Pub Date : 2021-09-13DOI: 10.23919/cinc53138.2021.9662779
Federico M. Muscato, V. Corino, L. Mainardi
The automatic detection and classification of cardiac abnormalities can assist physicians in making diagnoses, saving costs in modern healthcare systems. In this study we present an automatic algorithm for classification of cardiac abnormalities included in the CinC's challenge 2021 dataset consisting of twelve-lead, six-lead, three-lead, and two-lead ECGs (team: Polimi_1). For each set of leads an ensemble of three deep learning models, trained on three different subsets, was developed. These subsets, obtained by splitting the recordings with the most frequent classes, had more balanced distributions for training and were used to train the 3 classifiers. The trained models were modified Residual Networks with a Squeeze-and-Excitation module. This module is based on the intuition of channel attention: the basic idea of this approach is to apply a weight to the Convolutional channels based on their relevance in learning before propagating to the next layer. For evaluation, we submitted our model to the official phase of the PhysioNet/Computing in Cardiology Challenge 2021. The model received scores of 0.47, 0.46, 0.45, 0.48 and 0.45 (ranked 14th, 13th, 15th, 10th, and 13th out of 39 teams) on 12-lead, 6-lead, 4-lead, 3-lead, 2-lead hidden test set, respectively; placing us in the 11th position for the mean of the 12-lead, 3-lead, and 2-lead scores.
心脏异常的自动检测和分类可以帮助医生做出诊断，节省现代医疗保健系统的成本。在这项研究中，我们提出了一种用于心脏异常分类的自动算法，该算法包含在CinC的挑战2021数据集中，该数据集由12导联、6导联、3导联和2导联心电图组成(团队:Polimi_1)。对于每组线索，开发了三个深度学习模型的集合，在三个不同的子集上进行训练。这些子集是通过将记录与最频繁的类分开得到的，具有更平衡的训练分布，并用于训练3个分类器。训练后的模型是带有挤压-激励模块的改进残差网络。该模块基于通道注意的直觉:该方法的基本思想是在传播到下一层之前，根据卷积通道在学习中的相关性对其应用权重。为了进行评估，我们将我们的模型提交给了PhysioNet/Computing in Cardiology Challenge 2021的官方阶段。模型在12-lead、6-lead、4-lead、3-lead、2-lead隐藏测试集上的得分分别为0.47、0.46、0.45、0.48、0.45(在39支队伍中排名第14、13、15、10、13);这让我们在12分，3分和2分的平均得分中排名第11位。
{"title":"Ensemble Learning of Modified Residual Networks for Classifying ECG with Different Set of Leads","authors":"Federico M. Muscato, V. Corino, L. Mainardi","doi":"10.23919/cinc53138.2021.9662779","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662779","url":null,"abstract":"The automatic detection and classification of cardiac abnormalities can assist physicians in making diagnoses, saving costs in modern healthcare systems. In this study we present an automatic algorithm for classification of cardiac abnormalities included in the CinC's challenge 2021 dataset consisting of twelve-lead, six-lead, three-lead, and two-lead ECGs (team: Polimi_1). For each set of leads an ensemble of three deep learning models, trained on three different subsets, was developed. These subsets, obtained by splitting the recordings with the most frequent classes, had more balanced distributions for training and were used to train the 3 classifiers. The trained models were modified Residual Networks with a Squeeze-and-Excitation module. This module is based on the intuition of channel attention: the basic idea of this approach is to apply a weight to the Convolutional channels based on their relevance in learning before propagating to the next layer. For evaluation, we submitted our model to the official phase of the PhysioNet/Computing in Cardiology Challenge 2021. The model received scores of 0.47, 0.46, 0.45, 0.48 and 0.45 (ranked 14th, 13th, 15th, 10th, and 13th out of 39 teams) on 12-lead, 6-lead, 4-lead, 3-lead, 2-lead hidden test set, respectively; placing us in the 11th position for the mean of the 12-lead, 3-lead, and 2-lead scores.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128412730","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662818
M. Lange, Eugene Kwan, R. MacLeod, R. Ranjan
Personalized computational models used to guide ablation heavily depend on late gadolinium enhanced images for scar and gray area estimation. The estimation has a high degree of uncertainty, but it is unclear how sensitive the simulation outcome is to the specific scar. In this work, we study the sensitivity of the simulation outcome on the scar. Two personalized left atrial models were generated for a de-novo and a redo atrial. In control setting scar and gray area were obtained by thresholding LGE-MRI images at 70%, and 60% of the maximum myocardial intensity, respectively. This was compared against segmentations, generated by dilating, or eroding the control segmentation by one pixel, and increasing or decreasing the threshold by 5%. The outcomes were normal capture without further activity, extra beats with additional activity but not sustained, sustained arrhythmia with activity until the end of the simulation, and no capture. We found normally captured beats were not affected in redo cases but did change in de-novo ablation. However, extra beats were likely to change to arrhythmia when adding or subtracting scar. Sustained arrhythmia was sensitive to a reduction in scar size. This reiterates that attention is need when determining appropriate thresholds for scar and gray area.
