Machine learning model to predict evolution of pulseless electrical activity during in-hospital cardiac arrest.

Background: During pulseless electrical activity (PEA) the cardiac mechanical and electrical functions are dissociated, a phenomenon occurring in 25-42% of in-hospital cardiac arrest (IHCA) cases. Accurate evaluation of the likelihood of a PEA patient transitioning to return of spontaneous circulation (ROSC) may be vital for the successful resuscitation. The aim: We sought to develop a model to automatically discriminate between PEA rhythms with favorable and unfavorable evolution to ROSC. Methods: A dataset of 190 patients, 120 with ROSC, were acquired with defibrillators from different vendors in three hospitals. The ECG and the transthoracic impedance (TTI) signal were processed to compute 16 waveform features. Logistic regression models where designed integrating both automated features and characteristics annotated in the QRS to identify PEAs with better prognosis leading to ROSC. Cross validation techniques were applied, both patient-specific and stratified, to evaluate the performance of the algorithm. Results: The best model consisted in a three feature algorithm that exhibited median (interquartile range) Area Under the Curve/Balanced accuracy/Sensitivity/Specificity of 80.3(9.9)/75.6(8.0)/ 77.4(15.2)/72.3(16.4) %, respectively. Conclusions: Information hidden in the waveforms of the ECG and TTI signals, along with QRS complex features, can predict the progression of PEA. Automated methods as the one proposed in this study, could contribute to assist in the targeted treatment of PEA in IHCA.

Cardiac output estimation using ballistocardiography: a feasibility study in healthy subjects.

There is no reliable automated non-invasive solution for monitoring circulation and guiding treatment in prehospital emergency medicine. Cardiac output (CO) monitoring might provide a solution, but CO monitors are not feasible/practical in the prehospital setting. Non-invasive ballistocardiography (BCG) measures heart contractility and tracks CO changes. This study analyzed the feasibility of estimating CO using morphological features extracted from BCG signals. In 20 healthy subjects ECG, carotid/abdominal BCG, and invasive arterial blood pressure based CO were recorded. BCG signals were adaptively processed to isolate the circulatory component from carotid (CCc) and abdominal (CCa) BCG. Then, 66 features were computed on a beat-to-beat basis to characterize amplitude/duration/area/length of the fluctuation in CCc and CCa. Subjects' data were split into development set (75%) to select the best feature subset with which to build a machine learning model to estimate CO and validation set (25%) to evaluate model's performance. The model showed a mean absolute error, percentage error and 95% limits of agreement of 0.83 L/min, 30.2% and - 2.18-1.89 L/min respectively in the validation set. BCG showed potential to reliably estimate/track CO. This method is a promising first step towards an automated, non-invasive and reliable CO estimator that may be tested in prehospital emergencies.

Heart murmur detection from phonocardiogram recordings: The George B. Moody PhysioNet Challenge 2022.

Cardiac auscultation is an accessible diagnostic screening tool that can help to identify patients with heart murmurs, who may need follow-up diagnostic screening and treatment for abnormal cardiac function. However, experts are needed to interpret the heart sounds, limiting the accessibility of cardiac auscultation in resource-constrained environments. Therefore, the George B. Moody PhysioNet Challenge 2022 invited teams to develop algorithmic approaches for detecting heart murmurs and abnormal cardiac function from phonocardiogram (PCG) recordings of heart sounds. For the Challenge, we sourced 5272 PCG recordings from 1452 primarily pediatric patients in rural Brazil, and we invited teams to implement diagnostic screening algorithms for detecting heart murmurs and abnormal cardiac function from the recordings. We required the participants to submit the complete training and inference code for their algorithms, improving the transparency, reproducibility, and utility of their work. We also devised an evaluation metric that considered the costs of screening, diagnosis, misdiagnosis, and treatment, allowing us to investigate the benefits of algorithmic diagnostic screening and facilitate the development of more clinically relevant algorithms. We received 779 algorithms from 87 teams during the Challenge, resulting in 53 working codebases for detecting heart murmurs and abnormal cardiac function from PCG recordings. These algorithms represent a diversity of approaches from both academia and industry, including methods that use more traditional machine learning techniques with engineered clinical and statistical features as well as methods that rely primarily on deep learning models to discover informative features. The use of heart sound recordings for identifying heart murmurs and abnormal cardiac function allowed us to explore the potential of algorithmic approaches for providing more accessible diagnostic screening in resource-constrained environments. The submission of working, open-source algorithms and the use of novel evaluation metrics supported the reproducibility, generalizability, and clinical relevance of the research from the Challenge.

