Adaptive rhythm sequencing: A method for dynamic rhythm classification during CPR

Continuous Assessment of Ventricular Fibrillation Prognostic Status during CPR: Implications for Resuscitation

BACKGROUND

Ventricular fibrillation (VF) waveform measures reflect myocardial physiologic status. Continuous assessment of VF prognosis using such measures could guide resuscitation, but has not been possible due to CPR artifact in the ECG. A recently-validated VF measure (termed VitalityScore), which estimates the probability (0-100%) of return-of-rhythm (ROR) after shock, can assess VF during CPR, suggesting potential for continuous application during resuscitation.

OBJECTIVE

We evaluated VF using VitalityScore to characterize VF prognostic status continuously during resuscitation.

METHODS

We characterized VF using VitalityScore during 60 seconds of CPR and 10 seconds of subsequent pre-shock CPR interruption in patients with out-of-hospital VF arrest. VitalityScore utility was quantified using area under the receiver operating characteristic curve (AUC). VitalityScore trends over time were estimated using mixed-effects models, and associations between trends and ROR were evaluated using logistic models. A sensitivity analysis characterized VF during protracted (100-second) periods of CPR.

RESULTS

We evaluated 724 VF episodes among 434 patients. After an initial decline from 0-8 seconds following VF onset, VitalityScore increased slightly during CPR from 8-60 seconds (slope: 0.18 %/min). During the first 10 seconds of subsequent pre-shock CPR interruption, VitalityScore declined (slope: -14 %/min). VitalityScore predicted ROR throughout CPR with AUCs 0.73-0.75. Individual VitalityScore trends during 8-60 seconds of CPR were marginally associated with subsequent ROR (adjusted odds ratio for interquartile slope change (OR)=1.10, p=0.21), and became significant with protracted (≥100 seconds) CPR duration (OR=1.28, p=0.006).

CONCLUSION

VF prognostic status can be continuously evaluated during resuscitation, a development that could translate to patient-specific resuscitation strategies.

A method for continuous rhythm classification and early detection of ventricular fibrillation during CPR

AIM

We developed a method which continuously classifies the ECG rhythm during CPR in order to guide clinical care.

METHODS

We conducted a retrospective study of 432 patients treated following out-of-hospital cardiac arrest. Continuous ECG sequences from two-minute CPR cycles were extracted from defibrillator recordings and further divided into five-second clips. We developed an algorithm using wavelet analysis, hidden semi-Markov modeling, and random forest classification. The algorithm classifies individual clips as asystole, organized rhythm, ventricular fibrillation, or Inconclusive, while integrating information from previous clips. Classifications were compared to manual annotations to estimate accuracy in an independent validation dataset. Continuous sequences were classified as shockable, non-shockable, or Inconclusive; classifications were used to compute shock sensitivity and specificity.

RESULTS

Of 432 patient-cases, 290 were used for development and 142 for validation. In the 12,294 validation ECG clips during CPR, accuracies were 0.88 (95% CI 0.85-0.91) for asystole, 0.98 (95% CI 0.98-0.99) for organized rhythm, and 0.97 (95% CI 0.96-0.97) for ventricular fibrillation, with 43% classified as Inconclusive. Of 457 continuous sequences, shock sensitivity was 0.90 (95% CI 0.86, 0.93), shock specificity was 0.98 (95% CI 0.93, 0.99), and 7% were Inconclusive. Median delay to ventricular fibrillation recognition was 10 (IQR 5, 32) seconds.

CONCLUSION

An automated algorithm continuously classified the primary resuscitation rhythms-asystole, organized rhythms, and ventricular fibrillation-with 88-98% accuracy, enabling accurate shock advisory guidance during most two-minute CPR cycles. Additional investigation is required to understand how algorithm implementation could affect rescuer actions and clinical outcomes.

