Airway Strategy and Ventilation Rates in the Pragmatic Airway Resuscitation Trial

BACKGROUND

We sought to describe ventilation rates during out-of-hospital cardiac arrest (OHCA) resuscitation and their associations with airway management strategy and outcomes.

METHODS

We analyzed continuous end-tidal carbon dioxide capnography data from adult OHCA enrolled in the Pragmatic Airway Resuscitation Trial (PART). Using automated signal processing techniques, we determined continuous ventilation rate for consecutive 10-second epochs after airway insertion. We defined hypoventilation as a ventilation rate <6 breaths/min. We defined hyperventilation as a ventilation rate >12 breaths/min. We compared differences in total and percentage post-airway hyper- and hypoventilation between airway interventions (laryngeal tube (LT) vs. endotracheal intubation (ETI). We also determined associations between hypo-/hyperventilation and OHCA outcomes (ROSC, 72-hour survival, hospital survival, hospital survival with favorable neurologic status).

RESULTS

Adequate post-airway capnography were available for 1,010 (LT n=714, ETI n=296) of 3,004 patients. Median ventilation rates were: LT 8.0 (IQR 6.5-9.6) breaths/min, ETI 7.9 (6.5-9.7) breaths/min. Total duration and percentage of post-airway time with hypoventilation were similar between LT and ETI: median 1.8 vs. 1.7 minutes, p=0.94; median 10.5% vs. 11.5%, p=0.60. Total duration and percentage of post-airway time with hyperventilation were similar between LT and ETI: median 0.4 vs. 0.4 minutes, p=0.91; median 2.1% vs. 1.9%, p=0.99. Hypo- and hyperventilation exhibited limited associations with OHCA outcomes.

CONCLUSION

In the PART Trial, EMS personnel delivered post-airway ventilations at rates satisfying international guidelines, with only limited hypo- or hyperventilation. Hypo- and hyperventilation durations did not differ between airway management strategy and exhibited uncertain associations with OCHA outcomes.

Analysis of Few-Shot Techniques for Fungal Plant Disease Classification and Evaluation of Clustering Capabilities Over Real Datasets

Plant fungal diseases are one of the most important causes of crop yield losses. Therefore, plant disease identification algorithms have been seen as a useful tool to detect them at early stages to mitigate their effects. Although deep-learning based algorithms can achieve high detection accuracies, they require large and manually annotated image datasets that is not always accessible, specially for rare and new diseases. This study focuses on the development of a plant disease detection algorithm and strategy requiring few plant images (Few-shot learning algorithm). We extend previous work by using a novel challenging dataset containing more than 100,000 images. This dataset includes images of leaves, panicles and stems of five different crops (barley, corn, rape seed, rice, and wheat) for a total of 17 different diseases, where each disease is shown at different disease stages. In this study, we propose a deep metric learning based method to extract latent space representations from plant diseases with just few images by means of a Siamese network and triplet loss function. This enhances previous methods that require a support dataset containing a high number of annotated images to perform metric learning and few-shot classification. The proposed method was compared over a traditional network that was trained with the cross-entropy loss function. Exhaustive experiments have been performed for validating and measuring the benefits of metric learning techniques over classical methods. Results show that the features extracted by the metric learning based approach present better discriminative and clustering properties. Davis-Bouldin index and Silhouette score values have shown that triplet loss network improves the clustering properties with respect to the categorical-cross entropy loss. Overall, triplet loss approach improves the DB index value by 22.7% and Silhouette score value by 166.7% compared to the categorical cross-entropy loss model. Moreover, the F-score parameter obtained from the Siamese network with the triplet loss performs better than classical approaches when there are few images for training, obtaining a 6% improvement in the F-score mean value. Siamese networks with triplet loss have improved the ability to learn different plant diseases using few images of each class. These networks based on metric learning techniques improve clustering and classification results over traditional categorical cross-entropy loss networks for plant disease identification.

Novel application of thoracic impedance to characterize ventilations during cardiopulmonary resuscitation in the pragmatic airway resuscitation trial

BACKGROUND

Significant challenges exist in measuring ventilation quality during out-of-hospital cardiopulmonary arrest (OHCA) outcomes. Since ventilation is associated with outcomes in cardiac arrest, tools that objectively describe ventilation dynamics are needed. We sought to characterize thoracic impedance (TI) oscillations associated with ventilation waveforms in the Pragmatic Airway Resuscitation Trial (PART).