Beyond Heart Murmur Detection: Automatic Murmur Grading From Phonocardiogram.

OBJECTIVE: Murmurs are abnormal heart sounds, identified by experts through cardiac auscultation. The murmur grade, a quantitative measure of the murmur intensity, is strongly correlated with the patient's clinical condition. This work aims to estimate each patient's murmur grade (i.e., absent, soft, loud) from multiple auscultation location phonocardiograms (PCGs) of a large population of pediatric patients from a low-resource rural area. METHODS: The Mel spectrogram representation of each PCG recording is given to an ensemble of 15 convolutional residual neural networks with channel-wise attention mechanisms to classify each PCG recording. The final murmur grade for each patient is derived based on the proposed decision rule and considering all estimated labels for available recordings. The proposed method is cross-validated on a dataset consisting of 3456 PCG recordings from 1007 patients using a stratified ten-fold cross-validation. Additionally, the method was tested on a hidden test set comprised of 1538 PCG recordings from 442 patients. RESULTS: The overall cross-validation performances for patient-level murmur gradings are 86.3% and 81.6% in terms of the unweighted average of sensitivities and F1-scores, respectively. The sensitivities (and F1-scores) for absent, soft, and loud murmurs are 90.7% (93.6%), 75.8% (66.8%), and 92.3% (84.2%), respectively. On the test set, the algorithm achieves an unweighted average of sensitivities of 80.4% and an F1-score of 75.8%. CONCLUSIONS: This study provides a potential approach for algorithmic pre-screening in low-resource settings with relatively high expert screening costs. SIGNIFICANCE: The proposed method represents a significant step beyond detection of murmurs, providing characterization of intensity, which may provide an enhanced classification of clinical outcomes.

Issues in the automated classification of multilead ecgs using heterogeneous labels and populations.

Objective.The standard twelve-lead electrocardiogram (ECG) is a widely used tool for monitoring cardiac function and diagnosing cardiac disorders. The development of smaller, lower-cost, and easier-to-use ECG devices may improve access to cardiac care in lower-resource environments, but the diagnostic potential of these devices is unclear. This work explores these issues through a public competition: the 2021 PhysioNet Challenge. In addition, we explore the potential for performance boosting through a meta-learning approach.Approach.We sourced 131,149 twelve-lead ECG recordings from ten international sources. We posted 88,253 annotated recordings as public training data and withheld the remaining recordings as hidden validation and test data. We challenged teams to submit containerized, open-source algorithms for diagnosing cardiac abnormalities using various ECG lead combinations, including the code for training their algorithms. We designed and scored the algorithms using an evaluation metric that captures the risks of different misdiagnoses for 30 conditions. After the Challenge, we implemented a semi-consensus voting model on all working algorithms.Main results.A total of 68 teams submitted 1,056 algorithms during the Challenge, providing a variety of automated approaches from both academia and industry. The performance differences across the different lead combinations were smaller than the performance differences across the different test databases, showing that generalizability posed a larger challenge to the algorithms than the choice of ECG leads. A voting model improved performance by 3.5%.Significance.The use of different ECG lead combinations allowed us to assess the diagnostic potential of reduced-lead ECG recordings, and the use of different data sources allowed us to assess the generalizability of the algorithms to diverse institutions and populations. The submission of working, open-source code for both training and testing and the use of a novel evaluation metric improved the reproducibility, generalizability, and applicability of the research conducted during the Challenge.

The CirCor DigiScope Dataset: From Murmur Detection to Murmur Classification.

