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

Artificial intelligence in the management of patient-ventilator asynchronies: A scoping review

BACKGROUND

Patient-ventilator asynchronies (PVAs) are frequent complications in mechanically ventilated patients, contributing to adverse outcomes such as ventilator-induced lung injury, prolonged mechanical ventilation, and increased mortality. Artificial intelligence (AI) has emerged as a promising tool for enhancing PVA detection, prediction, and optimization. Despite its growing potential, the full scope of AI applications in this field and persistent gaps in evidence remain inadequately explored.

OBJECTIVE

This scoping review examines the breadth of AI-based approaches for managing PVAs, identifying key methodologies, evaluating research trends, and highlighting limitations in the current literature.

METHODS

A comprehensive search was conducted in PubMed, Embase, Science Direct, IEEE Xplore, and the Cochrane Library without time restrictions. Extracted data included study objectives, AI methodologies, patient populations, performance metrics, and clinical outcomes. The findings were synthesized into thematic categories to map advancements and research gaps.

RESULTS

Twenty-six studies were identified that applied AI techniques to detect, predict, or optimize PVAs. The included studies employed a range of AI methodologies, including convolutional neural networks, long short-term memory networks, and hybrid algorithms. These models demonstrated high predictive performance, with accuracy ranging from 87 % to 99 % and AUROC values exceeding 0.98 for detecting complex asynchronous events. AI systems also showed promise in predicting weaning success and optimizing ventilatory settings through real-time patient-specific adjustments. However, challenges such as reliance on single-center datasets, inconsistencies in calibration, and limited implementation of explainable AI frameworks restrict their clinical applicability.

CONCLUSIONS

AI holds transformative potential in managing PVAs by enabling real-time detection, improved weaning prediction, and personalized ventilatory strategies. However, significant challenges remain, particularly the need for multi-center validation, standardized reporting protocols, and randomized controlled trials to evaluate clinical efficacy. Addressing these gaps is essential for integrating AI into routine critical care practice and transitioning from theoretical models to practical, real-world applications in intensive care units.

Artificial Intelligence for the Detection of Patient-Ventilator Asynchrony

Patient-ventilator asynchrony (PVA) is a challenge to invasive mechanical ventilation characterized by misalignment of ventilatory support and patient respiratory effort. PVA is highly prevalent and associated with adverse clinical outcomes, including increased work of breathing, oxygen consumption, and risk of barotrauma. Artificial intelligence (AI) is a potentially transformative solution offering capabilities for automated detection of PVA. This narrative review characterizes the landscape of AI models designed for PVA detection and quantification. A comprehensive literature search identified 13 studies, spanning diverse settings and patient populations. Machine learning (ML) techniques, derivation datasets, types of asynchronies detected, and performance metrics were assessed to provide a contemporary view of AI in this domain. We reviewed 166 articles published between 1989 and April 2024, of which 13 were included, encompassing 332 participants and analyzing >5.8 million breaths. Patient counts ranged between 8 and 107 and breath data ranged between 1,375 and 4.2 M. The reason for invasive mechanical ventilation use was given as ARDS in three articles, whereas the remainder had different invasive mechanical ventilation indications. Various ML methods as well as newer deep learning techniques were used to address PVA types. Sensitivity and specificity of 10 of the 13 models were >0.9, and 8 models reported accuracy of >0.9. AI models have significant potential to address PVA in invasive mechanical ventilation, displaying high accuracy across various populations and asynchrony types. This showcases their potential to accurately detect and quantify PVA. Future work should focus on model validation in diverse clinical settings and patient populations.

Let's get in sync: current standing and future of AI-based detection of patient-ventilator asynchrony

BACKGROUND

Patient-ventilator asynchrony (PVA) is a mismatch between the patient's respiratory drive/effort and the ventilator breath delivery. It occurs frequently in mechanically ventilated patients and has been associated with adverse events and increased duration of ventilation. Identifying PVA through visual inspection of ventilator waveforms is highly challenging and time-consuming. Automated PVA detection using Artificial Intelligence (AI) has been increasingly studied, potentially offering real-time monitoring at the bedside. In this review, we discuss advances in automatic detection of PVA, focusing on developments of the last 15 years.

