Assessment of respiratory system mechanics by artificial neural networks: an exploratory study

Patient–Ventilator Interaction Testing Using the Electromechanical Lung Simulator xPULM™ during V/A-C and PSV Ventilation Mode

During mechanical ventilation, a disparity between flow, pressure and volume demands of the patient and the assistance delivered by the mechanical ventilator often occurs. This paper introduces an alternative approach of simulating and evaluating patient–ventilator interactions with high fidelity using the electromechanical lung simulator xPULM™. The xPULM™ approximates respiratory activities of a patient during alternating phases of spontaneous breathing and apnea intervals while connected to a mechanical ventilator. Focusing on different triggering events, volume assist-control (V/A-C) and pressure support ventilation (PSV) modes were chosen to test patient–ventilator interactions. In V/A-C mode, a double-triggering was detected every third breathing cycle, leading to an asynchrony index of 16.67%, which is classified as severe. This asynchrony causes a significant increase of peak inspiratory pressure (7.96 ± 6.38 vs. 11.09 ± 0.49 cmH2O, p < 0.01)) and peak expiratory flow (−25.57 ± 8.93 vs. 32.90 ± 0.54 L/min, p < 0.01) when compared to synchronous phases of the breathing simulation. Additionally, events of premature cycling were observed during PSV mode. In this mode, the peak delivered volume during simulated spontaneous breathing phases increased significantly (917.09 ± 45.74 vs. 468.40 ± 31.79 mL, p < 0.01) compared to apnea phases. Various dynamic clinical situations can be approximated using this approach and thereby could help to identify undesired patient–ventilation interactions in the future. Rapidly manufactured ventilator systems could also be tested using this approach.

Analytics with artificial intelligence to advance the treatment of acute respiratory distress syndrome

Artificial intelligence (AI) has found its way into clinical studies in the era of big data. Acute respiratory distress syndrome (ARDS) or acute lung injury (ALI) is a clinical syndrome that encompasses a heterogeneous population. Management of such heterogeneous patient population is a big challenge for clinicians. With accumulating ALI datasets being publicly available, more knowledge could be discovered with sophisticated analytics. We reviewed literatures with big data analytics to understand the role of AI for improving the caring of patients with ALI/ARDS. Many studies have utilized the electronic medical records (EMR) data for the identification and prognostication of ARDS patients. As increasing number of ARDS clinical trials data is open to public, secondary analysis on these combined datasets provide a powerful way of finding solution to clinical questions with a new perspective. AI techniques such as Classification and Regression Tree (CART) and artificial neural networks (ANN) have also been successfully used in the investigation of ARDS problems. Individualized treatment of ARDS could be implemented with a support from AI as we are now able to classify ARDS into many subphenotypes by unsupervised machine learning algorithms. Interestingly, these subphenotypes show different responses to a certain intervention. However, current analytics involving ARDS have not fully incorporated information from omics such as transcriptome, proteomics, daily activities and environmental conditions. AI technology is assisting us to interpret complex data of ARDS patients and enable us to further improve the management of ARDS patients in future with individual treatment plans.

Robustness of two different methods of monitoring respiratory system compliance during mechanical ventilation

Robustness measures the performance of estimation methods when they work under non-ideal conditions. We compared the robustness of artificial neural networks (ANNs) and multilinear fitting (MLF) methods in estimating respiratory system compliance (C_RS) during mechanical ventilation (MV). Twenty-four anaesthetized pigs underwent MV. Airway pressure, flow and volume were recorded at fixed intervals after the induction of acute lung injury. After consecutive mechanical breaths, an inspiratory pause (BIP) was applied in order to calculate C_RS using the interrupter technique. From the breath preceding the BIP, ANN and MLF had to compute C_RS in the presence of two types of perturbations: transient sensor disconnection (TD) and random noise (RN). Performance of the two methods was assessed according to Bland and Altman. The ANN presented a higher bias and scatter than MLF during the application of RN, except when RN was lower than 2% of peak airway pressure. During TD, MLF algorithm showed a higher bias and scatter than ANN. After the application of RN, ANN and MLF maintain a stable performance, although MLF shows better results. ANNs have a more stable performance and yield a more robust estimation of C_RS than MLF in conditions of transient sensor disconnection.