Machine learning and feature engineering for predicting pulse presence during chest compressions

Current resuscitation protocols require pausing chest compressions during cardiopulmonary resuscitation (CPR) to check for a pulse. However, pausing CPR when a patient is pulseless can worsen patient outcomes. Our objective was to design and evaluate an ECG-based algorithm that predicts pulse presence with or without CPR. We evaluated 383 patients being treated for out-of-hospital cardiac arrest with real-time ECG, impedance and audio recordings. Paired ECG segments having an organized rhythm immediately preceding a pulse check (during CPR) and during the pulse check (without CPR) were extracted. Patients were randomly divided into 60% training and 40% test groups. From training data, we developed an algorithm to predict the clinical pulse presence based on the wavelet transform of the bandpass-filtered ECG. Principal component analysis was used to reduce dimensionality, and we then trained a linear discriminant model using three principal component modes as input features. Overall, 38% (351/912) of checks had a spontaneous pulse. AUCs for predicting pulse presence with and without CPR on test data were 0.84 (95% CI (0.80, 0.88)) and 0.89 (95% CI (0.86, 0.92)), respectively. This ECG-based algorithm demonstrates potential to improve resuscitation by predicting the presence of a spontaneous pulse without pausing CPR with moderate accuracy.

A method to predict ventricular fibrillation shock outcome during chest compressions

BACKGROUND

Out-of-hospital ventricular fibrillation (VF) cardiac arrest is a leading cause of death. Quantitative analysis of the VF electrocardiogram (ECG) can predict patient outcomes and could potentially enable a patient-specific, guided approach to resuscitation. However, VF analysis during resuscitation is confounded by cardiopulmonary resuscitation (CPR) artifact in the ECG, challenging continuous application to guide therapy throughout resuscitation. We therefore sought to design a method to predict VF shock outcomes during CPR.

METHODS

Study data included 4577 5-s VF segments collected during and without CPR prior to defibrillation attempts in N = 1151 arrest patients. Using training data (460 patients), an algorithm was designed to predict the VF shock outcomes of defibrillation success (return of organized ventricular rhythm) and functional survival (Cerebral Performance Category 1-2). The algorithm was designed with variable-frequency notch filters to reduce CPR artifact in the ECG based on real-time chest compression rate. Ten ECG features and three dichotomous patient characteristics were developed to predict outcomes. These variables were combined using support vector machines and logistic regression. Algorithm performance was evaluated by area under the receiver operating characteristic curve (AUC) to predict outcomes in validation data (691 patients).

RESULTS

AUC (95% Confidence Interval) for predicting defibrillation success was 0.74 (0.71-0.77) during CPR and 0.77 (0.74-0.79) without CPR. AUC for predicting functional survival was 0.75 (0.72-0.78) during CPR and 0.76 (0.74-0.79) without CPR.

CONCLUSION

A novel algorithm predicted defibrillation success and functional survival during ongoing CPR following VF arrest, providing a potential proof-of-concept towards real-time guidance of resuscitation therapy.

Rhythm Analysis during Cardiopulmonary Resuscitation Using Convolutional Neural Networks

Chest compressions during cardiopulmonary resuscitation (CPR) induce artifacts in the ECG that may provoque inaccurate rhythm classification by the algorithm of the defibrillator. The objective of this study was to design an algorithm to produce reliable shock/no-shock decisions during CPR using convolutional neural networks (CNN). A total of 3319 ECG segments of 9 s extracted during chest compressions were used, whereof 586 were shockable and 2733 nonshockable. Chest compression artifacts were removed using a Recursive Least Squares (RLS) filter, and the filtered ECG was fed to a CNN classifier with three convolutional blocks and two fully connected layers for the shock/no-shock classification. A 5-fold cross validation architecture was adopted to train/test the algorithm, and the proccess was repeated 100 times to statistically characterize the performance. The proposed architecture was compared to the most accurate algorithms that include handcrafted ECG features and a random forest classifier (baseline model). The median (90% confidence interval) sensitivity, specificity, accuracy and balanced accuracy of the method were 95.8% (94.6-96.8), 96.1% (95.8-96.5), 96.1% (95.7-96.4) and 96.0% (95.5-96.5), respectively. The proposed algorithm outperformed the baseline model by 0.6-points in accuracy. This new approach shows the potential of deep learning methods to provide reliable diagnosis of the cardiac rhythm without interrupting chest compression therapy.