{"title":"Computer Simulations Outcomes of Left Atrial Arrhythmia Induction are Highly Sensitive to Scar and Fibrosis Determination","authors":"M. Lange, Eugene Kwan, R. MacLeod, R. Ranjan","doi":"10.23919/cinc53138.2021.9662818","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662818","url":null,"abstract":"Personalized computational models used to guide ablation heavily depend on late gadolinium enhanced images for scar and gray area estimation. The estimation has a high degree of uncertainty, but it is unclear how sensitive the simulation outcome is to the specific scar. In this work, we study the sensitivity of the simulation outcome on the scar. Two personalized left atrial models were generated for a de-novo and a redo atrial. In control setting scar and gray area were obtained by thresholding LGE-MRI images at 70%, and 60% of the maximum myocardial intensity, respectively. This was compared against segmentations, generated by dilating, or eroding the control segmentation by one pixel, and increasing or decreasing the threshold by 5%. The outcomes were normal capture without further activity, extra beats with additional activity but not sustained, sustained arrhythmia with activity until the end of the simulation, and no capture. We found normally captured beats were not affected in redo cases but did change in de-novo ablation. However, extra beats were likely to change to arrhythmia when adding or subtracting scar. Sustained arrhythmia was sensitive to a reduction in scar size. This reiterates that attention is need when determining appropriate thresholds for scar and gray area.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122147706","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662903
Petra Novotna, Tomáš Vičar, Jakub Hejc, M. Ronzhina
Detection of cardiac arrhythmias is still an ongoing challenge. Here we focus on premature ventricular contraction (PVC) and premature atrial contraction (PAC) and introduce a deep-learning-based method for PVC/PAC localization in ECG. Our method is based on involving the time series with non-zero values corresponding to the ground truth PVC/PAC positions into the training process. To improve the efficiency of deep model training, the transition between the non-zero and zero areas in the train output time series was smoothed by introducing a Gaussian function. When applied to the new ECGs, the output signal (time series including Gaussians) is processed by a robust peak detector with Bayesian optimization of threshold, minimal distance and peak prominence. Positions of the detected peaks correspond to the desired PVC/PAC positions. The proposed method was evaluated on China Physiological Signal Challenge 2018 (CPSC2018) using own-created ground truth positions of PVC/PAC. The proposed method reached F1 score 0.923 and 0.688 for PAC and PVC, respectively, which is better than our previous results obtained via multiple instance learning-based method.
