Replicating human expertise of mechanical ventilation waveform analysis in detecting patient-ventilator cycling asynchrony using machine learning

A Clinician's Guide to Understanding Bias in Critical Clinical Prediction Models

This narrative review focuses on the role of clinical prediction models in supporting informed decision-making in critical care, emphasizing their 2 forms: traditional scores and artificial intelligence (AI)-based models. Acknowledging the potential for both types to embed biases, the authors underscore the importance of critical appraisal to increase our trust in models. The authors outline recommendations and critical care examples to manage risk of bias in AI models. The authors advocate for enhanced interdisciplinary training for clinicians, who are encouraged to explore various resources (books, journals, news Web sites, and social media) and events (Datathons) to deepen their understanding of risk of bias.

Detection and quantitative analysis of patient-ventilator interactions in ventilated infants by deep learning networks

BACKGROUND

The study of patient-ventilator interactions (PVI) in mechanically ventilated neonates is limited by the lack of unified PVI definitions and tools to perform large scale analyses.

METHODS

An observational study was conducted in 23 babies randomly selected from 170 neonates who were ventilated with SIPPV-VG, SIMV-VG or PSV-VG mode for at least 12 h. 500 breaths were randomly selected and manually annotated from each recording to train convolutional neural network (CNN) models for PVI classification.

RESULTS

The average asynchrony index (AI) over all recordings was 52.5%. The most frequently occurring PVIs included expiratory work (median: 28.4%, interquartile range: 23.2-40.2%), late cycling (7.6%, 2.8-10.2%), failed triggering (4.6%, 1.2-6.2%) and late triggering (4.4%, 2.8-7.4%). Approximately 25% of breaths with a PVI had two or more PVIs occurring simultaneously. Binary CNN classifiers were developed for PVIs affecting ≥1% of all breaths (n = 7) and they achieved F1 scores of >0.9 on the test set except for early triggering where it was 0.809.

CONCLUSIONS

PVIs occur frequently in neonates undergoing conventional mechanical ventilation with a significant proportion of breaths containing multiple PVIs. We have developed computational models for seven different PVIs to facilitate automated detection and further evaluation of their clinical significance in neonates.

IMPACT

The study of patient-ventilator interactions (PVI) in mechanically ventilated neonates is limited by the lack of unified PVI definitions and tools to perform large scale analyses. By adapting a recent taxonomy of PVI definitions in adults, we have manually annotated neonatal ventilator waveforms to determine prevalence and co-occurrence of neonatal PVIs. We have also developed binary deep learning classifiers for common PVIs to facilitate their automatic detection and quantification.

The effects of real-time waveform analysis software on patient ventilator synchronization during pressure support ventilation: a randomized crossover physiological study

BACKGROUND

Patient-ventilator asynchrony commonly occurs during pressure support ventilation (PSV). IntelliSync + software (Hamilton Medical AG, Bonaduz, Switzerland) is a new ventilation technology that continuously analyzes ventilator waveforms to detect the beginning and end of patient inspiration in real time. This study aimed to evaluate the physiological effect of IntelliSync + software on inspiratory trigger delay time, delta airway (Paw) and esophageal (Pes) pressure drop during the trigger phase, airway occlusion pressure at 0.1 s (P0.1), and hemodynamic variables.

METHODS

A randomized crossover physiologic study was conducted in 14 mechanically ventilated patients under PSV. Patients were randomly assigned to receive conventional flow trigger and cycling, inspiratory trigger synchronization (I-sync), cycle synchronization (C-sync), and inspiratory trigger and cycle synchronization (I/C-sync) for 15 min at each step. Other ventilator settings were kept constant. Paw, Pes, airflow, P0.1, respiratory rate, SpO2, and hemodynamic variables were recorded. The primary outcome was inspiratory trigger and cycle delay time between each intervention. Secondary outcomes were delta Paw and Pes drop during the trigger phase, P0.1, SpO2, and hemodynamic variables.

RESULTS

The time to initiate the trigger was significantly shorter with I-sync compared to baseline (208.9±91.7 vs. 301.4±131.7 msec; P = 0.002) and I/C-sync compared to baseline (222.8±94.0 vs. 301.4±131.7 msec; P = 0.005). The I/C-sync group had significantly lower delta Paw and Pes drop during the trigger phase compared to C-sync group (-0.7±0.4 vs. -1.2±0.8 cmH2O; P = 0.028 and - 1.8±2.2 vs. -2.8±3.2 cmH2O; P = 0.011, respectively). No statistically significant differences were found in cycle delay time, P0.1 and other physiological variables between the groups.

CONCLUSIONS

IntelliSync + software reduced inspiratory trigger delay time compared to the conventional flow trigger system during PSV mode. However, no significant improvements in cycle delay time and other physiological variables were observed with IntelliSync + software.

TRIAL REGISTRATION

This study was registered in the Thai Clinical Trial Registry (TCTR20200528003; date of registration 28/05/2020).

Quantifiable identification of flow-limited ventilator dyssynchrony with the deformed lung ventilator model

BACKGROUND

Ventilator dyssynchrony (VD) can worsen lung injury and is challenging to detect and quantify due to the complex variability in the dyssynchronous breaths. While machine learning (ML) approaches are useful for automating VD detection from the ventilator waveform data, scalable severity quantification and its association with pathogenesis and ventilator mechanics remain challenging.

