Using the features of the time and volumetric capnogram for classification and prediction

Diagnosis and Severity Assessment of COPD Using a Novel Fast-Response Capnometer and Interpretable Machine Learning

INTRODUCTION

Spirometry is the gold standard for COPD diagnosis and severity determination, but is technique-dependent, nonspecific, and requires administration by a trained healthcare professional. There is a need for a fast, reliable, and precise alternative diagnostic test. This study's aim was to use interpretable machine learning to diagnose COPD and assess severity using 75-second carbon dioxide (CO2) breath records captured with TidalSense's N-TidalTM capnometer.

METHOD

For COPD diagnosis, machine learning algorithms were trained and evaluated on 294 COPD (including GOLD stages 1-4) and 705 non-COPD participants. A logistic regression model was also trained to distinguish GOLD 1 from GOLD 4 COPD with the output probability used as an index of severity.

RESULTS

The best diagnostic model achieved an AUROC of 0.890, sensitivity of 0.771, specificity of 0.850 and positive predictive value (PPV) of 0.834. Evaluating performance on all test capnograms that were confidently ruled in or out yielded PPV of 0.930 and NPV of 0.890. The severity determination model yielded an AUROC of 0.980, sensitivity of 0.958, specificity of 0.961 and PPV of 0.958 in distinguishing GOLD 1 from GOLD 4. Output probabilities from the severity determination model produced a correlation of 0.71 with percentage predicted FEV1.

CONCLUSION

The N-TidalTM device could be used alongside interpretable machine learning as an accurate, point-of-care diagnostic test for COPD, particularly in primary care as a rapid rule-in or rule-out test. N-TidalTM also could be effective in monitoring disease progression, providing a possible alternative to spirometry for disease monitoring.

A machine learning algorithm for detecting abnormal patterns in continuous capnography and pulse oximetry monitoring

Continuous capnography monitors patient ventilation but can be susceptible to artifact, resulting in alarm fatigue. Development of smart algorithms may facilitate accurate detection of abnormal ventilation, allowing intervention before patient deterioration. The objective of this analysis was to use machine learning (ML) to classify combined waveforms of continuous capnography and pulse oximetry as normal or abnormal. We used data collected during the observational, prospective PRODIGY trial, in which patients receiving parenteral opioids underwent continuous capnography and pulse oximetry monitoring while on the general care floor [1]. Abnormal ventilation segments in the data stream were reviewed by nine experts and inter-rater agreement was assessed. Abnormal segments were defined as the time series 60s before and 30s after an abnormal pattern was detected. Normal segments (90s continuous monitoring) were randomly sampled and filtered to discard sequences with missing values. Five ML models were trained on extracted features and optimized towards an Fβ score with β = 2. The results show a high inter-rater agreement (> 87%), allowing 7,858 sequences (2,944 abnormal) to be used for model development. Data were divided into 80% training and 20% test sequences. The XGBoost model had the highest Fβ score of 0.94 (with β = 2), showcasing an impressive recall of 0.98 against a precision of 0.83. This study presents a promising advancement in respiratory monitoring, focusing on reducing false alarms and enhancing accuracy of alarm systems. Our algorithm reliably distinguishes normal from abnormal waveforms. More research is needed to define patterns to distinguish abnormal ventilation from artifacts.

Artificial intelligence in emergency medicine. A systematic literature review

Motivation and objective: Emergency medicine is becoming a popular application area for artificial intelligence methods but remains less investigated than other healthcare branches. The need for time-sensitive decision-making on the basis of high data volumes makes the use of quantitative technologies inevitable. However, the specifics of healthcare regulations impose strict requirements for such applications. Published contributions cover separate parts of emergency medicine and use disparate data and algorithms. This study aims to systematize the relevant contributions, investigate the main obstacles to artificial intelligence applications in emergency medicine, and propose directions for further studies.

METHODS

The contributions selection process was conducted with systematic electronic databases querying and filtering with respect to established exclusion criteria. Among the 380 papers gathered from IEEE Xplore, ACM Digital Library, Springer Library, ScienceDirect, and Nature databases 116 were considered to be a part of the survey. The main features of the selected papers are the focus on emergency medicine and the use of machine learning or deep learning algorithms.