{"title":"Deep-Learning Premature Contraction Localization Using Gaussian Based Predicted Data","authors":"Petra Novotna, Tomáš Vičar, Jakub Hejc, M. Ronzhina","doi":"10.23919/cinc53138.2021.9662903","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662903","url":null,"abstract":"Detection of cardiac arrhythmias is still an ongoing challenge. Here we focus on premature ventricular contraction (PVC) and premature atrial contraction (PAC) and introduce a deep-learning-based method for PVC/PAC localization in ECG. Our method is based on involving the time series with non-zero values corresponding to the ground truth PVC/PAC positions into the training process. To improve the efficiency of deep model training, the transition between the non-zero and zero areas in the train output time series was smoothed by introducing a Gaussian function. When applied to the new ECGs, the output signal (time series including Gaussians) is processed by a robust peak detector with Bayesian optimization of threshold, minimal distance and peak prominence. Positions of the detected peaks correspond to the desired PVC/PAC positions. The proposed method was evaluated on China Physiological Signal Challenge 2018 (CPSC2018) using own-created ground truth positions of PVC/PAC. The proposed method reached F1 score 0.923 and 0.688 for PAC and PVC, respectively, which is better than our previous results obtained via multiple instance learning-based method.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126868599","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662935
B. U. Demirel, Adnan Harun Dogan, M. A. Al Faruque
This paper proposes an efficient convolutional neural network to detect 26 different classes of cardiac activities from different numbers of leads in the Phys-ionetlComputing data in the Cardiology Challenge 2021. The proposed CNN architecture is designed to utilize heart rate variation features from ECG recordings and wave-form morphologies of heartbeats simultaneously. Also, the designed architecture is flexible for the implementation of a different number of leads with a varied length of ECG recordings. The proposed algorithm achieved a score of 0.38 using only 2 channels ofECG on all recordings for the hidden test set of the challenge, placing us 21, 20, 19, 20, 20th (Team name: METU-19) out of 39 teams for 12, 6, 4, 3, and 2-leads respectively. These results show the potential of an efficient, flexible novel neural network for beat-by-beat classification of raw ECG recordings to a complex multi-class multi-label classification problem.
{"title":"Two Might Do: A Beat-by-Beat Classification of Cardiac Abnormalities Using Deep Learning with Domain-Specific Features","authors":"B. U. Demirel, Adnan Harun Dogan, M. A. Al Faruque","doi":"10.23919/cinc53138.2021.9662935","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662935","url":null,"abstract":"This paper proposes an efficient convolutional neural network to detect 26 different classes of cardiac activities from different numbers of leads in the Phys-ionetlComputing data in the Cardiology Challenge 2021. The proposed CNN architecture is designed to utilize heart rate variation features from ECG recordings and wave-form morphologies of heartbeats simultaneously. Also, the designed architecture is flexible for the implementation of a different number of leads with a varied length of ECG recordings. The proposed algorithm achieved a score of 0.38 using only 2 channels ofECG on all recordings for the hidden test set of the challenge, placing us 21, 20, 19, 20, 20th (Team name: METU-19) out of 39 teams for 12, 6, 4, 3, and 2-leads respectively. These results show the potential of an efficient, flexible novel neural network for beat-by-beat classification of raw ECG recordings to a complex multi-class multi-label classification problem.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127092741","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662675
P. Priya, Srinivasan Jayaraman
The interaction mechanisms of Hydroxychloroquine (HCQ) in a COVID-19 infected ventricle and its vulnerability to arrhythmogenesis for different dosage levels is not clearly understood. To address this, a 2D transmural anisotropic ventricular tissue model consisting of endocardial, midmyocardial and epicardial myocytes are config-uredfor mild and severe COVID-19 conditions as well as for three dosage levels of HCQ $1 mu M, 10 mu M$ and 100 $mu M)$. Results show that under control and mild COVID conditions, increasing the dosage of HCQ prolongs the QT interval as well as QRS duration, although under severe COVID-19 conditions, inverted T-waves are observed. In addition, on pacing with premature beats (PBs), it is observed that under all condition, premature ventricular complexes (PVCs) are created at $1 mu M$ and $10 mu M$ HCQ. However, the PVCs are sustained for a longer duration in presence of $10 mu M$ HCQ. ST elevation is observed under mild COVID-19 conditions and $1 mu M$ HCQ and reentrant arrhythmic activity is generated in severe COVID-19 conditions and $10 mu M$ HCQ dosage. Under all conditions, $100 mu M$ HCQ doesn't generate arrhythmia or PVCs in presence of PBs. This in-silico ventricular model indicates that the dosage of HCQ as well as pacing sequence influences the appearance of arrhythmic activity and could help in guiding HCQ therapy.