A Method to Detect Presence of Chest Compressions During Resuscitation Using Transthoracic Impedance

OBJECTIVE

Interruptions in chest compressions during treatment of out-of-hospital cardiac arrest are associated with lower likelihood of successful resuscitation. Real-time automated detection of chest compressions may improve CPR administration during resuscitation, and could facilitate application of next-generation ECG algorithms that employ different parameters depending on compression state. In contrast to accelerometer sensors, transthoracic impedance (TTI) is commonly acquired by defibrillators. We sought to develop and evaluate the performance of a TTI-based algorithm to automatically detect chest compressions.

METHODS

Five-second TTI segments were collected from patients with out-of-hospital cardiac arrest treated by one of four defibrillator models. Segments with and without chest compressions were collected prior to each of the first four defibrillation shocks (when available) from each case. Patients were divided randomly into 40% training and 60% validation groups. From the training segments, we identified spectral and time-domain features of the TTI associated with compressions. We used logistic regression to predict compression state from these features. Performance was measured by sensitivity and specificity in the validation set. The relationship between performance and TTI segment length was also evaluated.

RESULTS

The algorithm was trained using 1859 segments from 460 training patients. Validation sensitivity and specificity were >98% using 2727 segments from 691 validation patients. Validation performance was significantly reduced using segments shorter than 3.2 s.

CONCLUSIONS

A novel method can reliably detect the presence of chest compressions using TTI. These results suggest potential to provide real-time feedback in order to improve CPR performance or facilitate next-generation ECG rhythm algorithms during resuscitation.

The performance of a new shock advisory algorithm to reduce interruptions during CPR

OBJECTIVE

To explore a new algorithm and strategy for rhythm analysis during chest compressions (CCs), and to improve the efficiency of cardiopulmonary resuscitation (CPR) by minimizing interruptions.

METHODS

The clinical data and ECG of patients with sudden cardiac arrest (CA) from three hospitals in China were collected with Philips MRx monitor/defibrillators. The length of each analyzed ECG segment was 23 s, the first 11.5 s was selected to contain CPR compressions, the next 5 s had no compressions, and the last 6.5 s had no requirement. Three experienced emergency doctors annotated the ECG segments without compression artifacts. A two-step analysis through CPR (ATC) algorithm was applied to the selected data. The first step was analysis during chest compressions. If a shockable rhythm was not detected, compression-free analysis followed. The results of the ATC algorithm were compared with the annotations by the physicians, to determine the sensitivity and specificity of the algorithm.

RESULTS

In total 166 CA patients were included with 100 out-of-hospital cardiac arrest (OHCA) patients and 66 in-hospital cardiac arrest (IHCA) patients. A total of 1578 ECG segments were analyzed, including 115 (7.3%) shockable rhythms, 1278 (81.0%) non-shockable rhythms, and 185 (11.7%) intermediate/unknown rhythms. The specificity of all non-shockable rhythms was 99.8% at the end of chest compressions, and 99.5% after analysis without compression artifact. 70.5% of ventricular fibrillation (VF) rhythms were detected by the end of chest compressions. After the CC-free analysis, 93.6% of VF was identified.

CONCLUSION

The ATC algorithm achieved sensitivity of 93.6% and specificity of 99.5% after the two-step analysis, and 70.5% of the patients with shockable rhythms did not require CC-free analysis. Such an approach has the potential to substantially reduce CC interruptions when identifying shockable rhythms.

A Machine Learning Shock Decision Algorithm for Use During Piston-Driven Chest Compressions

GOAL

Accurate shock decision methods during piston-driven cardiopulmonary resuscitation (CPR) would contribute to improve therapy and increase cardiac arrest survival rates. The best current methods are computationally demanding, and their accuracy could be improved. The objective of this work was to introduce a computationally efficient algorithm for shock decision during piston-driven CPR with increased accuracy.

METHODS

The study dataset contains 201 shockable and 844 nonshockable ECG segments from 230 cardiac arrest patients treated with the LUCAS-2 mechanical CPR device. Compression artifacts were removed using the state-of-the-art adaptive filters, and shock/no-shock discrimination features were extracted from the stationary wavelet transform analysis of the filtered ECG, and fed to a support vector machine (SVM) classifier. Quasi-stratified patient wise nested cross-validation was used for feature selection and SVM hyperparameter optimization. The procedure was repeated 50 times to statistically characterize the results.