METHODS

We analyzed CPR process files collected from adult OHCA enrolled in PART. We limited the analysis to cases with simultaneous capnography ventilation recordings at the Dallas-Fort Worth site. We identified ventilation waveforms in the thoracic impedance signal by applying automated signal processing with adaptive filtering techniques to remove overlying artifacts from chest compressions. We correlated detected ventilations with the end-tidal capnography signals. We determined the amplitudes (Ai, Ae) and durations (Di, De) of both insufflation and exhalation phases. We compared differences between laryngeal tube (LT) and endotracheal intubation (ETI) airway management during mechanical or manual chest compressions using Mann-Whitney U-test.

RESULTS

We included 303 CPR process cases in the analysis; 209 manual (77 ETI, 132 LT), 94 mechanical (41 ETI, 53 LT). Ventilation Ai and Ae were higher for ETI than LT in both manual (ETI: Ai 0.71 Ω, Ae 0.70 Ω vs LT: Ai 0.46 Ω, Ae 0.45 Ω; p < 0.01 respectively) and mechanical chest compressions (ETI: Ai 1.22 Ω, Ae 1.14 Ω VS LT: Ai 0.74 Ω, Ae 0.68 Ω; p < 0.01 respectively). Ventilations per minute, duration of TI amplitude insufflation and exhalation did not differ among groups.

CONCLUSION

Compared with LT, ETI thoracic impedance ventilation insufflation and exhalation amplitude were higher while duration did not differ. TI may provide a novel approach to characterizing ventilation during OHCA.

Methodology and framework for the analysis of cardiopulmonary resuscitation quality in large and heterogeneous cardiac arrest datasets

BACKGROUND

Out-of-hospital cardiac arrest (OHCA) data debriefing and clinical research often require the retrospective analysis of large datasets containing defibrillator files from different vendors and clinical annotations by the emergency medical services.

AIM

To introduce and evaluate a methodology to automatically extract cardiopulmonary resuscitation (CPR) quality data in a uniform and systematic way from OHCA datasets from multiple heterogeneous sources.

METHODS

A dataset of 2236 OHCA cases from multiple defibrillator models and manufacturers was analyzed. Chest compressions were automatically identified using the thoracic impedance and compression depth signals. Device event time-stamps and clinical annotations were used to set the start and end of the analysis interval, and to identify periods with spontaneous circulation. A manual audit of the automatic annotations was conducted and used as gold standard. Chest compression fraction (CCF), rate (CCR) and interruption ratio were computed as CPR quality variables. The unsigned error between the automated procedure and the gold standard was calculated.

RESULTS

Full-episode median errors below 2% in CCF, 1 min^-1 in CCR, and 1.5% in interruption ratio, were measured for all signals and devices. The proportion of cases with large errors (>10% in CCF and interruption ratio, and >10 min^-1 in CCR) was below 10%. Errors were lower for shorter sub-intervals of interest, like the airway insertion interval.

CONCLUSIONS

An automated methodology was validated to accurately compute CPR metrics in large and heterogeneous OHCA datasets. Automated processing of defibrillator files and the associated clinical annotations enables the aggregation and analysis of CPR data from multiple sources.

Tidal volume measurements via transthoracic impedance waveform characteristics: The effect of age, body mass index and gender. A single centre interventional study

BACKGROUND AND AIM

Measuring tidal volumes (TV) during bag-valve ventilation is challenging in the clinical setting. The ventilation waveform amplitude of the transthoracic impedance (TTI-amplitude) correlates well with TV for an individual, but poorer between patients. We hypothesized that TV to TTI-amplitude relations could be improved when adjusted for morphometric variables like body mass index (BMI), gender or age, and that TTI-amplitude cut-offs for ventilations with adequate TV (>400ml) could be established.

MATERIALS AND METHODS

Twenty-one consenting adults (9 female, and 9 overall overweight) during positive pressure ventilation in anaesthesia before scheduled surgery were included. Seventeen ventilator modes were used (⩾ five breaths per mode) to adjust different TVs (150-800 ml), ventilation frequencies (10-30 min^-1) and insufflation times (0.5-3.5 s). TTI from the defibrillation pads was filtered to obtain ventilation TTI-amplitudes. Linear regression models were fitted between target and explanatory variables, and compared (coefficient of determination, R²).