{"title":"Influence of Hydroxychloroquine Dosage on the Occurrence of Arrhythmia in COVID-19 Infected Ventricle","authors":"P. Priya, Srinivasan Jayaraman","doi":"10.23919/cinc53138.2021.9662675","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662675","url":null,"abstract":"The interaction mechanisms of Hydroxychloroquine (HCQ) in a COVID-19 infected ventricle and its vulnerability to arrhythmogenesis for different dosage levels is not clearly understood. To address this, a 2D transmural anisotropic ventricular tissue model consisting of endocardial, midmyocardial and epicardial myocytes are config-uredfor mild and severe COVID-19 conditions as well as for three dosage levels of HCQ $1 mu M, 10 mu M$ and 100 $mu M)$. Results show that under control and mild COVID conditions, increasing the dosage of HCQ prolongs the QT interval as well as QRS duration, although under severe COVID-19 conditions, inverted T-waves are observed. In addition, on pacing with premature beats (PBs), it is observed that under all condition, premature ventricular complexes (PVCs) are created at $1 mu M$ and $10 mu M$ HCQ. However, the PVCs are sustained for a longer duration in presence of $10 mu M$ HCQ. ST elevation is observed under mild COVID-19 conditions and $1 mu M$ HCQ and reentrant arrhythmic activity is generated in severe COVID-19 conditions and $10 mu M$ HCQ dosage. Under all conditions, $100 mu M$ HCQ doesn't generate arrhythmia or PVCs in presence of PBs. This in-silico ventricular model indicates that the dosage of HCQ as well as pacing sequence influences the appearance of arrhythmic activity and could help in guiding HCQ therapy.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"108 4","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114026631","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662849
G. Luongo, S. Schuler, M. Rivolta, O. Dössel, R. Sassi, A. Loewe
Atrial flutter (AFl) is a common heart rhythm disorder driven by different self-sustaining electrophysiological atrial mechanisms. In this work, we tried to automatically distinguish the macro-mechanism sustaining the arrhythmia in an individual patient using the non-invasive 12-lead electrocardiogram (ECG). We implemented a concurrent clustering and classification algorithm (CCC) to discriminate the clinical classes and look for potential similarities between patient features in each class, thus suggesting that these patients would require a similar treatment. The CCC performance was then compared to a standard supervised technique (K-nearest neighbor, KNN). 3-class classification (macro-reentry right atrium, macro-reentry left atrium, and others) achieved 48.3% and 72.0% CCC and KNN accuracy, respectively. 4-class classification (tri-cuspidal reentry, mitral reentry, fig-8 macro-reentry, and others) achieved 41.6% and 71.2% CCC and KNN accuracy, respectively. Our results show that a clustering approach does not improve the performance of AFl classification because the semi-supervised method leads to clusters that are strongly overlapping between the different ground truth classes. In contrast, the supervised learning approach shows potential for the classification, although constrained by the complexity and the multiple variables that influence the underlying mechanisms.