FINDINGS AND DISCUSSION

The selected papers were classified into two branches: diagnostics-specific and triage-specific. The former ones are focused on either diagnosis prediction or decision support. The latter covers such applications as mortality, outcome, admission prediction, condition severity estimation, and urgent care prediction. The observed contributions are highly specialized within a single disease or medical operation and often use privately collected retrospective data, making them incomparable. These and other issues can be addressed by creating an end-to-end solution based on human-machine interaction.

CONCLUSION

Artificial intelligence applications are finding their place in emergency medicine, while most of the corresponding studies remain isolated and lack higher generalization and more sophisticated methodology, which can be a matter of forthcoming improvements.

Deep learning classification of capnography waveforms: secondary analysis of the PRODIGY study

Capnography monitors trigger high priority 'no breath' alarms when CO2 measurements do not exceed a given threshold over a specified time-period. False alarms occur when the underlying breathing pattern is stable, but the alarm is triggered when the CO2 value reduces even slightly below the threshold. True 'no breath' events can be falsely classified as breathing if waveform artifact causes an aberrant spike in CO2 values above the threshold. The aim of this study was to determine the accuracy of a deep learning approach to classifying segments of capnography waveforms as either 'breath' or 'no breath'. A post hoc secondary analysis of data from 9 North American sites included in the PRediction of Opioid-induced Respiratory Depression In Patients Monitored by capnoGraphY (PRODIGY) study was conducted. We used a convolutional neural network to classify 15 s capnography waveform segments drawn from a random sample of 400 participants. Loss was calculated over batches of 32 using the binary cross-entropy loss function with weights updated using the Adam optimizer. Internal-external validation was performed by iteratively fitting the model using data from all but one hospital and then assessing its performance in the remaining hospital. The labelled dataset consisted of 10,391 capnography waveform segments. The neural network's accuracy was 0.97, precision was 0.97 and recall was 0.96. Performance was consistent across hospitals in internal-external validation. The neural network could reduce false capnography alarms. Further research is needed to compare the frequency of alarms derived from the neural network with the standard approach.

Low-Order Mechanistic Models for Volumetric and Temporal Capnography: Development, Validation, and Application

OBJECTIVE

Develop low-order mechanistic models accounting quantitatively for, and identifiable from, the capnogram - the CO 2 concentration in exhaled breath, recorded over time (Tcap) or exhaled volume (Vcap).

METHODS

The airflow model's single "alveolar" compartment has compliance and inertance, and feeds a resistive unperfused airway comprising a laminar-flow region followed by a turbulent-mixing region. The gas-mixing model tracks mixing-region CO 2 concentration, fitted breath-by-breath to the measured capnogram, yielding estimates of model parameters that characterize the capnogram.

RESULTS

For the 17 examined records (310 breaths) of airflow, airway pressure and Tcap from ventilated adult patients, the models fit closely (mean rmse 1% of end-tidal CO 2 concentration on Vcap; 1.7% on Tcap). The associated parameters (4 for Vcap, 5 for Tcap) for each exhalation, and airflow parameters for the corresponding forced inhalation, are robustly estimated, and consonant with literature values. The models also allow, using Tcap alone, estimation of the entire exhaled airflow waveform to within a scaling. This suggests new Tcap-based tests, analogous to spirometry but with normal breathing, for discriminating chronic obstructive pulmonary disease (COPD) from congestive heart failure (CHF). A version trained on 15 exhalations from each of 24 COPD/24 CHF Tcap records from one hospital, then tested 100 times with 15 random exhalations from each of 27 COPD/31 CHF Tcap records at another, gave mean accuracy 80.6% (stdev 2.1%). Another version, tested on 29 COPD/32 CHF, yielded AUROC 0.84.

CONCLUSION

Our mechanistic models closely fit Tcap and Vcap measurements, and yield subject-specific parameter estimates.

SIGNIFICANCE

This can inform cardiorespiratory care.

Machine diagnosis of chronic obstructive pulmonary disease using a novel fast-response capnometer

BACKGROUND

Although currently most widely used in mechanical ventilation and cardiopulmonary resuscitation, features of the carbon dioxide (CO₂) waveform produced through capnometry have been shown to correlate with V/Q mismatch, dead space volume, type of breathing pattern, and small airway obstruction. This study applied feature engineering and machine learning techniques to capnography data collected by the N-Tidal™ device across four clinical studies to build a classifier that could distinguish CO₂ recordings (capnograms) of patients with COPD from those without COPD.