Cardiac auscultation is one of the most cost-effective techniques used to detect and identify many heart conditions. Computer-assisted decision systems based on auscultation can support physicians in their decisions. Unfortunately, the application of such systems in clinical trials is still minimal since most of them only aim to detect the presence of extra or abnormal waves in the phonocardiogram signal, i.e., only a binary ground truth variable (normal vs abnormal) is provided. This is mainly due to the lack of large publicly available datasets, where a more detailed description of such abnormal waves (e.g., cardiac murmurs) exists. To pave the way to more effective research on healthcare recommendation systems based on auscultation, our team has prepared the currently largest pediatric heart sound dataset. A total of 5282 recordings have been collected from the four main auscultation locations of 1568 patients, in the process, 215780 heart sounds have been manually annotated. Furthermore, and for the first time, each cardiac murmur has been manually annotated by an expert annotator according to its timing, shape, pitch, grading, and quality. In addition, the auscultation locations where the murmur is present were identified as well as the auscultation location where the murmur is detected more intensively. Such detailed description for a relatively large number of heart sounds may pave the way for new machine learning algorithms with a real-world application for the detection and analysis of murmur waves for diagnostic purposes.

A Machine Learning Model for the Prognosis of Pulseless Electrical Activity during Out-of-Hospital Cardiac Arrest.

Pulseless electrical activity (PEA) is characterized by the disassociation of the mechanical and electrical activity of the heart and appears as the initial rhythm in 20-30% of out-of-hospital cardiac arrest (OHCA) cases. Predicting whether a patient in PEA will convert to return of spontaneous circulation (ROSC) is important because different therapeutic strategies are needed depending on the type of PEA. The aim of this study was to develop a machine learning model to differentiate PEA with unfavorable (unPEA) and favorable (faPEA) evolution to ROSC. An OHCA dataset of 1921 5s PEA signal segments from defibrillator files was used, 703 faPEA segments from 107 patients with ROSC and 1218 unPEA segments from 153 patients with no ROSC. The solution consisted of a signal-processing stage of the ECG and the thoracic impedance (TI) and the extraction of the TI circulation component (ICC), which is associated with ventricular wall movement. Then, a set of 17 features was obtained from the ECG and ICC signals, and a random forest classifier was used to differentiate faPEA from unPEA. All models were trained and tested using patientwise and stratified 10-fold cross-validation partitions. The best model showed a median (interquartile range) area under the curve (AUC) of 85.7(9.8)% and a balance accuracy of 78.8(9.8)%, improving the previously available solutions at more than four points in the AUC and three points in balanced accuracy. It was demonstrated that the evolution of PEA can be predicted using the ECG and TI signals, opening the possibility of targeted PEA treatment in OHCA.

Multimodal Algorithms for the Classification of Circulation States During Out-of-Hospital Cardiac Arrest.

GOAL: Identifying the circulation state during out-of-hospital cardiac arrest (OHCA) is essential to determine what life-saving therapies to apply. Currently algorithms discriminate circulation (pulsed rhythms, PR) from no circulation (pulseless electrical activity, PEA), but PEA can be classified into true (TPEA) and pseudo (PPEA) depending on cardiac contractility. This study introduces multi-class algorithms to automatically determine circulation states during OHCA using the signals available in defibrillators. METHODS: A cohort of 60 OHCA cases were used to extract a dataset of 2506 5-s segments, labeled as PR (1463), PPEA (364) and TPEA (679) using the invasive blood pressure, experimentally recorded through a radial/femoral cannulation. A multimodal algorithm using features obtained from the electrocardiogram, the thoracic impedance and the capnogram was designed. A random forest model was trained to discriminate three (TPEA/PPEA/PR) and two (PEA/PR) circulation states. The models were evaluated using repeated patient-wise 5-fold cross-validation, with the unweighted mean of sensitivities (UMS) and F 1-score as performance metrics. RESULTS: The best model for 3-class had a median (interquartile range, IQR) UMS and F 1 of 69.0% (68.0-70.1) and 61.7% (61.0-62.5), respectively. The best two class classifier had median (IQR) UMS and F 1 of 83.9% (82.9-84.5) and 76.2% (75.0-76.9), outperforming all previous proposals in over 3-points in UMS. CONCLUSIONS: The first multiclass OHCA circulation state classifier was demonstrated. The method improved previous algorithms for binary pulse/no-pulse decisions. SIGNIFICANCE: Automatic multiclass circulation state classification during OHCA could contribute to improve cardiac arrest therapy and improve survival rates.

Classification of 12-lead ECGs: the PhysioNet/Computing in Cardiology Challenge 2020.