{"title":"Semi-Supervised vs. Supervised Learning for Discriminating Atrial Flutter Mechanisms Using the 12-lead ECG","authors":"G. Luongo, S. Schuler, M. Rivolta, O. Dössel, R. Sassi, A. Loewe","doi":"10.23919/cinc53138.2021.9662849","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662849","url":null,"abstract":"Atrial flutter (AFl) is a common heart rhythm disorder driven by different self-sustaining electrophysiological atrial mechanisms. In this work, we tried to automatically distinguish the macro-mechanism sustaining the arrhythmia in an individual patient using the non-invasive 12-lead electrocardiogram (ECG). We implemented a concurrent clustering and classification algorithm (CCC) to discriminate the clinical classes and look for potential similarities between patient features in each class, thus suggesting that these patients would require a similar treatment. The CCC performance was then compared to a standard supervised technique (K-nearest neighbor, KNN). 3-class classification (macro-reentry right atrium, macro-reentry left atrium, and others) achieved 48.3% and 72.0% CCC and KNN accuracy, respectively. 4-class classification (tri-cuspidal reentry, mitral reentry, fig-8 macro-reentry, and others) achieved 41.6% and 71.2% CCC and KNN accuracy, respectively. Our results show that a clustering approach does not improve the performance of AFl classification because the semi-supervised method leads to clusters that are strongly overlapping between the different ground truth classes. In contrast, the supervised learning approach shows potential for the classification, although constrained by the complexity and the multiple variables that influence the underlying mechanisms.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"72 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114499751","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662817
M. Rivolta, R. Sassi, L. Mainardi, V. Corino
Aim of this study is to assess, using numerical simulations, the effect of different degrees of ischemia on spatial heterogeneity of ventricular repolarization (SHVR), as evaluated by the V-index. Twelve-lead electrocardiograms were simulated using EGCSIM. Different degrees of ischemia were simulated in three regions, i.e., left anterior descending artery (LAD), right coronary artery (RCA) and left circumflex artery (LCX), by varying the size of the ischemic region (35 mm vs 50 mm), the amplitude of action potentials (APs; maximum reduction of 50%), and by shortening the AP durations (maximum reduction of 35%). The time progression of ischemia was simulated on a time window of 8 minutes in which 30 Monte Carlo simulations of 70 beats were generated each minute. V-index significantly increased at $LCA$ and $RCA$ of 11.2 $pm$ 1.8 ms (+ 35.4%) and $12.6 pm 1.6ms (>+ 39.7%)$ with respect to baseline $(p < 0.05)$, for the ischemic region of 35 mm. The increment was larger for the 50 mm region, in which Vindex approximately doubled. On the other hand, ischemia at LCX resulted in small changes of V-index of about 2 ms for both region sizes $(p < 0.05)$. The study showed that the V-index depended on the ischemic location, its size and electrophysiological changes of APs.
{"title":"Effect of Ischemia on the Spatial Heterogeneity of Ventricular Repolarization: a Simulation Study","authors":"M. Rivolta, R. Sassi, L. Mainardi, V. Corino","doi":"10.23919/cinc53138.2021.9662817","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662817","url":null,"abstract":"Aim of this study is to assess, using numerical simulations, the effect of different degrees of ischemia on spatial heterogeneity of ventricular repolarization (SHVR), as evaluated by the V-index. Twelve-lead electrocardiograms were simulated using EGCSIM. Different degrees of ischemia were simulated in three regions, i.e., left anterior descending artery (LAD), right coronary artery (RCA) and left circumflex artery (LCX), by varying the size of the ischemic region (35 mm vs 50 mm), the amplitude of action potentials (APs; maximum reduction of 50%), and by shortening the AP durations (maximum reduction of 35%). The time progression of ischemia was simulated on a time window of 8 minutes in which 30 Monte Carlo simulations of 70 beats were generated each minute. V-index significantly increased at $LCA$ and $RCA$ of 11.2 $pm$ 1.8 ms (+ 35.4%) and $12.6 pm 1.6ms (>+ 39.7%)$ with respect to baseline $(p < 0.05)$, for the ischemic region of 35 mm. The increment was larger for the 50 mm region, in which Vindex approximately doubled. On the other hand, ischemia at LCX resulted in small changes of V-index of about 2 ms for both region sizes $(p < 0.05)$. The study showed that the V-index depended on the ischemic location, its size and electrophysiological changes of APs.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122287026","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662693
Rui Rodrigues, Paula Couto
We present an approach for automatic cardiac abnormality detection using two leads ECG. This approach was developed in the context of the Physionet/Computing in Cardiology Challenge 2021. Our model is decomposed into an Encoder and a Decoder. It is a huge neural network model with more than 36 million parameters. Although the Challenge training dataset consists of more than 88 thousand annotated ECGs, our model is extremely prone to overfitting to the training data. The encoder is a convolution neural network followed by three transformer encoder blocks. The decoder is a transformer encoder block followed by a feed forward neural network. To reduce the overfitting, we pretrain the Encoder in a semi-supervised way on three tasks. Given an ECG segment, L1, the first task is to detect the QRS on L1; the second task is to predict the ECG shape on an ECG segment, L2 following L1, given the QRS location on $L_{2}$; the third task is to predict the number of samples, after $L_{1}$ , before the next QRS. The Decoder weights were firstly estimated with the frozen Endoder pre-trained parameters and then the whole model parameters were fine-tunned. Our team, named matFCT, received a challenge score of 0.43 on the official test dataset. However, we were unable to qualify for ranking because we weren't able to submit the preprint to the Computing in Cardiology Conference before the deadline.
