是一种特殊情况)等同于最小化加权0-1损失。然后，我们利用这一事实，使用加权二元交叉熵损失来训练梯度增强决策树的集合。我们的模型通过使用过去20小时内所有测量的固定大小窗口作为特征向量来考虑数据的时间序列性质。数据输入的方式可以向模型提供与实时呈现给医疗保健专业人员相同的信息。我们使用5倍交叉验证来调整模型超参数。在每个评估集上，使用在训练集上给出最大效用的阈值来测量模型的性能。我们最好的模型在挑战的最终完整测试集(团队名称:SBU，排名:第6 /78)上实现了0.332的官方标准化效用得分。
{"title":"Early Prediction of Sepsis Using Gradient Boosting Decision Trees with Optimal Sample Weighting","authors":"Ibrahim Hammoud, I. Ramakrishnan, M. Henry","doi":"10.23919/CinC49843.2019.9005700","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005700","url":null,"abstract":"In this work, we describe our early sepsis prediction model for the PhysioNet/Computing in Cardiology Challenge 2019. We prove that maximizing a general family of utility functions (of which the challenge utility function is a special case) is equivalent to minimizing a weighted 0-1 loss. We then utilize this fact to train an ensemble of gradient boosting decision trees using a weighted binary cross-entropy loss.Our model takes the time-series nature of the data into account by using a fixed size window of all measurements within the last 20 hours as a feature vector. Data were imputed in a way that gives the same information to the model as present to healthcare professionals in real-time. We tune the model hyper-parameters using 5-fold cross-validation. The model performance was measured on each evaluation set using the threshold that gives the maximum utility on the training set. Our best model achieves an official normalized utility score of 0.332 on the final full test set of the challenge (Team name: SBU, rank: 6th/78).","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"1 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82912538","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005898
Vasileios Athanasiou, Z. Konkoli
A supervised learning technique is used to carefully train memristor models to predict at an early stage whether a patient in intensive care unit (ICU) has the sepsis. A memristor behaves as a resistor, with a (mem)resistance that changes over time within a bounded interval. The resistance value depends on the full history of an applied voltage difference across the element, in the same way as the state of the brain depends on what a person has experienced in the past. The information contained in a voltage difference time series can be encoded in the resistance value. Clinical variables measured subsequently each hour since the patient’s admittance in ICU are transformed into voltage difference signals with transformation functions. The training procedure involves the optimization of the transformation functions. The decision of whether to predict sepsis or not is taken by reading the value of the resistance. The authors have participated in the Physionet 2019 challenge with the name called "the memristive agents" and their best submission resulted to a utility score 0.20 on a hidden test data-set.
{"title":"Memristor Models for Early Detection of Sepsis in ICU Patients","authors":"Vasileios Athanasiou, Z. Konkoli","doi":"10.23919/CinC49843.2019.9005898","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005898","url":null,"abstract":"A supervised learning technique is used to carefully train memristor models to predict at an early stage whether a patient in intensive care unit (ICU) has the sepsis. A memristor behaves as a resistor, with a (mem)resistance that changes over time within a bounded interval. The resistance value depends on the full history of an applied voltage difference across the element, in the same way as the state of the brain depends on what a person has experienced in the past. The information contained in a voltage difference time series can be encoded in the resistance value. Clinical variables measured subsequently each hour since the patient’s admittance in ICU are transformed into voltage difference signals with transformation functions. The training procedure involves the optimization of the transformation functions. The decision of whether to predict sepsis or not is taken by reading the value of the resistance. The authors have participated in the Physionet 2019 challenge with the name called \"the memristive agents\" and their best submission resulted to a utility score 0.20 on a hidden test data-set.","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"33 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"83697676","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005893
L. Zhong, S. Leng, Xiaodan Zhao, J. Tan, R. Tan
The transition of right ventricle (RV) from a compensated to decompensated state contributes to survival in pulmonary arterial hypertension (PAH). However, little is known about the significance of right atrial (RA) dysfunction on disease progression in PAH. In this context, there has been growing interest in markers of RA myocardial dysfunction. Speckle tracking echocardiography, which has been principally used to measure the myocardial strain, is technically challenging in the RA due to the thin atrial wall. Feature tracking cardiovascular magnetic resonance (FT-CMR) software designed to derive myocardial strain from CMR cine images has become available for measurements of atrial longitudinal strain. However, in subjects with relatively vigorous tricuspid annular motion, contour tracking of the RA free wall segment adjacent to the tricuspid valve is adversely affected and becomes the source of errors. In contrast to FT-CMR, we present a rapid assessable strain parameter that requires the automatic tracking of only 3 anatomical reference points – thus avoiding the segment contour tracking near the insertion of the anterior leaflet into the tricuspid annulus.