RESULTS

Best results were obtained for a six-feature classifier with mean (standard deviation) sensitivity, specificity, and total accuracy of 97.5 (0.4), 98.2 (0.4), and 98.1 (0.3), respectively. The algorithm presented a five-fold reduction in computational demands when compared to the best available methods, while improving their balanced accuracy by 3 points.

CONCLUSIONS

The accuracy of the best available methods was improved while drastically reducing the computational demands.

SIGNIFICANCE

An efficient and accurate method for shock decisions during mechanical CPR is now available to improve therapy and contribute to increase cardiac arrest survival.

A Multistage Algorithm for ECG Rhythm Analysis During Piston-Driven Mechanical Chest Compressions

GOAL

An accurate rhythm analysis during cardiopulmonary resuscitation (CPR) would contribute to increase the survival from out-of-hospital cardiac arrest. Piston-driven mechanical compression devices are frequently used to deliver CPR. The objective of this paper was to design a method to accurately diagnose the rhythm during compressions delivered by a piston-driven device.

METHODS

Data was gathered from 230 out-of-hospital cardiac arrest patients treated with the LUCAS 2 mechanical CPR device. The dataset comprised 201 shockable and 844 nonshockable ECG segments, whereof 270 were asystole (AS) and 574 organized rhythm (OR). A multistage algorithm (MSA) was designed, which included two artifact filters based on a recursive least squares algorithm, a rhythm analysis algorithm from a commercial defibrillator, and an ECG-slope-based rhythm classifier. Data was partitioned randomly and patient-wise into training (60%) and test (40%) for optimization and validation, and statistically meaningful results were obtained repeating the process 500 times.

RESULTS

The mean (standard deviation) sensitivity (SE) for shockable rhythms, specificity (SP) for nonshockable rhythms, and the total accuracy of the MSA solution were: 91.7 (6.0), 98.1 (1.1), and 96.9 (0.9), respectively. The SP for AS and OR were 98.0 (1.7) and 98.1 (1.4), respectively.

CONCLUSIONS

The SE/SP were above the 90%/95% values recommended by the American Heart Association for shockable and nonshockable rhythms other than sinus rhythm, respectively.

SIGNIFICANCE

It is possible to accurately diagnose the rhythm during mechanical chest compressions and the results considerably improve those obtained by previous algorithms.

An automatic system for the comprehensive retrospective analysis of cardiac rhythms in resuscitation episodes

AIM

An automatic resuscitation rhythm annotator (ARA) would facilitate and enhance retrospective analysis of resuscitation data, contributing to a better understanding of the interplay between therapy and patient response. The objective of this study was to define, implement, and demonstrate an ARA architecture for complete resuscitation episodes, including chest compression pauses (CC-pauses) and chest compression intervals (CC-intervals).

METHODS

We analyzed 126.5h of ECG and accelerometer-based chest-compression depth data from 281 out-of-hospital cardiac arrest (OHCA) patients. Data were annotated by expert reviewers into asystole (AS), pulseless electrical activity (PEA), pulse-generating rhythm (PR), ventricular fibrillation (VF), and ventricular tachycardia (VT). Clinical pulse annotations were based on patient-charts and impedance measurements. An ARA was developed for CC-pauses, and was used in combination with a chest compression artefact removal filter during CC-intervals. The performance of the ARA was assessed in terms of the unweighted mean of sensitivities (UMS).

RESULTS

The UMS of the ARA were 75.0% during CC-pauses and 52.5% during CC-intervals, 55-points and 32.5-points over a random guess (20% for five categories). Filtering increased the UMS during CC-intervals by 5.2-points. Sensitivities for AS, PEA, PR, VF, and VT were 66.8%, 55.8%, 86.5%, 82.1% and 83.8% during CC-pauses; and 51.1%, 34.1%, 58.7%, 86.4%, and 32.1% during CC-intervals.