RESULTS

The TV to TTI-amplitude slope was 1.39 Ω/l (R²=0.52), with significant differences (p<0.05) between male/female (1.04 Ω/l vs 1.84 Ω/l) and normal/overweight subjects (1.65 Ω/l vs 1.04 Ω/l). The median (interquartile range) TTI-amplitude cut-off for adequate TV was 0.51 Ω(0.14-1.20) with significant differences between males and females (0.58 Ω/0.39 Ω), and normal and overweight subjects (0.52 Ω/0.46 Ω). The TV to TTI-amplitude model improved (R²=0.66) when BMI, age and gender were included.

CONCLUSIONS

TTI-amplitude to TV relations were established and cut-offs for ventilations with adequate TV determined. Patient morphometric variables related to gender, age and BMI explain part of the variability in the measurements.

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.

Shock decision algorithm for use during load distributing band cardiopulmonary resuscitation

AIM

Chest compressions delivered by a load distributing band (LDB) induce artefacts in the electrocardiogram. These artefacts alter shock decisions in defibrillators. The aim of this study was to demonstrate the first reliable shock decision algorithm during LDB compressions.

METHODS

The study dataset comprised 5813 electrocardiogram segments from 896 cardiac arrest patients during LDB compressions. Electrocardiogram segments were annotated by consensus as shockable (1154, 303 patients) or nonshockable (4659, 841 patients). Segments during asystole were used to characterize the LDB artefact and to compare its characteristics to those of manual artefacts from other datasets. LDB artefacts were removed using adaptive filters. A machine learning algorithm was designed for the shock decision after filtering, and its performance was compared to that of a commercial defibrillator's algorithm.

RESULTS

Median (90% confidence interval) compression frequencies were lower and more stable for the LDB than for the manual artefact, 80 min^-1 (79.9-82.9) vs. 104.4 min^-1 (48.5-114.0). The amplitude and waveform regularity (Pearson's correlation coefficient) were larger for the LDB artefact, with 5.5 mV (0.8-23.4) vs. 0.5 mV (0.1-2.2) (p < 0.001) and 0.99 (0.78-1.0) vs. 0.88 (0.55-0.98) (p < 0.001). The shock decision accuracy was significantly higher for the machine learning algorithm than for the defibrillator algorithm, with sensitivity/specificity pairs of 92.1/96.8% (machine learning) vs. 91.4/87.1% (defibrillator) (p < 0.001).

CONCLUSION

Compared to other cardiopulmonary resuscitation artefacts, removing the LDB artefact was challenging due to larger amplitudes and lower compression frequencies. The machine learning algorithm achieved clinically reliable shock decisions during LDB compressions.

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.

Airway strategy and chest compression quality in the Pragmatic Airway Resuscitation Trial

BACKGROUND

Chest compression (CC) quality is associated with improved out-of-hospital cardiopulmonary arrest (OHCA) outcomes. Airway management efforts may adversely influence CC quality. We sought to compare the effects of initial laryngeal tube (LT) and initial endotracheal intubation (ETI) airway management strategies upon chest compression fraction (CCF), rate and interruptions in the Pragmatic Airway Resuscitation Trial (PART).

METHODS

We analyzed CPR process files collected from adult OHCA enrolled in PART. We used automated signal processing techniques and a graphical user interface to calculate CC quality measures and defined interruptions as pauses in chest compressions longer than 3 s. We determined CC fraction, rate and interruptions (number and total duration) for the entire resuscitation and compared differences between LT and ETI using t-tests. We repeated the analysis stratified by time before, during and after airway insertion as well as by successive 3-min time segments. We also compared CC quality between single vs. multiple airway insertion attempts, as well as between bag-valve-mask (BVM-only) vs. ETI or LT.

RESULTS

Of 3004 patients enrolled in PART, CPR process data were available for 1996 (1001 LT, 995 ETI). Mean CPR analysis duration were: LT 22.6 ± 10.8 min vs. ETI 25.3 ± 11.3 min (p < 0.001). Mean CC fraction (LT 88% vs. ETI 87%, p = 0.05) and rate (LT 114 vs. ETI 114 compressions per minute (cpm), p = 0.59) were similar between LT and ETI. Median number of CC interruptions were: LT 11 vs. ETI 12 (p = 0.001). Total CC interruption duration was lower for LT than ETI (LT 160 vs. ETI 181 s, p = 0.002); this difference was larger before airway insertion (LT 56 vs. ETI 78 s, p < 0.001). There were no differences in CC quality when stratified by 3-min time epochs.