OBJECTIVE

We develop a systematic framework to quantify pathophysiological features observed in ventilator waveform signals such that they can be used to create feature-based severity stratification of VD breaths.

METHODS

A mathematical model was developed to represent the pressure and volume waveforms of individual breaths in a feature-based parametric form. Model estimates of respiratory effort strength were used to assess the severity of flow-limited (FL)-VD breaths compared to normal breaths. A total of 93,007 breath waveforms from 13 patients were analyzed.

RESULTS

A novel model-defined continuous severity marker was developed and used to estimate breath phenotypes of FL-VD breaths. The phenotypes had a predictive accuracy of over 97% with respect to the previously developed ML-VD identification algorithm. To understand the incidence of FL-VD breaths and their association with the patient state, these phenotypes were further successfully correlated with ventilator-measured parameters and electronic health records.

CONCLUSION

This work provides a computational pipeline to identify and quantify the severity of FL-VD breaths and paves the way for a large-scale study of VD causes and effects. This approach has direct application to clinical practice and in meaningful knowledge extraction from the ventilator waveform data.

The Development, Optimization, and Validation of Four Different Machine Learning Algorithms to Identify Ventilator Dyssynchrony

Invasive mechanical ventilation can worsen lung injury. Ventilator dyssynchrony (VD) may propagate ventilator-induced lung injury (VILI) and is challenging to detect and systematically monitor because each patient takes approximately 25,000 breaths a day yet some types of VD are rare, accounting for less than 1% of all breaths. Therefore, we sought to develop and validate accurate machine learning (ML) algorithms to detect multiple types of VD by leveraging esophageal pressure waveform data to quantify patient effort with airway pressure, flow, and volume data generated during mechanical ventilation, building a computational pipeline to facilitate the study of VD.

Materials and Methods

We collected ventilator waveform and esophageal pressure data from 30 patients admitted to the ICU. Esophageal pressure allows the measurement of transpulmonary pressure and patient effort. Waveform data were cleaned, features considered essential to VD detection were calculated, and a set of 10,000 breaths were manually labeled. Four ML algorithms were trained to classify each type of VD: logistic regression, support vector classification, random forest, and XGBoost.

Results

We trained ML models to detect different families and seven types of VD with high sensitivity (>90% and >80%, respectively). Three types of VD remained difficult for ML to classify because of their rarity and lack of sample size. XGBoost classified breaths with increased specificity compared to other ML algorithms.

Discussion

We developed ML models to detect multiple types of VD accurately. The ability to accurately detect multiple VD types addresses one of the significant limitations in understanding the role of VD in affecting patient outcomes.

Conclusion

ML models identify multiple types of VD by utilizing esophageal pressure data and airway pressure, flow, and volume waveforms. The development of such computational pipelines will facilitate the identification of VD in a scalable fashion, allowing for the systematic study of VD and its impact on patient outcomes.

Automated evaluation of typical patient-ventilator asynchronies based on lung hysteretic responses

BACKGROUND

Patient-ventilator asynchrony is common during mechanical ventilation (MV) in intensive care unit (ICU), leading to worse MV care outcome. Identification of asynchrony is critical for optimizing MV settings to reduce or eliminate asynchrony, whilst current clinical visual inspection of all typical types of asynchronous breaths is difficult and inefficient. Patient asynchronies create a unique pattern of distortions in hysteresis respiratory behaviours presented in pressure-volume (PV) loop.

METHODS

Identification method based on hysteretic lung mechanics and hysteresis loop analysis is proposed to delineate the resulted changes of lung mechanics in PV loop during asynchronous breathing, offering detection of both its incidence and 7 major types. Performance is tested against clinical patient data with comparison to visual inspection conducted by clinical doctors.

RESULTS

The identification sensitivity and specificity of 11 patients with 500 breaths for each patient are above 89.5% and 96.8% for all 7 types, respectively. The average sensitivity and specificity across all cases are 94.6% and 99.3%, indicating a very good accuracy. The comparison of statistical analysis between identification and human inspection yields the essential same clinical judgement on patient asynchrony status for each patient, potentially leading to the same clinical decision for setting adjustment.

CONCLUSIONS

The overall results validate the accuracy and robustness of the identification method for a bedside monitoring, as well as its ability to provide a quantified metric for clinical decision of ventilator setting. Hence, the method shows its potential to assist a more consistent and objective assessment of asynchrony without undermining the efficacy of the current clinical practice.

Impact on the ability of healthcare professionals to correctly identify patient-ventilator asynchronies of the simultaneous visualization of estimated muscle pressure curves on the ventilator display: a randomized study (Pmus study)

BACKGROUND

Patient-ventilator asynchronies are usually detected by visual inspection of ventilator waveforms but with low sensitivity, even when performed by experts in the field. Recently, estimation of the inspiratory muscle pressure (P_mus) waveforms through artificial intelligence algorithm has been proposed (Magnamed®, São Paulo, Brazil). We hypothesized that the display of these waveforms could help healthcare providers identify patient-ventilator asynchronies.

