Fusing Continuous-Valued Medical Labels Using a Bayesian Model

Zhu, Tingting; Dunkley, Nic; Behar, Joachim A.; Clifton, David A.; Clifford, Gari D.

doi:10.1007/s10439-015-1344-1

Cited by 20 publications

(24 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although, the PPG dataset utilized in this study was diligently annotated by multiple clinical adjudicators, the ECG dataset labels were obtained from automatic bedside Holter software, with a coarse overview by a single adjudicator for potential mislabeled cases of AF. Therefore, the ECG-based AF detection results provided in this work should be taken with some caution [11,63,64]. In fact, as demonstrated in Table 1 re-defining the labels as a function of AF burden had a significant impact on the performance of the classifier (AUC of 0.94 at 5% or more versus AUC of 0.97 at 75% or more AF burden over a 10 minute segment).…”

Section: Discussionmentioning

confidence: 87%

Detection of Paroxysmal Atrial Fibrillation using Attention-based Bidirectional Recurrent Neural Networks

Shashikumar

Shah

Clifford

et al. 2018

Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery &Amp; Data Mining

Self Cite

View full text Add to dashboard Cite

Detection of atrial fibrillation (AF), a type of cardiac arrhythmia, is difficult since many cases of AF are usually clinically silent and undiagnosed. In particular paroxysmal AF is a form of AF that occurs occasionally, and has a higher probability of being undetected. In this work, we present an attention based deep learning framework for detection of paroxysmal AF episodes from a sequence of windows. Time-frequency representation of 30 seconds recording windows, over a 10 minute data segment, are fed sequentially into a deep convolutional neural network for image-based feature extraction, which are then presented to a bidirectional recurrent neural network with an attention layer for AF detection. To demonstrate the effectiveness of the proposed framework for transient AF detection, we use a database of 24 hour Holter Electrocardiogram (ECG) recordings acquired from 2850 patients at the University of Virginia heart station. The algorithm achieves an AUC of 0.94 on the testing set, which exceeds the performance of baseline models. We also demonstrate the cross-domain generalizablity of the approach by adapting the learned model parameters from one recording modality (ECG) to another (photoplethysmogram) with improved AF detection performance. The proposed high accuracy, low false alarm algorithm for detecting paroxysmal AF has potential applications in long-term monitoring using wearable sensors. KEYWORDSatrial fibrillation, convolutional neural network, recurrent neural network, deep learning, transfer learning arXiv:1805.09133v1 [q-bio.NC]

show abstract

Section: Discussionmentioning

confidence: 87%

Detection of Paroxysmal Atrial Fibrillation using Attention-based Bidirectional Recurrent Neural Networks

Shashikumar

Shah

Clifford

et al. 2018

Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery &Amp; Data Mining

Self Cite

View full text Add to dashboard Cite

show abstract

“…It is also important to note that too many naive voters (more than 13 in the case of the 2015 Challenge) can reduce the accuracy of the label or answer. In Zhu et al (2014) and Zhu et al (2015) a voting system for algorithm (and human) annotations of physiological data was described, which incorporates both the physiology and the individual annotator's accuracy as a function of objective features (such as signal quality) to produce a weighted voting scheme to guarantee that all voters added extra information. We suggest that such approaches will become ever more important as computational power becomes increasingly less expensive.…”