OBJECTIVE: Vast 12-lead ECGs repositories provide opportunities to develop new machine learning approaches for creating accurate and automatic diagnostic systems for cardiac abnormalities. However, most 12-lead ECG classification studies are trained, tested, or developed in single, small, or relatively homogeneous datasets. In addition, most algorithms focus on identifying small numbers of cardiac arrhythmias that do not represent the complexity and difficulty of ECG interpretation. This work addresses these issues by providing a standard, multi-institutional database and a novel scoring metric through a public competition: the PhysioNet/Computing in Cardiology Challenge 2020. APPROACH: A total of 66 361 12-lead ECG recordings were sourced from six hospital systems from four countries across three continents; 43 101 recordings were posted publicly with a focus on 27 diagnoses. For the first time in a public competition, we required teams to publish open-source code for both training and testing their algorithms, ensuring full scientific reproducibility. MAIN RESULTS: A total of 217 teams submitted 1395 algorithms during the Challenge, representing a diversity of approaches for identifying cardiac abnormalities from both academia and industry. As with previous Challenges, high-performing algorithms exhibited significant drops ([Formula: see text]10%) in performance on the hidden test data. SIGNIFICANCE: Data from diverse institutions allowed us to assess algorithmic generalizability. A novel evaluation metric considered different misclassification errors for different cardiac abnormalities, capturing the outcomes and risks of different diagnoses. Requiring both trained models and code for training models improved the generalizability of submissions, setting a new bar in reproducibility for public data science competitions.

Towards the Prediction of Rearrest during Out-of-Hospital Cardiac Arrest.

A secondary arrest is frequent in patients that recover spontaneous circulation after an out-of-hospital cardiac arrest (OHCA). Rearrest events are associated to worse patient outcomes, but little is known on the heart dynamics that lead to rearrest. The prediction of rearrest could help improve OHCA patient outcomes. The aim of this study was to develop a machine learning model to predict rearrest. A random forest classifier based on 21 heart rate variability (HRV) and electrocardiogram (ECG) features was designed. An analysis interval of 2 min after recovery of spontaneous circulation was used to compute the features. The model was trained and tested using a repeated cross-validation procedure, on a cohort of 162 OHCA patients (55 with rearrest). The median (interquartile range) sensitivity (rearrest) and specificity (no-rearrest) of the model were 67.3% (9.1%) and 67.3% (10.3%), respectively, with median areas under the receiver operating characteristics and the precision-recall curves of 0.69 and 0.53, respectively. This is the first machine learning model to predict rearrest, and would provide clinically valuable information to the clinician in an automated way.

Capnography: A support tool for the detection of return of spontaneous circulation in out-of-hospital cardiac arrest.

BACKGROUND: Automated detection of return of spontaneous circulation (ROSC) is still an unsolved problem during cardiac arrest. Current guidelines recommend the use of capnography, but most automatic methods are based on the analysis of the ECG and thoracic impedance (TI) signals. This study analysed the added value of EtCO2 for discriminating pulsed (PR) and pulseless (PEA) rhythms and its potential to detect ROSC. MATERIALS AND METHODS: A total of 426 out-of-hospital cardiac arrest cases, 117 with ROSC and 309 without ROSC, were analysed. First, EtCO2 values were compared for ROSC and no ROSC cases. Second, 5098 artefact free 3-s long segments were automatically extracted and labelled as PR (3639) or PEA (1459) using the instant of ROSC annotated by the clinician on scene as gold standard. Machine learning classifiers were designed using features obtained from the ECG, TI and the EtCO2 value. Third, the cases were retrospectively analysed using the classifier to discriminate cases with and without ROSC. RESULTS: EtCO2 values increased significantly from 41 mmHg 3-min before ROSC to 57 mmHg 1-min after ROSC, and EtCO2 was significantly larger for PR than for PEA, 46 mmHg/20 mmHg (p < 0.05). Adding EtCO2 to the machine learning models increased their area under the curve (AUC) by over 2 percentage points. The combination of ECG, TI and EtCO2 had an AUC for the detection of pulse of 0.92. Finally, the retrospective analysis showed a sensitivity and specificity of 96.6% and 94.5% for the detection of ROSC and no-ROSC cases, respectively. CONCLUSION: Adding EtCO2 improves the performance of automatic algorithms for pulse detection based on ECG and TI. These algorithms can be used to identify pulse on site, and to retrospectively identify cases with ROSC.