METHODS

A prospective single-center randomized study with parallel assignment was conducted to assess whether the display of the estimated P_mus waveform would improve the correct identification of asynchronies in simulated clinical scenarios. The primary outcome was the mean asynchrony detection rate (sensitivity). Physicians and respiratory therapists who work in intensive care units were randomized to control or intervention group. In both groups, participants analyzed pressure and flow waveforms of 49 different scenarios elaborated using the ASL-5000 lung simulator. In the intervention group the estimated P_mus waveform was displayed in addition to pressure and flow waveforms.

RESULTS

A total of 98 participants were included, 49 per group. The sensitivity per participant in identifying asynchronies was significantly higher in the P_mus group (65.8 ± 16.2 vs. 52.94 ± 8.42, p < 0.001). This effect remained when stratifying asynchronies by type.

CONCLUSIONS

We showed that the display of the P_mus waveform improved the ability of healthcare professionals to recognize patient-ventilator asynchronies by visual inspection of ventilator tracings. These findings require clinical validation.

TRIAL REGISTRATION

ClinicalTrials.gov: NTC05144607. Retrospectively registered 3 December 2021.

A Comparative Study on Predication of Appropriate Mechanical Ventilation Mode through Machine Learning Approach

Ventilation mode is one of the most crucial ventilator settings, selected and set by knowledgeable critical care therapists in a critical care unit. The application of a particular ventilation mode must be patient-specific and patient-interactive. The main aim of this study is to provide a detailed outline regarding ventilation mode settings and determine the best machine learning method to create a deployable model for the appropriate selection of ventilation mode on a per breath basis. Per-breath patient data is utilized, preprocessed and finally a data frame is created consisting of five feature columns (inspiratory and expiratory tidal volume, minimum pressure, positive end-expiratory pressure, and previous positive end-expiratory pressure) and one output column (output column consisted of modes to be predicted). The data frame has been split into training and testing datasets with a test size of 30%. Six machine learning algorithms were trained and compared for performance, based on the accuracy, F1 score, sensitivity, and precision. The output shows that the Random-Forest Algorithm was the most precise and accurate in predicting all ventilation modes correctly, out of the all the machine learning algorithms trained. Thus, the Random-Forest machine learning technique can be utilized for predicting optimal ventilation mode setting, if it is properly trained with the help of the most relevant data. Aside from ventilation mode, control parameter settings, alarm settings and other settings may also be adjusted for the mechanical ventilation process utilizing appropriate machine learning, particularly deep learning approaches.

Automated detection and classification of patient-ventilator asynchrony by means of machine learning and simulated data

BACKGROUND AND OBJECTIVE

Mechanical ventilation is a lifesaving treatment for critically ill patients in an Intensive Care Unit (ICU) or during surgery. However, one potential harm of mechanical ventilation is related to patient-ventilator asynchrony (PVA). PVA can cause discomfort to the patient, damage to the lungs, and an increase in the length of stay in the ICU and on the ventilator. Therefore, automated detection algorithms are being developed to detect and classify PVAs, with the goal of optimizing mechanical ventilation. However, the development of these algorithms often requires large labeled datasets; these are generally difficult to obtain, as their collection and labeling is a time-consuming and labor-intensive task, which needs to be performed by clinical experts.

METHODS

In this work, we aimed to develop a computer algorithm for the automatic detection and classification of PVA. The algorithm employs a neural network for the detection of the breath of the patient. The development of the algorithm was aided by simulations from a recently published model of the patient-ventilator interaction.

RESULTS

The proposed method was effective, providing an algorithm with reliable detection and classification results of over 90% accuracy. Besides presenting a detection and classification algorithm for a variety of PVAs, here we show that using simulated data in combination with clinical data increases the variability in the training dataset, leading to a gain in performance and generalizability.

CONCLUSIONS

In the future, these algorithms can be utilized to gain a better understanding of the clinical impact of PVAs and help clinicians to better monitor their ventilation strategies.

Neural Network-Enabled Identification of Weak Inspiratory Efforts during Pressure Support Ventilation Using Ventilator Waveforms

During pressure support ventilation (PSV), excessive assist results in weak inspiratory efforts and promotes diaphragm atrophy and delayed weaning. The aim of this study was to develop a classifier using a neural network to identify weak inspiratory efforts during PSV, based on the ventilator waveforms. Recordings of flow, airway, esophageal and gastric pressures from critically ill patients were used to create an annotated dataset, using data from 37 patients at 2-5 different levels of support, computing the inspiratory time and effort for every breath. The complete dataset was randomly split, and data from 22 patients (45,650 breaths) were used to develop the model. Using a One-Dimensional Convolutional Neural Network, a predictive model was developed to characterize the inspiratory effort of each breath as weak or not, using a threshold of 50 cmH₂O*s/min. The following results were produced by implementing the model on data from 15 different patients (31,343 breaths). The model predicted weak inspiratory efforts with a sensitivity of 88%, specificity of 72%, positive predictive value of 40%, and negative predictive value of 96%. These results provide a 'proof-of-concept' for the ability of such a neural-network based predictive model to facilitate the implementation of personalized assisted ventilation.