CONCLUSION

In the PART trial, compared with ETI, LT was associated with shorter total CC interruption duration but not other CC quality measures. CC quality may be associated with OHCA airway management.

End-tidal carbon dioxide (ETCO2) and ventricular fibrillation amplitude spectral area (AMSA) for shock outcome prediction in out-of-hospital cardiac arrest. Are they two sides of the same coin?

AIM

Ventricular fibrillation amplitude spectral area (AMSA) and end-tidal carbon dioxide (ETCO₂) are predictors of shock success, understood as restoration of an organized rhythm, and return of spontaneous circulation (ROSC). However, little is known about their combined use. We aimed to assess the prediction accuracy when combined, and to clarify if they are correlated in out of hospital cardiac arrest' victims.

MATERIALS AND METHODS

Records acquired by external defibrillators in out-of-hospital cardiac arrest patients of the Lombardia Cardiac Arrest registry were processed. The 1-min pre-shock ETCO₂ median value (METCO₂) was computed from the capnogram and AMSA (2-48 mV.Hz range) computed applying the Fast Fourier Transform to a 2-second pre-shock filtered ECG interval (0.5-30 Hz). Support Vector Machine (SVM) predictive models based on METCO₂, AMSA and their combination were fit; results were given as the area under the curve (AUC) of the receiver operating characteristic (ROC) curves.

RESULTS

We considered 112 patients with 391 shocks delivered. METCO₂ and AMSA were predictors of shock success [AUC (IQR) of the ROC curve: 0.59 (0.56-0.62); 0.68 (0.65-0.72), respectively] and of ROSC [0.56 (0.53-0.59); 0.74 (0.71-0.78),]. Their combination in a SVM model increased the accuracy for predicting shock success [AUC (IQR) of the ROC curve: 0.71 (0.68-0.75)] and ROSC [0.77 (0.73-0.8)]. AMSA and METCO₂ were significantly correlated only in patients who achieved ROSC (rho = 0.33 p = 0.03).

CONCLUSIONS

AMSA and ETCO₂ predict shock success and ROSC after every shock, and their predictive power increases if combined. Notably, they were correlated only in patients who achieved ROSC.

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.

Factors affecting the course of resuscitation from cardiac arrest with pulseless electrical activity in children and adolescents

BACKGROUND

Although in-hospital pediatric cardiac arrests and cardiopulmonary resuscitation occur >15,000/year in the US, few studies have assessed which factors affect the course of resuscitation in these patients. We investigated transitions from Pulseless Electrical Activity (PEA) to Ventricular Fibrillation/pulseless Ventricular Tachycardia (VF/pVT), Return of Spontaneous Circulation (ROSC) and recurrences from ROSC to PEA in children and adolescents with in-hospital cardiac arrest.

METHODS

Episodes of cardiac arrest at the Children's Hospital of Philadelphia were prospectively registered. Defibrillators that recorded chest compression depth/rate and ventilation rate were applied. CPR variables, patient characteristics and etiology, and dynamic factors (e.g. the proportion of time spent in PEA or ROSC) were entered as time-varying covariates for the transition intensities under study.

RESULTS

In 67 episodes of CPR in 59 patients (median age 15 years) with cardiac arrest, there were 52 transitions from PEA to ROSC, 22 transitions from PEA to VF/pVT, and 23 recurrences of PEA from ROSC. Except for a nearly significant effect of mean compression depth beyond a threshold of 5.7 cm, only dynamic factors that evolved during CPR favored a transition from PEA to ROSC. The latter included a lower proportion of PEA over the last 5 min and a higher proportion of ROSC over the last 5 min. Factors associated with PEA to VF/pVT development were age, weight, the proportion spent in VF/pVT or PEA the last 5 min, and the general transition intensity, while PEA recurrence from ROSC only depended on the general transition intensity.

CONCLUSION

The clinical course during pediatric cardiac arrest was mainly influenced by dynamic factors associated with time in PEA and ROSC. Transitions from PEA to ROSC seemed to be favored by deeper compressions.

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.