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.

ECG-based pulse detection during cardiac arrest using random forest classifier.

Sudden cardiac arrest is one of the leading causes of death in the industrialized world. Pulse detection is essential for the recognition of the arrest and the recognition of return of spontaneous circulation during therapy, and it is therefore crucial for the survival of the patient. This paper introduces the first method based exclusively on the ECG for the automatic detection of pulse during cardiopulmonary resuscitation. Random forest classifier is used to efficiently combine up to nine features from the time, frequency, slope, and regularity analysis of the ECG. Data from 191 cardiac arrest patients was used, and 1177 ECG segments were processed, 796 with pulse and 381 without pulse. A leave-one-patient out cross validation approach was used to train and test the algorithm. The statistical distributions of sensitivity (SE) and specificity (SP) for pulse detection were estimated using 500 patient-wise bootstrap partitions. The mean (std) SE/SP for nine-feature classifier was 88.4 (1.8) %/89.7 (1.4) %, respectively. The designed algorithm only requires 4-s-long ECG segments and could be integrated in any commercial automated external defibrillator. The method permits to detect the presence of pulse accurately, minimizing interruptions in cardiopulmonary resuscitation therapy, and could contribute to improve survival from cardiac arrest.

A Robust Machine Learning Architecture for a Reliable ECG Rhythm Analysis during CPR.

Chest compressions delivered during cardiopulmonary resuscitation (CPR) induce artifacts in the ECG that may make the shock advice algorithms (SAA) of defibrillators inaccurate. There is evidence that methods consisting of adaptive filters that remove the CPR artifact followed by machine learning (ML) based algorithms are able to make reliable shock/no-shock decisions during compressions. However, there is room for improvement in the performance of these methods. The objective was to design a robust ML framework for a reliable shock/no-shock decision during CPR. The study dataset contained 596 shockable and 1697 nonshockable ECG segments obtained from 273 cases of out-of-hospital cardiac arrest. Shock/no-shock labels were adjudicated by expert reviewers using ECG intervals without artifacts. First, CPR artifacts were removed from the ECG using a Least Mean Squares (LMS) filter. Then, 38 shock/no-shock decision features based on the Stationary Wavelet Transform (SWT) were extracted from the filtered ECG. A wapper-based feature selection method was applied to select the 6 best features for classification. Finally, 4 state-of-the-art ML classifiers were tested to make the shock/no-shock decision. These diagnoses were compared with the rhythm annotations to compute the Sensitivity (Se) and Specificity (Sp). All classifiers achieved an Se above 94.5%, Sp above 95.5% and an accuracy around 96.0%. They all exceeded the 90% Se and 95% Sp minimum values recommended by the American Heart Association.

Convolutional Recurrent Neural Networks to Characterize the Circulation Component in the Thoracic Impedance during Out-of-Hospital Cardiac Arrest.

Pulse detection during out-of-hospital cardiac arrest remains challenging for both novel and expert rescuers because current methods are inaccurate and time-consuming. There is still a need to develop automatic methods for pulse detection, where the most challenging scenario is the discrimination between pulsed rhythms (PR, pulse) and pulseless electrical activity (PEA, no pulse). Thoracic impedance (TI) acquired through defibrillation pads has been proven useful for detecting pulse as it shows small fluctuations with every heart beat. In this study we analyse the use of deep learning techniques to detect pulse using only the TI signal. The proposed neural network, composed by convolutional and recurrent layers, outperformed state of the art methods, and achieved a balanced accuracy of 90% for segments as short as 3 s.

Deep Neural Networks for ECG-Based Pulse Detection during Out-of-Hospital Cardiac Arrest.