Research Progress of Respiratory Disease and Idiopathic Pulmonary Fibrosis Based on Artificial Intelligence

Machine Learning (ML) is an algorithm based on big data, which learns patterns from the previously observed data through classifying, predicting, and optimizing to accomplish specific tasks. In recent years, there has been rapid development in the field of ML in medicine, including lung imaging analysis, intensive medical monitoring, mechanical ventilation, and there is need for intubation etiology prediction evaluation, pulmonary function evaluation and prediction, obstructive sleep apnea, such as biological information monitoring and so on. ML can have good performance and is a great potential tool, especially in the imaging diagnosis of interstitial lung disease. Idiopathic pulmonary fibrosis (IPF) is a major problem in the treatment of respiratory diseases, due to the abnormal proliferation of fibroblasts, leading to lung tissue destruction. The diagnosis mainly depends on the early detection of imaging and early treatment, which can effectively prolong the life of patients. If the computer can be used to assist the examination results related to the effects of fibrosis, a timely diagnosis of such diseases will be of great value to both doctors and patients. We also previously proposed a machine learning algorithm model that can play a good clinical guiding role in early imaging prediction of idiopathic pulmonary fibrosis. At present, AI and machine learning have great potential and ability to transform many aspects of respiratory medicine and are the focus and hotspot of research. AI needs to become an invisible, seamless, and impartial auxiliary tool to help patients and doctors make better decisions in an efficient, effective, and acceptable way. The purpose of this paper is to review the current application of machine learning in various aspects of respiratory diseases, with the hope to provide some help and guidance for clinicians when applying algorithm models.

Development and external validation of a machine learning-based prediction model for the cancer-related fatigue diagnostic screening in adult cancer patients: a cross-sectional study in China

PURPOSE

Cancer-related fatigue (CRF) is the most common symptom in cancer patients and may interfere with patients' daily activities and decrease survival rate. However, the etiology of CRF has not been identified. Diagnosing CRF is challenging. Thus, our study aimed to develop a CRF prediction model in cancer patients, using data that healthcare professionals routinely obtained from electronic health records (EHRs) based on the 3P model and externally validate this model in an independent dataset collected from another hospital.

METHODS

Between April 2022 and September 2022, a cross-sectional study was conducted on adult cancer patients at two first-class tertiary hospitals in China. Data that healthcare professionals routinely obtained from electronic health records (EHRs) based on the 3P model were collected. The outcome measure was according to ICD-10 diagnostic criteria for CRF. Data from one hospital (n = 305) were used for model development and internal validation. An independent data set from another hospital (n = 260) was utilized for external validation. logistic regression, random forest (RF), Naive Bayes (NB), and extreme gradient boosting (XGBoost) were constructed and compared. The model performance was evaluated in terms of both discrimination and calibration.

RESULTS

The prevalence of CRF in the two centers was 57.9% and 56.1%, respectively. The Random Forest model achieved the highest AUC of 0.86 among the four types of classifiers in the internal validation. The AUC of RF and NB were above 0.7 in the external validation, suggesting that the models also have an acceptable generalization ability.

CONCLUSIONS

The incidence of CRF remains high and deserves more attention. The fatigue prediction model based on the 3P theory can accurately predict the risk of CRF. Nonlinear algorithms such as Random Forest and Naive Bayes are more suitable for diagnosing and evaluating symptoms.

Reconstructing asynchrony for mechanical ventilation using a hysteresis loop virtual patient model

BACKGROUND

Patient-specific lung mechanics during mechanical ventilation (MV) can be identified from measured waveforms of fully ventilated, sedated patients. However, asynchrony due to spontaneous breathing (SB) effort can be common, altering these waveforms and reducing the accuracy of identified, model-based, and patient-specific lung mechanics.

METHODS

Changes in patient-specific lung elastance over a pressure-volume (PV) loop, identified using hysteresis loop analysis (HLA), are used to detect the occurrence of asynchrony and identify its type and pattern. The identified HLA parameters are then combined with a nonlinear mechanics hysteresis loop model (HLM) to extract and reconstruct ventilated waveforms unaffected by asynchronous breaths. Asynchrony magnitude can then be quantified using an energy-dissipation metric, Easyn, comparing PV loop area between model-reconstructed and original, altered asynchronous breathing cycles. Performance is evaluated using both test-lung experimental data with a known ground truth and clinical data from four patients with varying levels of asynchrony.

RESULTS

Root mean square errors for reconstructed PV loops are within 5% for test-lung experimental data, and 10% for over 90% of clinical data. Easyn clearly matches known asynchrony magnitude for experimental data with RMS errors < 4.1%. Clinical data performance shows 57% breaths having Easyn > 50% for Patient 1 and 13% for Patient 2. Patient 3 only presents 20% breaths with Easyn > 10%. Patient 4 has Easyn = 0 for 96% breaths showing accuracy in a case without asynchrony.

CONCLUSIONS

Experimental test-lung validation demonstrates the method's reconstruction accuracy and generality in controlled scenarios. Clinical validation matches direct observations of asynchrony in incidence and quantifies magnitude, including cases without asynchrony, validating its robustness and potential efficacy as a clinical real-time asynchrony monitoring tool.