{"title":"Semi-Supervised Learning for ECG Classification","authors":"Rui Rodrigues, Paula Couto","doi":"10.23919/cinc53138.2021.9662693","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662693","url":null,"abstract":"We present an approach for automatic cardiac abnormality detection using two leads ECG. This approach was developed in the context of the Physionet/Computing in Cardiology Challenge 2021. Our model is decomposed into an Encoder and a Decoder. It is a huge neural network model with more than 36 million parameters. Although the Challenge training dataset consists of more than 88 thousand annotated ECGs, our model is extremely prone to overfitting to the training data. The encoder is a convolution neural network followed by three transformer encoder blocks. The decoder is a transformer encoder block followed by a feed forward neural network. To reduce the overfitting, we pretrain the Encoder in a semi-supervised way on three tasks. Given an ECG segment, L1, the first task is to detect the QRS on L1; the second task is to predict the ECG shape on an ECG segment, L2 following L1, given the QRS location on $L_{2}$; the third task is to predict the number of samples, after $L_{1}$ , before the next QRS. The Decoder weights were firstly estimated with the frozen Endoder pre-trained parameters and then the whole model parameters were fine-tunned. Our team, named matFCT, received a challenge score of 0.43 on the official test dataset. However, we were unable to qualify for ranking because we weren't able to submit the preprint to the Computing in Cardiology Conference before the deadline.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"93 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122694890","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662947
Olivier Meste, S. Zeemering, Joël M. H. Karel, T. Lankveld, U. Schotten, H. Crijns, R. Peeters, P. Bonizzi
In contrast to electrograms, Body-Surface Potential Mapping (BSPM) records the global atrial activity, at the expenses of a lower spatial accuracy. The aim of this study is to investigate whether BSPM recordings can discriminate persistent patients undergoing electrical cardiover-sion, based on the body-surface normalized AF spatial frequency distribution. High-density BSPMs (120 anterior, 64 posterior electrodes) were recorded in 63 patients with persistent AF. For each patient and electrode recording, the frequency content of AF was analyzed on the raw signal, and also by means of the normalized correlation function, and by Singular Spectrum Analysis (SSA). In order to compare the body-surface spatial distributions of AF frequency in all patients, these distributions were first normalized, before performing statistical analysis. We found that the distribution of AF frequency on the body-surface, and its interpretation, are strongly dependent on the specific method employed. Moreover, the estimated body-surface AF frequency was greater over the central posterior and the right anterior BSPM locations. Finally, SSA-based decomposition followed by frequency analysis could discriminate AF patients recurring 4 to 6 weeks after electrical cardioversion from those who did not, based on the frequency content in the proximity of V1.