{"title":"Impaired Right Atrial Strain is Associated with Decompensated Hemodynamics in Pulmonary Arterial Hypertension","authors":"L. Zhong, S. Leng, Xiaodan Zhao, J. Tan, R. Tan","doi":"10.23919/CinC49843.2019.9005893","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005893","url":null,"abstract":"The transition of right ventricle (RV) from a compensated to decompensated state contributes to survival in pulmonary arterial hypertension (PAH). However, little is known about the significance of right atrial (RA) dysfunction on disease progression in PAH. In this context, there has been growing interest in markers of RA myocardial dysfunction. Speckle tracking echocardiography, which has been principally used to measure the myocardial strain, is technically challenging in the RA due to the thin atrial wall. Feature tracking cardiovascular magnetic resonance (FT-CMR) software designed to derive myocardial strain from CMR cine images has become available for measurements of atrial longitudinal strain. However, in subjects with relatively vigorous tricuspid annular motion, contour tracking of the RA free wall segment adjacent to the tricuspid valve is adversely affected and becomes the source of errors. In contrast to FT-CMR, we present a rapid assessable strain parameter that requires the automatic tracking of only 3 anatomical reference points – thus avoiding the segment contour tracking near the insertion of the anterior leaflet into the tricuspid annulus.","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"24 1","pages":"Page 1-Page 2"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90197772","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005457
S. Schellenberger, Kilin Shi, J. P. Wiedemann, F. Lurz, R. Weigel, A. Koelpin
Sepsis is a life-threatening condition that has to be treated at an early stage. Doctors use the Sequential Organ Failure Assessment score for the earliest possible recognition. In addition, the practitioner’s many years of experience help in order to facilitate an immediate response. Mortality decreases with every hour that sepsis is detected and treated with antibiotics. In this years PhysioNet/Computing in Cardiology Challenge the objective is to automatically detect sepsis six hours before the clinical prediction. This paper describes the implementation of an Long Short-Term Memory network for an early detection of sepsis in provided hourly physiological data. An utility score of 0.29 was achieved when testing on the full hidden test set. All entries were submitted using the team name "404: Sepsis not found".
败血症是一种危及生命的疾病，必须在早期治疗。医生使用序贯器官衰竭评估评分来尽早识别。此外，从业人员多年的经验有助于迅速作出反应。发现败血症并使用抗生素治疗的时间越长，死亡率就越低。在今年的PhysioNet/Computing In Cardiology挑战赛中，目标是在临床预测前6小时自动检测败血症。这篇论文描述了一个长短期记忆网络的实现，在提供的每小时生理数据中早期检测败血症。在对完整隐藏测试集进行测试时，实现了0.29的效用得分。所有参赛作品都以“404:败血症未找到”的团队名称提交。
{"title":"An Ensemble LSTM Architecture for Clinical Sepsis Detection","authors":"S. Schellenberger, Kilin Shi, J. P. Wiedemann, F. Lurz, R. Weigel, A. Koelpin","doi":"10.23919/CinC49843.2019.9005457","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005457","url":null,"abstract":"Sepsis is a life-threatening condition that has to be treated at an early stage. Doctors use the Sequential Organ Failure Assessment score for the earliest possible recognition. In addition, the practitioner’s many years of experience help in order to facilitate an immediate response. Mortality decreases with every hour that sepsis is detected and treated with antibiotics. In this years PhysioNet/Computing in Cardiology Challenge the objective is to automatically detect sepsis six hours before the clinical prediction. This paper describes the implementation of an Long Short-Term Memory network for an early detection of sepsis in provided hourly physiological data. An utility score of 0.29 was achieved when testing on the full hidden test set. All entries were submitted using the team name \"404: Sepsis not found\".","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"8 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90835087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005882
M. Calvo, V. Rolle, D. Romero, N. Béhar, P. Gomis, P. Mabo, Alfredo I. Hernández
Several autonomic markers were estimated overnight and during exercise and head-up tilt (HUT) testing for 44 BS patients, to design classifiers capable of distinguishing patients at different levels of risk. The classification performance of predictive models built from the optimization of a step-based machine-learning method were compared, so as to identify those autonomic protocols and markers best distinguishing between symptomatic and asymptomatic patients. Although exercise and HUT testing together led to better predictive results than when they were separately assessed, among all analyzed combinations, the night-based classifier presented the best performance (AUC = 95%), using the least amount of features. This optimal features subset was mostly composed of markers extracted between 4 a.m. - 5 a.m. Thus, results provide further evidence for the role of nighttime analysis, mainly during the last hours of sleep, for risk stratification in BS.