METHODS

Capnography data from four longitudinal observational studies (CBRS, GBRS, CBRS2 and ABRS) was analysed from 295 patients, generating a total of 88,186 capnograms. CO₂ sensor data was processed using TidalSense's regulated cloud platform, performing real-time geometric analysis on CO₂ waveforms to generate 82 physiologic features per capnogram. These features were used to train machine learning classifiers to discriminate COPD from 'non-COPD' (a group that included healthy participants and those with other cardiorespiratory conditions); model performance was validated on independent test sets.

RESULTS

The best machine learning model (XGBoost) performance provided a class-balanced AUROC of 0.985 ± 0.013, positive predictive value (PPV) of 0.914 ± 0.039 and sensitivity of 0.915 ± 0.066 for a diagnosis of COPD. The waveform features that are most important for driving classification are related to the alpha angle and expiratory plateau regions. These features correlated with spirometry readings, supporting their proposed properties as markers of COPD.

CONCLUSION

The N-Tidal™ device can be used to accurately diagnose COPD in near-real-time, lending support to future use in a clinical setting.

TRIAL REGISTRATION

Please see NCT03615365, NCT02814253, NCT04504838 and NCT03356288.

Digitization and Analysis of Capnography Using Image Processing Technique

The study of carbon dioxide expiration is called capnometry. The graphical representation of capnometry is called capnography. There is a growing interest in the usage of capnography as the usage has expanded toward the study of metabolism, circulation, lung perfusion and diffusion, quality of spontaneous respiration, and patency of airways outside of its typical usage in the anesthetic and emergency medicine field. The parameters of the capnograph could be classified as carbon dioxide (CO₂) concentration and time points and coordinates, slopes angle, volumetric studies, and functional transformation of wave data. Up to date, there is no gold standard device for the calculation of the capnographic parameters. Capnography digitization using the image processing technique could serve as an option. From the algorithm we developed, eight identical breath waves were tested by four investigators. The values of the parameters chosen showed no significant difference between investigators. Although there were no significant differences between any of the parameters tested, there were a few related parameters that were not calculable. Further testing after refinement of the algorithm could be done. As more capnographic parameters are being derived and rediscovered by clinicians and researchers alike for both lung and non-lung-related diseases, there is a dire need for data analysis and interpretation. Although the proposed algorithm still needs minor refinements and further large-scale testing, we proposed that the digitization of the capnograph via image processing technique could serve as an intellectual option as it is fast, convenient, easy to use, and efficient.

Carbon dioxide monitoring in the newborn infant

Carbon dioxide (CO₂ ) monitoring is vital during mechanical ventilation of newborn infants, as morbidity increases when CO₂ levels are inappropriate. Our aim was to review the uses and limitations of such noninvasive monitoring methods. Colorimetry is primarily utilized during resuscitation to determine whether successful intubation has occurred. False negative and positive results can however lead to delays in detecting tracheal versus esophageal intubation. Transcutaneous carbon dioxide sensors have limited use during resuscitation, but can be utilized to provide continuous trend data during on-going ventilation. End-tidal capnography can provide clinicians with quantitative end-tidal CO₂ (EtCO₂ ) values and a continuous real-time capnogram waveform trace. These devices are becoming more widely accepted for use in the neonatal population as the new devices are lightweight with minimal additional dead space. Nevertheless, they have been reported to have variable accuracy when compared to arterial CO₂ measurements, however, divergence of results may be related to disease severity rather than technological limitations. During resuscitation EtCO₂ can be detected by capnography more rapidly than by colorimetry. Furthermore, capnography can be currently utilized in neonatal research settings to determine the physiological dead space and ventilation inhomogeneity, and thus has potential to be beneficial to clinical care. In conclusion, novel modes of noninvasive carbon dioxide monitoring can be safely and reliably utilized in newborn infants during mechanical ventilation. Future randomized trials should aim to address which device provides the most optimal form of monitoring in different clinical contexts.

Use of capnography for prediction of obstruction severity in non-intubated COPD and asthma patients

BACKGROUND

Capnography waveform contains essential information regarding physiological characteristics of the airway and thus indicative of the level of airway obstruction. Our aim was to develop a capnography-based, point-of-care tool that can estimate the level of obstruction in patients with asthma and COPD.