State of the Art of Machine Learning-Enabled Clinical Decision Support in Intensive Care Units: Literature Review

Background

Modern clinical care in intensive care units is full of rich data, and machine learning has great potential to support clinical decision-making. The development of intelligent machine learning–based clinical decision support systems is facing great opportunities and challenges. Clinical decision support systems may directly help clinicians accurately diagnose, predict outcomes, identify risk events, or decide treatments at the point of care.

Objective

We aimed to review the research and application of machine learning–enabled clinical decision support studies in intensive care units to help clinicians, researchers, developers, and policy makers better understand the advantages and limitations of machine learning–supported diagnosis, outcome prediction, risk event identification, and intensive care unit point-of-care recommendations.

Methods

We searched papers published in the PubMed database between January 1980 and October 2020. We defined selection criteria to identify papers that focused on machine learning–enabled clinical decision support studies in intensive care units and reviewed the following aspects: research topics, study cohorts, machine learning models, analysis variables, and evaluation metrics.

Results

A total of 643 papers were collected, and using our selection criteria, 97 studies were found. Studies were categorized into 4 topics—monitoring, detection, and diagnosis (13/97, 13.4%), early identification of clinical events (32/97, 33.0%), outcome prediction and prognosis assessment (46/97, 47.6%), and treatment decision (6/97, 6.2%). Of the 97 papers, 82 (84.5%) studies used data from adult patients, 9 (9.3%) studies used data from pediatric patients, and 6 (6.2%) studies used data from neonates. We found that 65 (67.0%) studies used data from a single center, and 32 (33.0%) studies used a multicenter data set; 88 (90.7%) studies used supervised learning, 3 (3.1%) studies used unsupervised learning, and 6 (6.2%) studies used reinforcement learning. Clinical variable categories, starting with the most frequently used, were demographic (n=74), laboratory values (n=59), vital signs (n=55), scores (n=48), ventilation parameters (n=43), comorbidities (n=27), medications (n=18), outcome (n=14), fluid balance (n=13), nonmedicine therapy (n=10), symptoms (n=7), and medical history (n=4). The most frequently adopted evaluation metrics for clinical data modeling studies included area under the receiver operating characteristic curve (n=61), sensitivity (n=51), specificity (n=41), accuracy (n=29), and positive predictive value (n=23).

Conclusions

Early identification of clinical and outcome prediction and prognosis assessment contributed to approximately 80% of studies included in this review. Using new algorithms to solve intensive care unit clinical problems by developing reinforcement learning, active learning, and time-series analysis methods for clinical decision support will be greater development prospects in the future.

Timing of inspiratory muscle activity detected from airway pressure and flow during pressure support ventilation: the waveform method

Background

Whether respiratory efforts and their timing can be reliably detected during pressure support ventilation using standard ventilator waveforms is unclear. This would give the opportunity to assess and improve patient–ventilator interaction without the need of special equipment.

Methods

In 16 patients under invasive pressure support ventilation, flow and pressure waveforms were obtained from proximal sensors and analyzed by three trained physicians and one resident to assess patient’s spontaneous activity. A systematic method (the waveform method) based on explicit rules was adopted. Esophageal pressure tracings were analyzed independently and used as reference. Breaths were classified as assisted or auto-triggered, double-triggered or ineffective. For assisted breaths, trigger delay, early and late cycling (minor asynchronies) were diagnosed. The percentage of breaths with major asynchronies (asynchrony index) and total asynchrony time were computed.

Results

Out of 4426 analyzed breaths, 94.1% (70.4–99.4) were assisted, 0.0% (0.0–0.2) auto-triggered and 5.8% (0.4–29.6) ineffective. Asynchrony index was 5.9% (0.6–29.6). Total asynchrony time represented 22.4% (16.3–30.1) of recording time and was mainly due to minor asynchronies. Applying the waveform method resulted in an inter-operator agreement of 0.99 (0.98–0.99); 99.5% of efforts were detected on waveforms and agreement with the reference in detecting major asynchronies was 0.99 (0.98–0.99). Timing of respiratory efforts was accurately detected on waveforms: AUC for trigger delay, cycling delay and early cycling was 0.865 (0.853–0.876), 0.903 (0.892–0.914) and 0.983 (0.970–0.991), respectively.

Conclusions

Ventilator waveforms can be used alone to reliably assess patient’s spontaneous activity and patient–ventilator interaction provided that a systematic method is adopted.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-022-03895-4.

Identifying Patient-Ventilator Asynchrony on a Small Dataset Using Image-Based Transfer Learning

Mechanical ventilation is an essential life-support treatment for patients who cannot breathe independently. Patient-ventilator asynchrony (PVA) occurs when ventilatory support does not match the needs of the patient and is associated with a series of adverse clinical outcomes. Deep learning methods have shown a strong discriminative ability for PVA detection, but they require a large number of annotated data for model training, which hampers their application to this task. We developed a transfer learning architecture based on pretrained convolutional neural networks (CNN) and used it for PVA recognition based on small datasets. The one-dimensional signal was converted to a two-dimensional image, and features were extracted by the CNN using pretrained weights for classification. A partial dropping cross-validation technique was developed to evaluate model performance on small datasets. When using large datasets, the performance of the proposed method was similar to that of non-transfer learning methods. However, when the amount of data was reduced to 1%, the accuracy of transfer learning was approximately 90%, whereas the accuracy of the non-transfer learning was less than 80%. The findings suggest that the proposed transfer learning method can obtain satisfactory accuracies for PVA detection when using small datasets. Such a method can promote the application of deep learning to detect more types of PVA under various ventilation modes.