CONCLUSIONS

A general ARA architecture was defined and demonstrated on a comprehensive OHCA dataset. Results showed that semi-automatic resuscitation rhythm annotation, which may involve further revision/correction by clinicians for quality assurance, is feasible. The performance (UMS) dropped significantly during CC-intervals and sensitivity was lowest for PEA.

Ventricular fibrillation waveform measures combined with prior shock outcome predict defibrillation success during cardiopulmonary resuscitation

AIM

Amplitude Spectrum Area (AMSA) and Median Slope (MS) are ventricular fibrillation (VF) waveform measures that predict defibrillation shock success. Cardiopulmonary resuscitation (CPR) obscures electrocardiograms and must be paused for analysis. Studies suggest waveform measures better predict subsequent shock success when combined with prior shock success. We determined whether this relationship applies during CPR.

METHODS

AMSA and MS were calculated from 5-second pre-shock segments with and without CPR, and compared to logistic models combining each measure with prior return of organized rhythm (ROR).

RESULTS

VF segments from 692 patients were analyzed during CPR before 1372 shocks and without CPR before 1283 shocks. Combining waveform measures with prior ROR increased areas under receiver operating characteristic curves for AMSA/MS with CPR (0.66/0.68 to 0.73/0.74, p<0.001) and without CPR (0.71/0.72 to 0.76/0.76, p<0.001).

CONCLUSIONS

Prior ROR improves prediction of shock success during CPR, and may enable waveform measure calculation without chest compression pauses.

ECG-Based Classification of Resuscitation Cardiac Rhythms for Retrospective Data Analysis

OBJECTIVE

There is a need to monitor the heart rhythm in resuscitation to improve treatment quality. Resuscitation rhythms are categorized into: ventricular tachycardia (VT), ventricular fibrillation (VF), pulseless electrical activity (PEA), asystole (AS), and pulse-generating rhythm (PR). Manual annotation of rhythms is time-consuming and infeasible for large datasets. Our objective was to develop ECG-based algorithms for the retrospective and automatic classification of resuscitation cardiac rhythms.

METHODS

The dataset consisted of 1631 3-s ECG segments with clinical rhythm annotations, obtained from 298 out-of-hospital cardiac arrest patients. In total, 47 wavelet- and time-domain-based features were computed from the ECG. Features were selected using a wrapper-based feature selection architecture. Classifiers based on Bayesian decision theory, k-nearest neighbor, k-local hyperplane distance nearest neighbor, artificial neural network (ANN), and ensemble of decision trees were studied.

RESULTS

The best results were obtained for ANN classifier with Bayesian regularization backpropagation training algorithm with 14 features, which forms the proposed algorithm. The overall accuracy for the proposed algorithm was 78.5%. The sensitivities (and positive-predictive-values) for AS, PEA, PR, VF, and VT were 88.7% (91.0%), 68.9% (70.4%), 65.9% (69.0%), 86.2% (83.8%), and 78.8% (72.9%), respectively.

CONCLUSIONS

The results demonstrate that it is possible to classify resuscitation cardiac rhythms automatically, but the accuracy for the organized rhythms (PEA and PR) is low.

SIGNIFICANCE

We have made an important step toward making classification of resuscitation rhythms more efficient in the sense of minimal feedback from human experts.

Short ECG segments predict defibrillation outcome using quantitative waveform measures

AIM

Quantitative waveform measures of the ventricular fibrillation (VF) electrocardiogram (ECG) predict defibrillation outcome. Calculation requires an ECG epoch without chest compression artifact. However, pauses in CPR can adversely affect survival. Thus the potential use of waveform measures is limited by the need to pause CPR. We sought to characterize the relationship between the length of the CPR-free epoch and the ability to predict outcome.

METHODS

We conducted a retrospective investigation using the CPR-free ECG prior to first shock among out-of-hospital VF cardiac arrest patients in a large metropolitan region (n=442). Amplitude Spectrum Area (AMSA) and Median Slope (MS) were calculated using ECG epochs ranging from 5s to 0.2s. The relative ability of the measures to predict return of organized rhythm (ROR) and neurologically-intact survival was evaluated at different epoch lengths by calculating the area under the receiver operating characteristic curve (AUC) using the 5-s epoch as the referent group.