{"title":"Multivariate Classification of Brugada Syndrome Patients Based on the Autonomic Response During Sleep, Exercise and Head-up Tilt Testing","authors":"M. Calvo, V. Rolle, D. Romero, N. Béhar, P. Gomis, P. Mabo, Alfredo I. Hernández","doi":"10.23919/CinC49843.2019.9005882","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005882","url":null,"abstract":"Several autonomic markers were estimated overnight and during exercise and head-up tilt (HUT) testing for 44 BS patients, to design classifiers capable of distinguishing patients at different levels of risk. The classification performance of predictive models built from the optimization of a step-based machine-learning method were compared, so as to identify those autonomic protocols and markers best distinguishing between symptomatic and asymptomatic patients. Although exercise and HUT testing together led to better predictive results than when they were separately assessed, among all analyzed combinations, the night-based classifier presented the best performance (AUC = 95%), using the least amount of features. This optimal features subset was mostly composed of markers extracted between 4 a.m. - 5 a.m. Thus, results provide further evidence for the role of nighttime analysis, mainly during the last hours of sleep, for risk stratification in BS.","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"24 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80918011","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005625
V. Sholohov, V. Zverev, A. Kursanov
We developed three-dimensional model of isolated myocardial muscular preparation that takes into account the coupling of excitation with contraction in the myocardium at the cellular and tissue levels. This model describes myocardium sample using approaches and methods developed in continuum mechanics. In the model, electromechanical interactions and mechano-electric feedbacks are realized both at the micro level and at the macro level. We used non-linear partial differential equations describing the deformation of the cardiac tissue, and a detailed "Ekaterinburg-Oxford" (EO) cellular model of the electrical and mechanical activity of cardiomyocytes. Electrical and mechanical interactions between the cells in tissue, as well as intracellular mechano-electric feedback beat-to-beat affect the functional characteristics of coupled cardiomyocytes further, adjusting their electrical and mechanical heterogeneity to the activation timing. Model analysis suggests that cooperative mechanisms of myofilament calcium activation contribute essentially to the generation of cellular functional heterogeneity in contracting cardiac tissue.
{"title":"Investigation of Mechanisms of Regulation of Electromechanical Function of Cardiomyocytes in the Biomechanical Model of Myocardium","authors":"V. Sholohov, V. Zverev, A. Kursanov","doi":"10.23919/CinC49843.2019.9005625","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005625","url":null,"abstract":"We developed three-dimensional model of isolated myocardial muscular preparation that takes into account the coupling of excitation with contraction in the myocardium at the cellular and tissue levels. This model describes myocardium sample using approaches and methods developed in continuum mechanics. In the model, electromechanical interactions and mechano-electric feedbacks are realized both at the micro level and at the macro level. We used non-linear partial differential equations describing the deformation of the cardiac tissue, and a detailed \"Ekaterinburg-Oxford\" (EO) cellular model of the electrical and mechanical activity of cardiomyocytes. Electrical and mechanical interactions between the cells in tissue, as well as intracellular mechano-electric feedback beat-to-beat affect the functional characteristics of coupled cardiomyocytes further, adjusting their electrical and mechanical heterogeneity to the activation timing. Model analysis suggests that cooperative mechanisms of myofilament calcium activation contribute essentially to the generation of cellular functional heterogeneity in contracting cardiac tissue.","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"6 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80365611","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005751
Yale Chang, Jonathan Rubin, G. Boverman, S. Vij, Asif Rahman, A. Natarajan, S. Parvaneh
Early prediction of sepsis onset can notify clinicians to provide timely interventions to patients to improve their clinical outcomes. The key question motivating this work is: given a retrospective patient cohort consisting of multivariate clinical time series (e.g., vital signs and lab measurement) and patients' demographics, how to build a model to predict the onset of sepsis six hours earlier? To tackle this challenge, we first used a recurrent imputation for time series (RITS) approach to impute missing values in multivariate clinical time series. Second, we applied temporal convolutional networks (TCN) to the RITS-imputed data. Compared to other sequence prediction models, TCN can effectively control the size of sequence history. Third, when defining the loss function, we assigned custom time- dependent weights to different types of errors. We achieved 9th place (team name = prna, utility score = 0.328) at the 2019 PhysioNet Computing in Cardiology Challenge, which evaluated our proposed model on a real-world sepsis patient cohort. At a follow-up ‘hackathon’ event, held by the challenge organizers, an improved version of our algorithm achieved 2nd place (utility score = 0.342).