{"title":"Body-Surface Atrial Signals Analysis Based on Spatial Frequency Distribution: Comparison Between Different Signal Transformations","authors":"Olivier Meste, S. Zeemering, Joël M. H. Karel, T. Lankveld, U. Schotten, H. Crijns, R. Peeters, P. Bonizzi","doi":"10.23919/cinc53138.2021.9662947","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662947","url":null,"abstract":"In contrast to electrograms, Body-Surface Potential Mapping (BSPM) records the global atrial activity, at the expenses of a lower spatial accuracy. The aim of this study is to investigate whether BSPM recordings can discriminate persistent patients undergoing electrical cardiover-sion, based on the body-surface normalized AF spatial frequency distribution. High-density BSPMs (120 anterior, 64 posterior electrodes) were recorded in 63 patients with persistent AF. For each patient and electrode recording, the frequency content of AF was analyzed on the raw signal, and also by means of the normalized correlation function, and by Singular Spectrum Analysis (SSA). In order to compare the body-surface spatial distributions of AF frequency in all patients, these distributions were first normalized, before performing statistical analysis. We found that the distribution of AF frequency on the body-surface, and its interpretation, are strongly dependent on the specific method employed. Moreover, the estimated body-surface AF frequency was greater over the central posterior and the right anterior BSPM locations. Finally, SSA-based decomposition followed by frequency analysis could discriminate AF patients recurring 4 to 6 weeks after electrical cardioversion from those who did not, based on the frequency content in the proximity of V1.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125950028","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 : 2021-09-13DOI: 10.23919/cinc53138.2021.9662795
P. Jurák, P. Leinveber, F. Plesinger, K. Čurila, I. Viscor, V. Vondra, M. Matejkova, L. Znojilova, R. Smíšek, J. Lipoldova, F. Prinzen, J. Halámek
Background: We introduce a new technology that uses the ultra-high-frequency components (150–1000 Hz) of the electrocardiogram (UHF-ECG). Method: The UHF-ECG components represent weak signals generated by the depolarization of myocardial cells. The amplitude of UHF oscillations decreases with distance from the source. This property and the different timing of depolarization in the ventricles' volume enable mapping of the ventricular activation from the chest ECG leads. Because of a low signal-to-noise ratio of UHF oscillations, averaging must be performed. Single recording thus lasts 30 seconds and more. Results: UHF-ECG defines the time-spatial distribution of myocardial electrical activity. Corresponding numerical parameters are electrical dyssynchrony (e-DYS) and the duration of local depolarization (Vd). UHF ventricular depolarization maps present details of electrical activation. Conclusion: The UHF-ECG uses a new source of information originating in ventricular volumes that is different from the standard ECG. It provides information about the volumetric electrical activation associated with mechanical contraction. Its primary clinical utilization is in cardiac resynchronization, pacing optimization, and conduction system pacing.
{"title":"Ultra-High-Frequency Electrocardiography","authors":"P. Jurák, P. Leinveber, F. Plesinger, K. Čurila, I. Viscor, V. Vondra, M. Matejkova, L. Znojilova, R. Smíšek, J. Lipoldova, F. Prinzen, J. Halámek","doi":"10.23919/cinc53138.2021.9662795","DOIUrl":"https://doi.org/10.23919/cinc53138.2021.9662795","url":null,"abstract":"Background: We introduce a new technology that uses the ultra-high-frequency components (150–1000 Hz) of the electrocardiogram (UHF-ECG). Method: The UHF-ECG components represent weak signals generated by the depolarization of myocardial cells. The amplitude of UHF oscillations decreases with distance from the source. This property and the different timing of depolarization in the ventricles' volume enable mapping of the ventricular activation from the chest ECG leads. Because of a low signal-to-noise ratio of UHF oscillations, averaging must be performed. Single recording thus lasts 30 seconds and more. Results: UHF-ECG defines the time-spatial distribution of myocardial electrical activity. Corresponding numerical parameters are electrical dyssynchrony (e-DYS) and the duration of local depolarization (Vd). UHF ventricular depolarization maps present details of electrical activation. Conclusion: The UHF-ECG uses a new source of information originating in ventricular volumes that is different from the standard ECG. It provides information about the volumetric electrical activation associated with mechanical contraction. Its primary clinical utilization is in cardiac resynchronization, pacing optimization, and conduction system pacing.","PeriodicalId":126746,"journal":{"name":"2021 Computing in Cardiology (CinC)","volume":"203 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123731580","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：