RESULTS

Compared to the 5-s epoch, AMSA performance declined significantly only after reducing epoch length to 0.2s for ROR (AUC 0.77-0.74, p=0.03) and with epochs of ≤0.6s for neurologically-intact survival (AUC 0.72-0.70, p=0.04). MS performance declined significantly with epochs of ≤0.8s for ROR (AUC 0.78-0.77, p=0.04) and with epochs ≤1.6s for neurologically-intact survival (AUC 0.72-0.71, p=0.04).

CONCLUSION

Waveform measures predict defibrillation outcome using very brief ECG epochs, a quality that may enable their use in current resuscitation algorithms designed to limit CPR interruption.

An Enhanced Adaptive Filtering Method for Suppressing Cardiopulmonary Resuscitation Artifact

Cardiopulmonary resuscitation (CPR) must be interrupted for reliable rhythm analysis in current automatic external defibrillators because of artifacts produced by chest compressions. However, interruptions in CPR adversely affect the restoration of spontaneous circulation and survival. Suppressing CPR artifacts by digital signal processing techniques is a promising method to enable rhythm analysis during chest compressions, which would eliminate CPR interruptions for rhythm analysis. Although numerous methods have been developed to suppress CPR artifacts, the accuracy of rhythm analysis is still inadequate due to the residual artifact components in the filtered signal. This study proposes an enhanced adaptive filtering method to suppress CPR artifacts. A total of 183 shockable and 453 nonshockable segments of ECG signal, together with CPR-related reference signal, were extracted from 233 out of hospital cardiac arrest patients. The method was optimized on a training set with 85 shockable and 211 nonshockable segments, and evaluated on a testing set with 98 shockable and 242 nonshockable segments. Compared with artifact corrupted ECG signals, the signal-to-noise ratio (SNR) increased from -9.8 ± 12.5 to 11.2 ± 11.8 dB, and the accuracy was improved from 74.1% to 92.0% after filtering with the proposed method. Compared with the traditional adaptive filter, the SNR was improved by 1.7 dB and the accuracy was improved by 5.6 points. These results indicated that the proposed method could effectively suppress the chest compression related artifacts and improve the accuracy of rhythm analysis during uninterrupted CPR.

Combining Amplitude Spectrum Area with Previous Shock Information Using Neural Networks Improves Prediction Performance of Defibrillation Outcome for Subsequent Shocks in Out-Of-Hospital Cardiac Arrest Patients

Objective

Quantitative ventricular fibrillation (VF) waveform analysis is a potentially powerful tool to optimize defibrillation. However, whether combining VF features with additional attributes that related to the previous shock could enhance the prediction performance for subsequent shocks is still uncertain.

Methods

A total of 528 defibrillation shocks from 199 patients experienced out-of-hospital cardiac arrest were analyzed in this study. VF waveform was quantified using amplitude spectrum area (AMSA) from defibrillator's ECG recordings prior to each shock. Combinations of AMSA with previous shock index (PSI) or/and change of AMSA (ΔAMSA) between successive shocks were exercised through a training dataset including 255shocks from 99patientswith neural networks. Performance of the combination methods were compared with AMSA based single feature prediction by area under receiver operating characteristic curve(AUC), sensitivity, positive predictive value (PPV), negative predictive value (NPV) and prediction accuracy (PA) through a validation dataset that was consisted of 273 shocks from 100patients.

Results

A total of61 (61.0%) patients required subsequent shocks (N = 173) in the validation dataset. Combining AMSA with PSI and ΔAMSA obtained highest AUC (0.904 vs. 0.819, p<0.001) among different combination approaches for subsequent shocks. Sensitivity (76.5% vs. 35.3%, p<0.001), NPV (90.2% vs. 76.9%, p = 0.007) and PA (86.1% vs. 74.0%, p = 0.005)were greatly improved compared with AMSA based single feature prediction with a threshold of 90% specificity.

Conclusion

In this retrospective study, combining AMSA with previous shock information using neural networks greatly improves prediction performance of defibrillation outcome for subsequent shocks.