{"title":"A Multi-Task Imputation and Classification Neural Architecture for Early Prediction of Sepsis from Multivariate Clinical Time Series","authors":"Yale Chang, Jonathan Rubin, G. Boverman, S. Vij, Asif Rahman, A. Natarajan, S. Parvaneh","doi":"10.23919/CinC49843.2019.9005751","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005751","url":null,"abstract":"Early prediction of sepsis onset can notify clinicians to provide timely interventions to patients to improve their clinical outcomes. The key question motivating this work is: given a retrospective patient cohort consisting of multivariate clinical time series (e.g., vital signs and lab measurement) and patients' demographics, how to build a model to predict the onset of sepsis six hours earlier? To tackle this challenge, we first used a recurrent imputation for time series (RITS) approach to impute missing values in multivariate clinical time series. Second, we applied temporal convolutional networks (TCN) to the RITS-imputed data. Compared to other sequence prediction models, TCN can effectively control the size of sequence history. Third, when defining the loss function, we assigned custom time- dependent weights to different types of errors. We achieved 9th place (team name = prna, utility score = 0.328) at the 2019 PhysioNet Computing in Cardiology Challenge, which evaluated our proposed model on a real-world sepsis patient cohort. At a follow-up ‘hackathon’ event, held by the challenge organizers, an improved version of our algorithm achieved 2nd place (utility score = 0.342).","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"81 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76713456","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005844
Muhammad Haziq Kamarul Azman, Olivier Meste, D. Latcu, K. Kadir
Atrial flutter presents quasi-periodic atrial activity due to circular depolarization. Given the different structure of right and left atria, spatiotemporal variability should be different. This was analyzed using recurrence quantification analysis. Autocorrelation signals were estimated from the unthresholded recurrence plot, calculated with a properly processed ECG to remove variability related to external sources (noise, respiratory motion, T wave overlap). Simple features were considered from the autocorre-lation that attempts to describe the atrial activity in terms of range of recurrence and periodicity. Linear classification using support vector machines and logistic regression both allowed good classification performance (max accuracy 0.8 for both). Feature selection showed that right and left AFL have significantly different cycle lengths (right vs. left: 230.63 ms vs. 206.50 ms, p < 0.01).
心房扑动由于圆去极化而呈现准周期性心房活动。由于左右心房结构不同，其时空变异性也不同。使用递归定量分析进行分析。从无阈值递归图中估计自相关信号，并使用经过适当处理的ECG进行计算，以去除与外部源(噪声、呼吸运动、T波重叠)相关的变异性。简单的特征是考虑自相关，试图描述心房活动的复发和周期性的范围。使用支持向量机的线性分类和逻辑回归都有很好的分类性能(两者的最大准确率为0.8)。特征选择结果显示，左、右AFL的周期长度存在显著差异(左、右分别为230.63 ms和206.50 ms, p < 0.01)。
{"title":"Non-Invasive Localization of Atrial Flutter Circuit Using Recurrence Quantification Analysis and Machine Learning","authors":"Muhammad Haziq Kamarul Azman, Olivier Meste, D. Latcu, K. Kadir","doi":"10.23919/CinC49843.2019.9005844","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005844","url":null,"abstract":"Atrial flutter presents quasi-periodic atrial activity due to circular depolarization. Given the different structure of right and left atria, spatiotemporal variability should be different. This was analyzed using recurrence quantification analysis. Autocorrelation signals were estimated from the unthresholded recurrence plot, calculated with a properly processed ECG to remove variability related to external sources (noise, respiratory motion, T wave overlap). Simple features were considered from the autocorre-lation that attempts to describe the atrial activity in terms of range of recurrence and periodicity. Linear classification using support vector machines and logistic regression both allowed good classification performance (max accuracy 0.8 for both). Feature selection showed that right and left AFL have significantly different cycle lengths (right vs. left: 230.63 ms vs. 206.50 ms, p < 0.01).","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"21 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78999826","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005766
D. Ge, Alfredo I. Hernández
Robust, real-time apnea and hypopnea detection for monitoring patients suffering from sleep apnea syndrome (SAS) still represents an open problem due to the effect of noise artifacts, the complexity of respiratory patterns and inter-subject variability. We propose in this study the application of an original multi-feature probabilistic detector (MFPD) for SAS event detection during long-term monitoring recordings on three SAS patients. The nasal pressure signal is used as input to derive a set of respiratory features (variance, peak-to-peak amplitude and total respiration cycle) which are statistically characterized during time and used to provide a mono-feature detection probability in realtime. A centralized fusion approach based on the Kullback-Leibler divergence (KLD), optimally combines these mono-feature distributions in order to produce a final detection. While the optimal feature set selection lies beyond the scope of our study, we illustrate the ability to adapt each feature’s weight dynamically to make centralized fusion decisions. The method can be directly applied to data acquired from multiple sensors as long as features are synchronized. Our proposed fusion method achieves a very high sensitivity (94%) as compared with reference thresholding based methods in the literature.