METHODS

Two prospective observational studies conducted between September 2016 and May 2018 at Rabin Medical Center, Israel, included healthy, asthma and COPD patient groups. Each patient underwent spirometry test and continuous capnography, as part of, either methacholine challenge test for asthma diagnosis or bronchodilator reversibility test for asthma and COPD routine evaluation. Continuous capnography signal, divided into single breaths waveforms, were analyzed to identify waveform features, to create a predictive model for FEV1 using an artificial neural network. The gold standard for comparison was FEV1 measured with spirometry.

MEASUREMENTS AND MAIN RESULTS

Overall 160 patients analyzed. Model prediction included 32/88 waveform features and three demographic features (age, gender and height). The model showed excellent correlation with FEV1 (R = 0.84), R² achieved was 0.7 with mean square error of 0.13.

CONCLUSION

In this study we have developed a model to evaluate FEV1 in asthma and COPD patients. Using this model, as a point-of-care tool, we can evaluate the airway obstruction level without reliance on patient cooperation. Moreover, continuous FEV1 monitoring can identify disease fluctuations, response to treatment and guide therapy.

TRIAL REGISTRATION

clinical trials.gov, NCT02805114. Registered 17 June 2016, https://clinicaltrials.gov/ct2/show/NCT02805114.

Automatic classification of regular and irregular capnogram segments using time- and frequency-domain features: A machine learning-based approach

BACKGROUND

The quantitative features of a capnogram signal are important clinical metrics in assessing pulmonary function. However, these features should be quantified from the regular (artefact-free) segments of the capnogram waveform.

OBJECTIVE

This paper presents a machine learning-based approach for the automatic classification of regular and irregular capnogram segments.

METHODS

Herein, we proposed four time- and two frequency-domain features experimented with the support vector machine classifier through ten-fold cross-validation. MATLAB simulation was conducted on 100 regular and 100 irregular 15 s capnogram segments. Analysis of variance was performed to investigate the significance of the proposed features. Pearson's correlation was utilized to select the relatively most substantial ones, namely variance and the area under normalized magnitude spectrum. Classification performance, using these features, was evaluated against two feature sets in which either time- or frequency-domain features only were employed.

RESULTS

Results showed a classification accuracy of 86.5%, which outperformed the other cases by an average of 5.5%. The achieved specificity, sensitivity, and precision were 84%, 89% and 86.51%, respectively. The average execution time for feature extraction and classification per segment is only 36 ms.

CONCLUSION

The proposed approach can be integrated with capnography devices for real-time capnogram-based respiratory assessment. However, further research is recommended to enhance the classification performance.

Comparing Nasal End-Tidal Carbon Dioxide Measurement Variation and Agreement While Delivering Pulsed and Continuous Flow Oxygen in Volunteers and Patients

BACKGROUND

Supplemental oxygen is administered during procedural sedation to prevent hypoxemia. Continuous flow oxygen, the most widespread method, is generally adequate but distorts capnography. Pulsed flow oxygen is novel and ideally will not distort capnography. We have developed a prototype oxygen administration system designed to try to facilitate end-tidal carbon dioxide (ETCO2) measurement. We conducted a volunteer study (ClinicalTrials.gov, NCT02886312) to determine how much nasal ETCO2 measurements vary with oxygen flow rate. We also conducted a clinical study (NCT02962570) to determine the median difference and limits of agreement between ETCO2 measurements made with and without administering oxygen.

METHODS

Both studies were conducted at the University of Utah and participants acted as their own control. Inclusion criteria were age 18 years and older with an American Society of Anesthesiologists physical status of I-III. Exclusion criteria included acute respiratory distress syndrome, pneumonia, lung or cardiovascular disease, nasal/bronchial congestion, pregnancy, oxygen saturation measured by pulse oximetry <93%, and a procedure scheduled for <20 minutes. For the volunteer study, pulsed and continuous flow was administered at rates from 2 to 10 L/min using a single sequence of technique and flow. The median absolute deviation from the median value was analyzed for the primary outcome of ETCO2. For the clinical study, ETCO2 measurements (the primary outcome) were collected while administering pulsed and continuous flow at rates between 1 and 5 L/min and were compared to measurements without oxygen flow. Due to institutional review board requirements for patient safety, this study was not randomized. After completing the study, measurements with and without administering oxygen were analyzed to determine median differences and 95% limits of agreement for each administration technique.