An interpretable 1D convolutional neural network for detecting patient-ventilator asynchrony in mechanical ventilation

BACKGROUND AND OBJECTIVE

Patient-ventilator asynchrony (PVA) is the result of a mismatch between the need of patients and the assistance provided by the ventilator during mechanical ventilation. Because the poor interaction between the patient and the ventilator is associated with inferior clinical outcomes, effort should be made to identify and correct their occurrence. Deep learning has shown promising ability in PVA detection; however, lack of network interpretability hampers its application in clinic.

METHODS

We proposed an interpretable one-dimensional convolutional neural network (1DCNN) to detect four most manifestation types of PVA (double triggering, ineffective efforts during expiration, premature cycling and delayed cycling) under pressure control ventilation mode and pressure support ventilation mode. A global average pooling (GAP) layer was incorporated with the 1DCNN model to highlight the sections of the respiratory waveform the model focused on when making a classification. Dilation convolution and batch normalization were introduced to the 1DCNN model for compensating the reduction of performance caused by the GAP layer.

RESULTS

The proposed interpretable 1DCNN exhibited comparable performance with the state-of-the-art deep learning model in PVA detection. The F1 scores for the detection of four types of PVA under pressure control ventilation and pressure support ventilation modes were greater than 0.96. The critical sections of the waveform used to detect PVA were highlighted, and found to be well consistent with the understanding of the respective type of PVA by experts.

CONCLUSIONS

The findings suggest that the proposed 1DCNN can help detect PVA, and enhance the interpretability of the classification process to help clinicians better understand the results obtained from deep learning technology.

Automated detection and quantification of reverse triggering effort under mechanical ventilation

BACKGROUND

Reverse triggering (RT) is a dyssynchrony defined by a respiratory muscle contraction following a passive mechanical insufflation. It is potentially harmful for the lung and the diaphragm, but its detection is challenging. Magnitude of effort generated by RT is currently unknown. Our objective was to validate supervised methods for automatic detection of RT using only airway pressure (Paw) and flow. A secondary objective was to describe the magnitude of the efforts generated during RT.

METHODS

We developed algorithms for detection of RT using Paw and flow waveforms. Experts having Paw, flow and esophageal pressure (Pes) assessed automatic detection accuracy by comparison against visual assessment. Muscular pressure (Pmus) was measured from Pes during RT, triggered breaths and ineffective efforts.

RESULTS

Tracings from 20 hypoxemic patients were used (mean age 65 ± 12 years, 65% male, ICU survival 75%). RT was present in 24% of the breaths ranging from 0 (patients paralyzed or in pressure support ventilation) to 93.3%. Automatic detection accuracy was 95.5%: sensitivity 83.1%, specificity 99.4%, positive predictive value 97.6%, negative predictive value 95.0% and kappa index of 0.87. Pmus of RT ranged from 1.3 to 36.8 cmH₂0, with a median of 8.7 cmH₂0. RT with breath stacking had the highest levels of Pmus, and RTs with no breath stacking were of similar magnitude than pressure support breaths.

CONCLUSION

An automated detection tool using airway pressure and flow can diagnose reverse triggering with excellent accuracy. RT generates a median Pmus of 9 cmH₂O with important variability between and within patients.

TRIAL REGISTRATION

BEARDS, NCT03447288.

Computational analysis of neonatal ventilator waveforms and loops

BACKGROUND

Modern neonatal ventilators allow the downloading of their data with a high sampling rate. We wanted to develop an algorithm that automatically recognises and characterises ventilator inflations from ventilator pressure and flow data.

METHODS

We downloaded airway pressure and flow data with 100 Hz sampling rate from Dräger Babylog VN500 ventilators ventilating critically ill infants. We developed an open source Python package, Ventiliser, that includes a rule-based algorithm to automatically discretise ventilator data into a sequence of flow and pressure states and to recognise ventilator inflations and an information gain approach to identify inflation phases (inspiration, expiration) and sub-phases (pressure rise, pressure plateau, inspiratory hold etc.).

RESULTS

Ventiliser runs on a personal computer and analyses 24 h of ventilation in 2 min. With longer recordings, the processing time increases linearly. It generates a table reporting indices of each breath and its sub-phases. Ventiliser also allows visualisation of individual inflations as waveforms or loops. Ventiliser identified >97% of ventilator inflations and their sub-phases in an out-of-sample validation of manually annotated data. We also present detailed quantitative analysis and comparison of two 1-hour-long ventilation periods.

CONCLUSIONS

Ventiliser can analyse ventilation patterns and ventilator-patient interactions over long periods of mechanical ventilation.

IMPACT

We have developed a computational method to recognize and analyse ventilator inflations from raw data downloaded from ventilators of preterm and critically ill infants. There have been no previous reports on the computational analysis of neonatal ventilator data. We have made our program, Ventiliser, freely available. Clinicians and researchers can use Ventiliser to analyse ventilator inflations, waveforms and loops over long periods. Ventiliser can also be used to study ventilator-patient interactions.