{"title":"Multi-Feature Probabilistic Detector Applied to Apnea/Hypopnea Monitoring","authors":"D. Ge, Alfredo I. Hernández","doi":"10.23919/CinC49843.2019.9005766","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005766","url":null,"abstract":"Robust, real-time apnea and hypopnea detection for monitoring patients suffering from sleep apnea syndrome (SAS) still represents an open problem due to the effect of noise artifacts, the complexity of respiratory patterns and inter-subject variability. We propose in this study the application of an original multi-feature probabilistic detector (MFPD) for SAS event detection during long-term monitoring recordings on three SAS patients. The nasal pressure signal is used as input to derive a set of respiratory features (variance, peak-to-peak amplitude and total respiration cycle) which are statistically characterized during time and used to provide a mono-feature detection probability in realtime. A centralized fusion approach based on the Kullback-Leibler divergence (KLD), optimally combines these mono-feature distributions in order to produce a final detection. While the optimal feature set selection lies beyond the scope of our study, we illustrate the ability to adapt each feature’s weight dynamically to make centralized fusion decisions. The method can be directly applied to data acquired from multiple sensors as long as features are synchronized. Our proposed fusion method achieves a very high sensitivity (94%) as compared with reference thresholding based methods in the literature.","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"67 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75435560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-09-01DOI: 10.23919/CinC49843.2019.9005805
James Morrill, A. Kormilitzin, A. Nevado-Holgado, S. Swaminathan, Sam, Howison, Terry Lyons
Optimal feature selection leads to enhanced efficiency and accuracy when developing both supervised and unsupervised machine-learning models. In this work, a new signature-based regression model is proposed to automatically identify a patient's risk of sepsis based on physiological data streams and to make a positive or negative prediction ofsepsis for every time interval since admission to the intensive care unit. The gradient boosting machine algorithm that uses the features at the current time-points and the signature features extracted from the time-series to model the longitudinal effects ofsepsis yields the utility function score of 0.360 (officially ranked 1st, team name: ‘Can I get your Signature?’) on the full test set. The signature method shows a systematic and competitive approach to model sepsis by learning from health data streams.
在开发有监督和无监督机器学习模型时，最优特征选择可以提高效率和准确性。在这项工作中，提出了一种新的基于特征的回归模型，可以根据生理数据流自动识别患者的脓毒症风险，并在进入重症监护室后的每个时间间隔内对脓毒症进行阳性或阴性预测。梯度增强机算法使用当前时间点的特征和从时间序列中提取的签名特征来模拟脓毒症的纵向影响，在完整的测试集上，效用函数得分为0.360(官方排名第一，团队名称:Can I get your signature ?)签名方法显示了一种通过从健康数据流中学习来模拟败血症的系统和竞争性方法。
{"title":"The Signature-Based Model for Early Detection of Sepsis From Electronic Health Records in the Intensive Care Unit","authors":"James Morrill, A. Kormilitzin, A. Nevado-Holgado, S. Swaminathan, Sam, Howison, Terry Lyons","doi":"10.23919/CinC49843.2019.9005805","DOIUrl":"https://doi.org/10.23919/CinC49843.2019.9005805","url":null,"abstract":"Optimal feature selection leads to enhanced efficiency and accuracy when developing both supervised and unsupervised machine-learning models. In this work, a new signature-based regression model is proposed to automatically identify a patient's risk of sepsis based on physiological data streams and to make a positive or negative prediction ofsepsis for every time interval since admission to the intensive care unit. The gradient boosting machine algorithm that uses the features at the current time-points and the signature features extracted from the time-series to model the longitudinal effects ofsepsis yields the utility function score of 0.360 (officially ranked 1st, team name: ‘Can I get your Signature?’) on the full test set. The signature method shows a systematic and competitive approach to model sepsis by learning from health data streams.","PeriodicalId":6697,"journal":{"name":"2019 Computing in Cardiology (CinC)","volume":"9 1","pages":"Page 1-Page 4"},"PeriodicalIF":0.0,"publicationDate":"2019-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75078478","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：