Automatic Detection of Ventilations During Mechanical Cardiopulmonary Resuscitation

Feedback on chest compressions and ventilations during cardiopulmonary resuscitation (CPR) is important to improve survival from out-of-hospital cardiac arrest (OHCA). The thoracic impedance signal acquired by monitor-defibrillators during treatment can be used to provide feedback on ventilations, but chest compression components prevent accurate detection of ventilations. This study introduces the first method for accurate ventilation detection using the impedance while chest compressions are concurrently delivered by a mechanical CPR device. A total of 423 OHCA patients treated with mechanical CPR were included, 761 analysis intervals were selected which in total comprised 5 884 minutes and contained 34 864 ventilations. Ground truth ventilations were determined using the expired CO ₂ channel. The method uses adaptive signal processing to obtain the impedance ventilation waveform. Then, 14 features were calculated from the ventilation waveform and fed to a random forest (RF) classifier to discriminate false positive detections from actual ventilations. The RF feature importance was used to determine the best feature subset for the classifier. The method was trained and tested using stratified 10-fold cross validation (CV) partitions. The training/test process was repeated 20 times to statistically characterize the results. The best ventilation detector had a median (interdecile range, IDR) F ₁-score of 96.32 (96.26-96.37). When used to provide feedback in 1-min intervals, the median (IDR) error and relative error in ventilation rate were 0.002 (-0.334-0.572) min^-1 and 0.05 (-3.71-9.08)%, respectively. An accurate ventilation detector during mechanical CPR was demonstrated. The algorithm could be introduced in current equipment for feedback on ventilation rate and quality, and it could contribute to improve OHCA survival rates.

ECG-based Random Forest Classifier for Cardiac Arrest Rhythms

Rhythm annotation of out-of-hospital cardiac episodes (OHCA) is key for a better understanding of the interplay between resuscitation therapy and OHCA patient outcome. OHCA rhythms are classified in five categories, asystole (AS), pulseless electrical activity (PEA), pulsed rhythms (PR), ventricular fibrillation (VF) and ventricular tachycardia (VT). Manual OHCA annotation by expert clinicians is onerous and time consuming, so there is a need for accurate and automatic OHCA rhythm annotation methods. For this study 852 OHCA episodes of patients treated with Automated External Defibrillators (AED) by the Emergency Medical Services of the Basque Country were analyzed. Six expert clinicians reviewed the electrocardiogram (ECG) of 4214 AED rhythm analyses and annotated the rhythm. Their consensus decision was used as ground truth. There were a total of 2418 AS, 294 PR, 1008 PEA, 472 VF and 22 VT. The ECG analysis intervals were extracted and used to develop an automatic rhythm annotator. Data was partitioned patient-wise into training (70%) and test (30%). Performance was evaluated in terms of per class sensitivity (Se) and F-score (F1). The unweighted mean of sensitivity (UMS) and F-score were used as global performance metrics. The classification method is composed of a feature extraction and denoising stage based on the stationary wavelet transform of the ECG, and on a random forest classifier. The best model presented a per rhythm Se/F1 of 95.8/95.7, 43.3/52.2, 85.3/81.3, 94.2/96.1, 81.9/72.2 for AS, PR, PEA, VF and VT, respectively. The UMS for the test set was 80.2%, 2-points above that of previous solutions. This method could be used to retrospectively annotate large OHCA datasets and ameliorate the workload of manual OHCA rhythm annotation.

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.

Impedance Based Automatic Detection of Ventilations During Mechanical Cardiopulmonary Resuscitation

Monitoring ventilation rate is key to improve the quality of cardiopulmonary resuscitation (CPR) and increase the probability of survival in the event of an out-of-hospital cardiac arrest (OHCA). Ventilations produce discernible fluctuations in the thoracic impedance signal recorded by defibrillators. Impedance-based detection of ventilations during CPR is challenging due to chest compression artifacts. This study presents a method for an accurate detection of ventilations when chest compressions are delivered using a piston-driven mechanical device. Data from 223 OHCA patients were analyzed and 399 analysis segments totaling 3101 minutes of mechanical CPR were extracted. A total of 18327 ventilations were annotated using concurrent capnogram recordings. An adaptive least mean squares filter was used to remove compression artifacts. Potential ventilations were detected using a greedy peak detector, and the ventilation waveform was characterized using 8 waveform features. These features were used in a logistic regression classifier to discriminate true ventilations from false positives produced by the greedy peak detector. The classifier was trained and tested using patient wise 10-fold cross validation (CV), and 100 random CV partitions were created to statistically characterize the performance metrics. The peak detector presented a sensitivity (Se) of 99.30%, but a positive predictive value (PPV) of 54.43%. The best classifier configuration used 6 features and improved the mean (sd) Se and PPV of the detector to 93.20% (0.06) and 94.43% (0.04), respectively. When used to measure per minute ventilation rates for feedback to the rescuer, the mean (sd) absolute error in ventilation rate was 0.61 (1.64) min^-1. The first impedance-based method to accurately detect ventilations and give feedback on ventilation rate during mechanical CPR has been demonstrated.