Ventilator dyssynchrony - Detection, pathophysiology, and clinical relevance: A Narrative review

Mortality associated with the acute respiratory distress syndrome remains unacceptably high due in part to ventilator-induced lung injury (VILI). Ventilator dyssynchrony is defined as the inappropriate timing and delivery of a mechanical breath in response to patient effort and may cause VILI. Such deleterious patient–ventilator interactions have recently been termed patient self-inflicted lung injury. This narrative review outlines the detection and frequency of several different types of ventilator dyssynchrony, delineates the different mechanisms by which ventilator dyssynchrony may propagate VILI, and reviews the potential clinical impact of ventilator dyssynchrony. Until recently, identifying ventilator dyssynchrony required the manual interpretation of ventilator pressure and flow waveforms. However, computerized interpretation of ventilator waive forms can detect ventilator dyssynchrony with an area under the receiver operating curve of >0.80. Using such algorithms, ventilator dyssynchrony occurs in 3%–34% of all breaths, depending on the patient population. Moreover, two types of ventilator dyssynchrony, double-triggered and flow-limited breaths, are associated with the more frequent delivery of large tidal volumes >10 mL/kg when compared with synchronous breaths (54% [95% confidence interval (CI), 47%–61%] and 11% [95% CI, 7%–15%]) compared with 0.9% (95% CI, 0.0%–1.9%), suggesting a role in propagating VILI. Finally, a recent study associated frequent dyssynchrony-defined as >10% of all breaths-with an increase in hospital mortality (67 vs. 23%, P = 0.04). However, the clinical significance of ventilator dyssynchrony remains an area of active investigation and more research is needed to guide optimal ventilator dyssynchrony management.

A machine learning method for automatic detection and classification of patient-ventilator asynchrony

Patients suffering from respiratory failure are often put on assisted mechanical ventilation. Patient-ventilator asynchrony (PVA) can occur during mechanical ventilation, which cause damage to the lungs and has been linked to increased mortality in the intensive care unit. In current clinical practice PVA is still detected using visual inspection of the air pressure, flow, and volume curves, which is time-consuming and sensitive to subjective interpretation. Correct detection of the patient respiratory efforts is needed to properly asses the type of asynchrony. Therefore, we propose a method for automatic detection of the patient respiratory efforts using a one-dimensional convolution neural network. The proposed method was able to detect patient efforts with a sensitivity and precision of 98.6% and 97.3% for the inspiratory efforts, and 97.7% and 97.2% for the expiratory efforts. Besides allowing detection of PVA, combining the estimated timestamps of patient's inspiratory and expiratory efforts with the timings of the mechanical ventilator further allows for classification of the asynchrony type. In the future, the proposed method could support clinical decision making by informing clinicians on the quality of ventilation and providing actionable feedback for properly adjusting the ventilator settings.

AI in the Intensive Care Unit: Up-to-Date Review

AI is the latest technologic trend that likely will have a huge impact in medicine. AI's potential lies in its ability to process large volumes of data and perform complex pattern analyses. The ICU is an area of medicine that is particularly conducive to AI applications. Much AI ICU research currently is focused on improving high volumes of data on high-risk patients and making clinical workflow more efficient. Emerging topics of AI medicine in the ICU include AI sensors, sepsis prediction, AI in the NICU or SICU, and the legal role of AI in medicine. This review will cover the current applications of AI medicine in the ICU, potential pitfalls, and other AI medicine-related topics relevant for the ICU.

Leveraging IoTs and Machine Learning for Patient Diagnosis and Ventilation Management in the Intensive Care Unit

Future healthcare systems will rely heavily on clinical decision support systems (CDSS) to improve the decision-making processes of clinicians. To explore the design of future CDSS, we developed a research-focused CDSS for the management of patients in the intensive care unit that leverages Internet of Things (IoT) devices capable of collecting streaming physiologic data from ventilators and other medical devices. We then created machine learning (ML) models that could analyze the collected physiologic data to determine if the ventilator was delivering potentially harmful therapy and if a deadly respiratory condition, acute respiratory distress syndrome (ARDS), was present. We also present work to aggregate these models into a mobile application that can provide responsive, real-time alerts of changes in ventilation to providers. As illustrated in the recent COVID-19 pandemic, being able to accurately predict ARDS in newly infected patients can assist in prioritizing care. We show that CDSS may be used to analyze physiologic data for clinical event recognition and automated diagnosis, and we also highlight future research avenues for hospital CDSS.

Applications of artificial intelligence and machine learning in respiratory medicine

The past 5 years have seen an explosion of interest in the use of artificial intelligence (AI) and machine learning techniques in medicine. This has been driven by the development of deep neural networks (DNNs)-complex networks residing in silico but loosely modelled on the human brain-that can process complex input data such as a chest radiograph image and output a classification such as 'normal' or 'abnormal'. DNNs are 'trained' using large banks of images or other input data that have been assigned the correct labels. DNNs have shown the potential to equal or even surpass the accuracy of human experts in pattern recognition tasks such as interpreting medical images or biosignals. Within respiratory medicine, the main applications of AI and machine learning thus far have been the interpretation of thoracic imaging, lung pathology slides and physiological data such as pulmonary function tests. This article surveys progress in this area over the past 5 years, as well as highlighting the current limitations of AI and machine learning and the potential for future developments.