The automatic detection of pulse during out-of-hospital cardiac arrest (OHCA) is necessary for the early recognition of the arrest and the detection of return of spontaneous circulation (end of the arrest). The only signal available in every single defibrillator and valid for the detection of pulse is the electrocardiogram (ECG). In this study we propose two deep neural network (DNN) architectures to detect pulse using short ECG segments (5 s), i.e., to classify the rhythm into pulseless electrical activity (PEA) or pulse-generating rhythm (PR). A total of 3914 5-s ECG segments, 2372 PR and 1542 PEA, were extracted from 279 OHCA episodes. Data were partitioned patient-wise into training (80%) and test (20%) sets. The first DNN architecture was a fully convolutional neural network, and the second architecture added a recurrent layer to learn temporal dependencies. Both DNN architectures were tuned using Bayesian optimization, and the results for the test set were compared to state-of-the art PR/PEA discrimination algorithms based on machine learning and hand crafted features. The PR/PEA classifiers were evaluated in terms of sensitivity (Se) for PR, specificity (Sp) for PEA, and the balanced accuracy (BAC), the average of Se and Sp. The Se/Sp/BAC of the DNN architectures were 94.1%/92.9%/93.5% for the first one, and 95.5%/91.6%/93.5% for the second one. Both architectures improved the performance of state of the art methods by more than 1.5 points in BAC.

A novel technique to assess the quality of ventilation during pre-hospital cardiopulmonary resuscitation.

BACKGROUND: Devices that measure ventilation in the pre-hospital setting are deficient especially during early cardiopulmonary resuscitation (CPR) before placement of an advanced airway. Consequently, evidence is limited regarding the role of ventilation during early CPR and its effect on outcomes. OBJECTIVE: To develop software that automatically identifies ventilation waveforms recorded by defibrillators based on changes in transthoracic impedance during standard CPR. METHODS: This was an observational, retrospective analysis of non-traumatic pre-hospital cardiac arrest patients who received 30:2 CPR by emergency medical service rescuers. Data was collected from 550 cases recorded by the bioimpedance channel of defibrillators. Two expert clinicians independently assessed all episodes from the time of initial CPR until placement of an advanced airway, defined acceptable ventilation waveforms, and annotated the pauses between compressions with ventilation waveforms. We then developed software that incorporated the expert criteria and automatically annotated pauses with acceptable ventilations. RESULTS: A total of 7396 pauses were analyzed, mean(SD) duration of 30:2 CPR was 13 (8) min, with 13 (10) pauses/patient, and mean pause duration of 6 (3) s. Reviewer 1 and reviewer 2 identified 2375 and 2249 pauses with any acceptable ventilation, respectively, with an inter-rater reliability of 0.94. The novel software program reproduced expert annotation with excellent agreement (>0.8) and high accuracy, both sensitivity and specificity above 90%, compared to two reviewers. The software presented a substantial agreement with the reviewers (κ > 0.73) for ventilation counts in the pauses. CONCLUSION: We developed a novel and reliable strategy that enables investigation of ventilation quality during standard CPR using thoracic bioimpedance. This strategy would allow a timely and reliable automatic annotation of large scale resuscitation datasets.

Feasibility of the capnogram to monitor ventilation rate during cardiopulmonary resuscitation.

AIM: The rates of chest compressions (CCs) and ventilations are both important metrics to monitor the quality of cardiopulmonary resuscitation (CPR). Capnography permits monitoring ventilation, but the CCs provided during CPR corrupt the capnogram and compromise the accuracy of automatic ventilation detectors. The aim of this study was to evaluate the feasibility of an automatic algorithm based on the capnogram to detect ventilations and provide feedback on ventilation rate during CPR, specifically addressing intervals where CCs are delivered. METHODS: The dataset used to develop and test the algorithm contained in-hospital and out-of-hospital cardiac arrest episodes. The method relies on adaptive thresholding to detect ventilations in the first derivative of the capnogram. The performance of the detector was reported in terms of sensitivity (SE) and Positive Predictive Value (PPV). The overall performance was reported in terms of the rate error and errors in the hyperventilation alarms. Results were given separately for the intervals with CCs. RESULTS: A total of 83 episodes were considered, resulting in 4880min and 46,740 ventilations (8741 during CCs). The method showed an overall SE/PPV above 99% and 97% respectively, even in intervals with CCs. The error for the ventilation rate was below 1.8min-1 in any group, and >99% of the ventilation alarms were correctly detected. CONCLUSION: A method to provide accurate feedback on ventilation rate using only the capnogram is proposed. Its accuracy was proven even in intervals where canpography signal was severely corrupted by CCs. This algorithm could be integrated into monitor/defibrillators to provide reliable feedback on ventilation rate during CPR.