Mixed convolutional and long short-term memory network for the detection of lethal ventricular arrhythmia

Early defibrillation by an automated external defibrillator (AED) is key for the survival of out-of-hospital cardiac arrest (OHCA) patients. ECG feature extraction and machine learning have been successfully used to detect ventricular fibrillation (VF) in AED shock decision algorithms. Recently, deep learning architectures based on 1D Convolutional Neural Networks (CNN) have been proposed for this task. This study introduces a deep learning architecture based on 1D-CNN layers and a Long Short-Term Memory (LSTM) network for the detection of VF. Two datasets were used, one from public repositories of Holter recordings captured at the onset of the arrhythmia, and a second from OHCA patients obtained minutes after the onset of the arrest. Data was partitioned patient-wise into training (80%) to design the classifiers, and test (20%) to report the results. The proposed architecture was compared to 1D-CNN only deep learners, and to a classical approach based on VF-detection features and a support vector machine (SVM) classifier. The algorithms were evaluated in terms of balanced accuracy (BAC), the unweighted mean of the sensitivity (Se) and specificity (Sp). The BAC, Se, and Sp of the architecture for 4-s ECG segments was 99.3%, 99.7%, and 98.9% for the public data, and 98.0%, 99.2%, and 96.7% for OHCA data. The proposed architecture outperformed all other classifiers by at least 0.3-points in BAC in the public data, and by 2.2-points in the OHCA data. The architecture met the 95% Sp and 90% Se requirements of the American Heart Association in both datasets for segment lengths as short as 3-s. This is, to the best of our knowledge, the most accurate VF detection algorithm to date, especially on OHCA data, and it would enable an accurate shock no shock diagnosis in a very short time.

Value of capnography to predict defibrillation success in out-of-hospital cardiac arrest

BACKGROUND AND AIM

Unsuccessful defibrillation shocks adversely affect survival from out-of-hospital cardiac arrest (OHCA). Ventricular fibrillation (VF) waveform analysis is the tool-of-choice for the non-invasive prediction of shock success, but surrogate markers of perfusion like end-tidal CO2 (EtCO2) could improve the prediction. The aim of this study was to evaluate EtCO2 as predictor of shock success, both individually and in combination with VF-waveform analysis.

MATERIALS AND METHODS

In total 514 shocks from 214 OHCA patients (75 first shocks) were analysed. For each shock three predictors of defibrillation success were automatically calculated from the device files: two VF-waveform features, amplitude spectrum area (AMSA) and fuzzy entropy (FuzzyEn), and the median EtCO2 (MEtCO2) in the minute before the shock. Sensitivity, specificity, receiver operating characteristic (ROC) curves and area under the curve (AUC) were calculated, for each predictor individually and for the combination of MEtCO2 and VF-waveform predictors. Separate analyses were done for first shocks and all shocks.

RESULTS

MEtCO2 in first shocks was significantly higher for successful than for unsuccessful shocks (31mmHg/25mmHg, p<0.05), but differences were not significant for all shocks (32mmHg/29mmHg, p>0.05). MEtCO2 predicted shock success with an AUC of 0.66 for first shocks, but was not a predictor for all shocks (AUC 0.54). AMSA and FuzzyEn presented AUCs of 0.76 and 0.77 for first shocks, and 0.75 and 0.75 for all shocks. For first shocks, adding MEtCO2 improved the AUC of AMSA and FuzzyEn to 0.79 and 0.83, respectively.

CONCLUSIONS

MEtCO2 predicted defibrillation success only for first shocks. Adding MEtCO2 to VF-waveform analysis in first shocks improved prediction of shock success. VF-waveform features and MEtCO2 were automatically calculated from the device files, so these methods could be introduced in current defibrillators adding only new software.

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.