Artificial Intelligence in the Intensive Care Unit

This article is one of ten reviews selected from the Annual Update in Intensive Care and Emergency Medicine 2020. Other selected articles can be found online at https://www.biomedcentral.com/collections/annualupdate2020. Further information about the Annual Update in Intensive Care and Emergency Medicine is available from http://www.springer.com/series/8901.

ADAPTS: An Intelligent Sustainable Conceptual Framework for Engineering Projects

This paper presents a conceptual framework for the optimization of environmental sustainability in engineering projects, both for products and industrial facilities or processes. The main objective of this work is to propose a conceptual framework to help researchers to approach optimization under the criteria of sustainability of engineering projects, making use of current Machine Learning techniques. For the development of this conceptual framework, a bibliographic search has been carried out on the Web of Science. From the selected documents and through a hermeneutic procedure the texts have been analyzed and the conceptual framework has been carried out. A graphic representation pyramid shape is shown to clearly define the variables of the proposed conceptual framework and their relationships. The conceptual framework consists of 5 dimensions; its acronym is ADAPTS. In the base are: (1) the Application to which it is intended, (2) the available DAta, (3) the APproach under which it is operated, and (4) the machine learning Tool used. At the top of the pyramid, (5) the necessary Sensing. A study case is proposed to show its applicability. This work is part of a broader line of research, in terms of optimization under sustainability criteria.

Patient-ventilator asynchronies during mechanical ventilation: current knowledge and research priorities

BACKGROUND

Mechanical ventilation is common in critically ill patients. This life-saving treatment can cause complications and is also associated with long-term sequelae. Patient-ventilator asynchronies are frequent but underdiagnosed, and they have been associated with worse outcomes.

MAIN BODY

Asynchronies occur when ventilator assistance does not match the patient's demand. Ventilatory overassistance or underassistance translates to different types of asynchronies with different effects on patients. Underassistance can result in an excessive load on respiratory muscles, air hunger, or lung injury due to excessive tidal volumes. Overassistance can result in lower patient inspiratory drive and can lead to reverse triggering, which can also worsen lung injury. Identifying the type of asynchrony and its causes is crucial for effective treatment. Mechanical ventilation and asynchronies can affect hemodynamics. An increase in intrathoracic pressure during ventilation modifies ventricular preload and afterload of ventricles, thereby affecting cardiac output and hemodynamic status. Ineffective efforts can decrease intrathoracic pressure, but double cycling can increase it. Thus, asynchronies can lower the predictive accuracy of some hemodynamic parameters of fluid responsiveness. New research is also exploring the psychological effects of asynchronies. Anxiety and depression are common in survivors of critical illness long after discharge. Patients on mechanical ventilation feel anxiety, fear, agony, and insecurity, which can worsen in the presence of asynchronies. Asynchronies have been associated with worse overall prognosis, but the direct causal relation between poor patient-ventilator interaction and worse outcomes has yet to be clearly demonstrated. Critical care patients generate huge volumes of data that are vastly underexploited. New monitoring systems can analyze waveforms together with other inputs, helping us to detect, analyze, and even predict asynchronies. Big data approaches promise to help us understand asynchronies better and improve their diagnosis and management.

CONCLUSIONS

Although our understanding of asynchronies has increased in recent years, many questions remain to be answered. Evolving concepts in asynchronies, lung crosstalk with other organs, and the difficulties of data management make more efforts necessary in this field.

Intensive Care Unit Telemedicine in the Era of Big Data, Artificial Intelligence, and Computer Clinical Decision Support Systems

This article examines the history of the telemedicine intensive care unit (tele-ICU), the current state of clinical decision support systems (CDSS) in the tele-ICU, applications of machine learning (ML) algorithms to critical care, and opportunities to integrate ML with tele-ICU CDSS. The enormous quantities of data generated by tele-ICU systems is a major driver in the development of the large, comprehensive, heterogeneous, and granular data sets necessary to develop generalizable ML CDSS algorithms, and deidentification of these data sets expands opportunities for ML CDSS research.

Clinical intelligence: New machine learning techniques for predicting clinical drug response

Predicting the response, or sensitivity, of a clinical drug to a specific cancer type is an important research problem. By predicting the clinical drug response correctly, clinicians are able to understand patient-to-patient differences in drug sensitivity outcomes, which in turn results in lesser time spent and lower cost associated with identifying effective drug candidates. Although technological advances in high-throughput drug screening in cells led to the generation of a substantial amount of relevant data, the analysis of such data would be a challenging task. There is a critical need for advanced machine learning (ML) algorithms to generate accurate predictions of clinical drug response. A major goal of this work is to provide advanced ML tools to data analysts, who would in turn build prediction calculators to be incorporated into intelligent clinical decision support systems. Such innovative tools could be used to enhance patient-care, among other uses. To achieve this goal, we develop new ML techniques, including a transfer learning approach coupled with or without a boosting technique. Experimental results on real clinical data pertaining to breast cancer, multiple myeloma, and triple-negative cancer patients demonstrate the effectiveness and superiority of the proposed approaches compared to baseline approaches, including existing transfer learning methods.