RESULTS

Nineteen studies were identified. Multiple forms of AI have been used for the automated detection of PVA, including rule-based algorithms, machine learning and deep learning. Three licensed algorithms are currently reported. Results of algorithms are generally promising (average reported sensitivity, specificity and accuracy of 0.80, 0.93 and 0.92, respectively), but most algorithms are only available offline, can detect a small subset of PVAs (focusing mostly on ineffective effort and double trigger asynchronies), or remain in the development or validation stage (84% (16/19 of the reviewed studies)). Moreover, only in 58% (11/19) of the studies a reference method for monitoring patient's breathing effort was available. To move from bench to bedside implementation, data quality should be improved and algorithms that can detect multiple PVAs should be externally validated, incorporating measures for breathing effort as ground truth. Last, prospective integration and model testing/finetuning in different ICU settings is key.

CONCLUSIONS

AI-based techniques for automated PVA detection are increasingly studied and show potential. For widespread implementation to succeed, several steps, including external validation and (near) real-time employment, should be considered. Then, automated PVA detection could aid in monitoring and mitigating PVAs, to eventually optimize personalized mechanical ventilation, improve clinical outcomes and reduce clinician's workload.

The value of artificial intelligence for the treatment of mechanically ventilated intensive care unit patients: An early health technology assessment

PURPOSE

The health and economic consequences of artificial intelligence (AI) systems for mechanically ventilated intensive care unit patients often remain unstudied. Early health technology assessments (HTA) can examine the potential impact of AI systems by using available data and simulations. Therefore, we developed a generic health-economic model suitable for early HTA of AI systems for mechanically ventilated patients.

MATERIALS AND METHODS

Our generic health-economic model simulates mechanically ventilated patients from their hospitalisation until their death. The model simulates two scenarios, care as usual and care with the AI system, and compares these scenarios to estimate their cost-effectiveness.

RESULTS

The generic health-economic model we developed is suitable for estimating the cost-effectiveness of various AI systems. By varying input parameters and assumptions, the model can examine the cost-effectiveness of AI systems across a wide range of different clinical settings.

CONCLUSIONS

Using the proposed generic health-economic model, investors and innovators can easily assess whether implementing a certain AI system is likely to be cost-effective before an exact clinical impact is determined. The results of the early HTA can aid investors and innovators in deployment of AI systems by supporting development decisions, informing value-based pricing, clinical trial design, and selection of target patient groups.

An adversarial learning approach to generate pressure support ventilation waveforms for asynchrony detection

BACKGROUND AND OBJECTIVE

Mechanical ventilation is a life-saving treatment for critically-ill patients. During treatment, patient-ventilator asynchrony (PVA) can occur, which can lead to pulmonary damage, complications, and higher mortality. While traditional detection methods for PVAs rely on visual inspection by clinicians, in recent years, machine learning models are being developed to detect PVAs automatically. However, training these models requires large labeled datasets, which are difficult to obtain, as labeling is a labour-intensive and time-consuming task, requiring clinical expertise. Simulating the lung-ventilator interactions has been proposed to obtain large labeled datasets to train machine learning classifiers. However, the obtained data lacks the influence of different hardware, of servo-controlled algorithms, and different sources of noise. Here, we propose VentGAN, an adversarial learning approach to improve simulated data by learning the ventilator fingerprints from unlabeled clinical data.

METHODS

In VentGAN, the loss functions are designed to add characteristics of clinical waveforms to the generated results, while preserving the labels of the simulated waveforms. To validate VentGAN, we compare the performance for detection and classification of PVAs when training a previously developed machine learning algorithm with the original simulated data and with the data generated by VentGAN. Testing is performed on independent clinical data labeled by experts. The McNemar test is applied to evaluate statistical differences in the obtained classification accuracy.

RESULTS

VentGAN significantly improves the classification accuracy for late cycling, early cycling and normal breaths (p< 0.01); no significant difference in accuracy was observed for delayed inspirations (p = 0.2), while the accuracy decreased for ineffective efforts (p< 0.01).