Section: Discussionmentioning

confidence: 99%

False alarm reduction in critical care

et al. 2016

View full text Add to dashboard Cite

High false alarm rates in the ICU decrease quality of care by slowing staff response times while increasing patient delirium through noise pollution. The 2015 Physio-Net/Computing in Cardiology Challenge provides a set of 1,250 multi-parameter ICU data segments associated with critical arrhythmia alarms, and challenges the general research community to address the issue of false alarm suppression using all available signals. Each data segment was 5 minutes long (for real time analysis), ending at the time of the alarm. For retrospective analysis, we provided a further 30 seconds of data after the alarm was triggered. A total of 750 data segments were made available for training and 500 were held back for testing. Each alarm was reviewed by expert annotators, at least two of whom agreed that the alarm was either true or false. Challenge participants were invited to submit a complete, working algorithm to distinguish true from false alarms, and received a score based on their program's performance on the hidden test set. This score was based on the percentage of alarms correct, but with a penalty that weights the suppression of true alarms five times more heavily than acceptance of false alarms. We provided three example entries based on well-known, open source signal processing algorithms, to serve as a basis for comparison and as a starting point for participants to develop their own code. A total of 38 teams submitted a total of 215 entries in this year's Challenge. This editorial reviews the background issues for this Challenge, the design of the Challenge itself, the key achievements, and the follow-up research generated as a result of the Challenge, published in the concurrent special issue of Physiological Measurement. Additionally we make some recommendations for future changes in the field of patient monitoring as a result of the Challenge.

show abstract

“…This shows that the cluster type assignment can still be improved. One can for example consider a Bayesian voting approach, which has been proven to be more robust than simple majority voting [11]. Another improvement might come from better feature extraction.…”

Section: Discussionmentioning

confidence: 99%

Automated ECG Ventricular Beat Detection with Switching Kalman Filters

Oster

Tarassenko

2016

2016 Computing in Cardiology Conference (CinC)

View full text Add to dashboard Cite

The exponential rise in availability of clinical data, and especially physiological recordings made using wearables, creates a real need for highly accurate and fully automated analysis techniques. An automated detection of ventricular beat in the ECG is proposed, which is an extension of a recently published switching Kalman filter (skf) approach. The latter technique enables automatic selection of the most likely mode (beat type), and makes novelty detection possible by incorporating a mode for unknown morphologies (X-factor). The previously published technique is semi-supervised and relies on the manual annotation of the different clusters (or modes), thus making it less readily applicable in Big Data scenarios. Here we propose to extend the switching Kalman filter technique by automating the labelling of the modes. Each heartbeat in a mode was classified individually using a featurebased approach, and the cluster was assigned a given type by majority voting. Two different feature-based classifications were tested. First, ecgkit, a state-of-the-art toolkit recently made available online provides a heartbeat classification based on clustering and Linear Discriminant Analysis. Second, a Support Vector Machine (svm) approach was used with the same features as ecgkit. Therefore two different automated switching Kalman filter techniques were tested, ecgkit-skf and svm-skf, that differed only in the way the modes were classified. Both approaches were assessed on an independent subset of the MIT-BIH arrhythmia database (22 individual subjects, 30-minute recordings), and were compared to the semi-supervised switching Kalman filter approach (skf), as well as to the classification techniques, ecgkit and svm. F1 varied from 81.2% for ecgkit, 85.4% for svm, 91.8% for ecgkit-skf, 92.3% for svm-skf, and 98.6% for skf. The proposed combined techniques demonstrated improved automatic beat classification, compared to state-of-the-art fully automated techniques (ecgkit). Performances were however still lower than what was achieved with semi-supervised techniques (skf) highlighting the fact that some clusters were mislabeled. IntroductionWith the exponential rise in the acquisition of physiological data, often for phenotyping purposes, there is an increased importance for the extraction of meaningful information from this vast quantity of data. Cardiac aplications are no exception, and it is widely accepted that big data analytics in cardiology will lead to improved patient outcomes in cardiovacular disease.Several applications will require the development of robust and fully automated data analysis techniques, among them:(i) telemedicine and in particular mHealth applications as the tool for reaching a wider population and predicting the advent of serious pathologies [1] or (ii) the prevention, diagnosis and treatment of a wide range of serious and life-threatening illnesses. These applications are supported by the large databases such as Physionet [2], and longitudinal studies such as the UK Biobank[3].When it comes t...

show abstract

Fusing Continuous-Valued Medical Labels Using a Bayesian Model

Cited by 20 publications

References 20 publications

Detection of Paroxysmal Atrial Fibrillation using Attention-based Bidirectional Recurrent Neural Networks

Detection of Paroxysmal Atrial Fibrillation using Attention-based Bidirectional Recurrent Neural Networks

False alarm reduction in critical care

Automated ECG Ventricular Beat Detection with Switching Kalman Filters

Contact Info

Product

Resources

About