RESULTS

Thirty volunteers and 60 patients participated in these studies which ended after enrolling the predetermined number of participants. In volunteers, the median absolute deviation for ETCO2 measurements made while administering pulsed flow oxygen (0.89; 25%-75% quantiles: 0.3-1.2) was smaller than while administering continuous flow oxygen (3.93; 25%-75% quantiles: 2.2-6.2). In sedated patients, the median difference was larger during continuous flow oxygen (-6.8 mm Hg; 25%-75% quantiles: -12.5 to -2.1) than during pulsed flow oxygen (0.1 mm Hg; 25%-75% quantiles: -0.5 to 1.5). The 95% limits of agreement were also narrower during pulsed flow oxygen (-2.4 to 4.5 vs -30.5 to 2.4 mm Hg).

CONCLUSIONS

We have shown that nasal ETCO2 measurements while administering pulsed flow have little deviation and agree well with measurements made without administering oxygen. We have also demonstrated that ETCO2 measurements during continuous flow oxygen have large deviation and wide limits of agreement when compared with measurements made without administering oxygen.

An Enhanced Mechanistic Model For Capnography, With Application To CHF-COPD Discrimination

Capnography records CO₂ partial pressure in exhaled breath as a function of time or exhaled volume. Time-based capnography, which is our focus, is a point-of-care, noninvasive, effort-independent and widely available clinical monitoring modality. The generated waveform, or capnogram, reflects the ventilation-perfusion dynamics of the lung, and thus has value in the diagnosis of respiratory conditions such as chronic obstructive pulmonary disease (COPD). Effective discrimination between normal respiration and obstructive lung disease can be performed using capnogram-derived estimates of respiratory parameters in a simple mechanistic model of CO₂ exhalation. We propose an enhanced mechanistic model that can capture specific capnogram characteristics in congestive heart failure (CHF) by incorporating a representation of the inertance associated with fluid in the lungs. The 4 associated parameters are estimated on a breath-by-breath basis by fitting the model output to the exhalations in the measured capnogram. Estimated parameters from 40 exhalations of 7 CHF and 7 COPD patients were used as a training set to design a quadratic discriminator in the parameter space, aimed at distinguishing between CHF and COPD patients. The area under the ROC curve for the training set was 0.94, and the corresponding equal-error-rate value of approximately 0.1 suggests classification accuracies of the order of 90% are attainable. Applying this discriminator without modification to 40 exhalations from each CHF and COPD patient in a fresh test set, and deciding on a simple majority basis whether the patient has CHF or COPD, results in correctly labeling all 8 out of the 8 CHF patients and 6 out of the 8 COPD patients in the test set, corresponding to a classification accuracy of 87.5%.

Journal of Clinical Monitoring and Computing 2017 end of year summary: respiration

This paper reviews 32 papers or commentaries published in Journal of Clinical Monitoring and Computing in 2016, within the field of respiration. Papers were published covering airway management, ventilation and respiratory rate monitoring, lung mechanics and gas exchange monitoring, in vitro monitoring of lung mechanics, CO₂ monitoring, and respiratory and metabolic monitoring techniques.

Journal of Clinical Monitoring and Computing 2016 end of year summary: respiration

This paper reviews 16 papers or commentaries published in Journal of Clinical Monitoring and Computing in 2016, within the field of respiration. Papers were published covering peri- and post-operative monitoring of respiratory rate, perioperative monitoring of CO₂, modeling of oxygen gas exchange, and techniques for respiratory monitoring.

Current methodological and technical limitations of time and volumetric capnography in newborns

Although capnography is a standard tool in mechanically ventilated adult and pediatric patients, it has physiological and technical limitations in neonates. Gas exchange differs between small and adult lungs due to the greater impact of small airways on gas exchange, the higher impact of the apparatus dead space on measurements due to lower tidal volume and the occurrence of air leaks in intubated patients. The high respiratory rate and low tidal volume in newborns, especially those with stiff lungs, require main-stream sensors with fast response times and minimal dead-space or low suction flow when using side-stream measurements. If these technical requirements are not fulfilled, the measured end-tidal CO2 (P et CO 2 ), which should reflect the alveolar CO2 and the calculated airway dead spaces, can be misleading. The aim of this survey is to highlight the current limitations of capnography in very young patients to avoid pitfalls associated with the interpretation of capnographic parameters, and to describe further developments.