CONCLUSIONS

Generation of realistic synthetic data with labels by the proposed framework is feasible and represents a promising avenue for improving training of machine learning models.

Flow starvation during square-flow assisted ventilation detected by supervised deep learning techniques

BACKGROUND

Flow starvation is a type of patient-ventilator asynchrony that occurs when gas delivery does not fully meet the patients' ventilatory demand due to an insufficient airflow and/or a high inspiratory effort, and it is usually identified by visual inspection of airway pressure waveform. Clinical diagnosis is cumbersome and prone to underdiagnosis, being an opportunity for artificial intelligence. Our objective is to develop a supervised artificial intelligence algorithm for identifying airway pressure deformation during square-flow assisted ventilation and patient-triggered breaths.

METHODS

Multicenter, observational study. Adult critically ill patients under mechanical ventilation > 24 h on square-flow assisted ventilation were included. As the reference, 5 intensive care experts classified airway pressure deformation severity. Convolutional neural network and recurrent neural network models were trained and evaluated using accuracy, precision, recall and F1 score. In a subgroup of patients with esophageal pressure measurement (ΔPes), we analyzed the association between the intensity of the inspiratory effort and the airway pressure deformation.

RESULTS

6428 breaths from 28 patients were analyzed, 42% were classified as having normal-mild, 23% moderate, and 34% severe airway pressure deformation. The accuracy of recurrent neural network algorithm and convolutional neural network were 87.9% [87.6-88.3], and 86.8% [86.6-87.4], respectively. Double triggering appeared in 8.8% of breaths, always in the presence of severe airway pressure deformation. The subgroup analysis demonstrated that 74.4% of breaths classified as severe airway pressure deformation had a ΔPes > 10 cmH2O and 37.2% a ΔPes > 15 cmH2O.

CONCLUSIONS

Recurrent neural network model appears excellent to identify airway pressure deformation due to flow starvation. It could be used as a real-time, 24-h bedside monitoring tool to minimize unrecognized periods of inappropriate patient-ventilator interaction.

Automated characterization of patient-ventilator interaction using surface electromyography

BACKGROUND

Characterizing patient-ventilator interaction in critically ill patients is time-consuming and requires trained staff to evaluate the behavior of the ventilated patient.

METHODS

In this study, we recorded surface electromyography ([Formula: see text]) signals from the diaphragm and intercostal muscles and esophageal pressure ([Formula: see text]) in mechanically ventilated patients with ARDS. The sEMG recordings were preprocessed, and two different algorithms (triangle algorithm and adaptive thresholding algorithm) were used to automatically detect inspiratory patient effort. Based on the detected inspirations, major asynchronies (ineffective, auto-, and double triggers and double efforts), delayed and synchronous triggers were computationally classified. Reverse triggers were not considered in this study. Subsequently, asynchrony indices were calculated. For the validation of detected efforts, two experts manually annotated inspiratory patient activity in [Formula: see text], blinded toward each other, the [Formula: see text] signals, and the algorithmic results. We also classified patient-ventilator interaction and calculated asynchrony indices with manually detected inspirations in [Formula: see text] as a reference for automated asynchrony classification and asynchrony index calculation.

RESULTS

Spontaneous breathing activity was recognized in 22 out of the 36 patients included in the study. Evaluation of the accuracy of the algorithms using 3057 inspiratory efforts in [Formula: see text] demonstrated reliable detection performance for both methods. Across all datasets, we found a high sensitivity (triangle algorithm/adaptive thresholding algorithm: 0.93/0.97) and a high positive predictive value (0.94/0.89) against expert annotations in [Formula: see text]. The average delay of automatically detected inspiratory onset to the [Formula: see text] reference was [Formula: see text]79 ms/29 ms for the two algorithms. Our findings also indicate that automatic asynchrony index prediction is reliable. For both algorithms, we found the same deviation of [Formula: see text] to the [Formula: see text]-based reference.

CONCLUSIONS

Our study demonstrates the feasibility of automating the quantification of patient-ventilator asynchrony in critically ill patients using noninvasive sEMG. This may facilitate more frequent diagnosis of asynchrony and support improving patient-ventilator interaction.