Monitoring of total positive end-expiratory pressure during mechanical ventilation by artificial neural networks

Ventilation treatment of acute lung injury (ALI) requires the application of positive airway pressure at the end of expiration (PEEP_app) to avoid lung collapse. However, the total pressure exerted on the alveolar walls (PEEP_tot) is the sum of PEEP_app and intrinsic PEEP (PEEP_i), a hidden component. To measure PEEP_tot, ventilation must be discontinued with an end-expiratory hold maneuver (EEHM). We hypothesized that artificial neural networks (ANN) could estimate the PEEP_tot from flow and pressure tracings during ongoing mechanical ventilation. Ten pigs were mechanically ventilated, and the time constant of their respiratory system (τ_RS) was measured. We shortened their expiratory time (TE) according to multiples of τ_RS, obtaining different respiratory patterns (R_pat). Pressure (P_AW) and flow (V'_AW) at the airway opening during ongoing mechanical ventilation were simultaneously recorded, with and without the addition of external resistance. The last breath of each R_pat included an EEHM, which was used to compute the reference PEEP_tot. The entire protocol was repeated after the induction of ALI with i.v. injection of oleic acid, and 382 tracings were obtained. The ANN had to extract the PEEP_tot, from the tracings without an EEHM. ANN agreement with reference PEEP_tot was assessed with the Bland-Altman method. Bland Altman analysis of estimation error by ANN showed -0.40 ± 2.84 (expressed as bias ± precision) and ±5.58 as limits of agreement (data expressed as cmH₂O). The ANNs estimated the PEEP_tot well at different levels of PEEP_app under dynamic conditions, opening up new possibilities in monitoring PEEP_i in critically ill patients who require ventilator treatment.

Machine learning algorithms and forced oscillation measurements applied to the automatic identification of chronic obstructive pulmonary disease

The purpose of this study is to develop a clinical decision support system based on machine learning (ML) algorithms to help the diagnostic of chronic obstructive pulmonary disease (COPD) using forced oscillation (FO) measurements. To this end, the performances of classification algorithms based on Linear Bayes Normal Classifier, K nearest neighbor (KNN), decision trees, artificial neural networks (ANN) and support vector machines (SVM) were compared in order to the search for the best classifier. Four feature selection methods were also used in order to identify a reduced set of the most relevant parameters. The available dataset consists of 7 possible input features (FO parameters) of 150 measurements made in 50 volunteers (COPD, n = 25; healthy, n = 25). The performance of the classifiers and reduced data sets were evaluated by the determination of sensitivity (Se), specificity (Sp) and area under the ROC curve (AUC). Among the studied classifiers, KNN, SVM and ANN classifiers were the most adequate, reaching values that allow a very accurate clinical diagnosis (Se > 87%, Sp > 94%, and AUC > 0.95). The use of the analysis of correlation as a ranking index of the FOT parameters, allowed us to simplify the analysis of the FOT parameters, while still maintaining a high degree of accuracy. In conclusion, the results of this study indicate that the proposed classifiers may contribute to easy the diagnostic of COPD by using forced oscillation measurements.

ASSESSMENT AND CLASSIFICATION OF MECHANICAL STRENGTH COMPONENTS OF HUMAN FEMUR TRABECULAR BONE USING DIGITAL IMAGE PROCESSING AND NEURAL NETWORKS

In this work, the assessment of the mechanical strength of human femur trabecular bone and its classification into normal or abnormal are carried out using digital image processing and neural networks. The mechanical strength components of femur trabeculae, such as primary compressive (PC), primary tensile (PT), secondary tensile (ST), and Ward's triangle (WT), are delineated by the semiautomatic image processing procedure from the planar radiographic images (N = 90) of subjects that are acquired under controlled clinical settings. Parameters such as apparent mineralization and total area of the individual mechanical strength components are calculated for normal and abnormal samples. The data are trained with neural networks and validated. The classifications are carried out using feed-forward neural networks trained with the standard backpropagation algorithm. The abnormal and normal outputs are validated by sensitivity and specificity measurements. The observation shows that the investigation of bone mechanical strength at the various strength components is useful in classifying normal and abnormal human femur trabeculae from conventional radiographs. Furthermore, the results confirm the effectiveness of the neural network–based classification of femur trabeculae into normal and abnormal conditions. The sensitivity and specificity are found to be 100% and 80%, respectively. In this paper, the methodology, data collection procedures, and neural network–based analysis and results are discussed in detail.

EXPERIMENTAL ANALYSIS ON HUMAN RESPIRATORY DYNAMICS USING FLOW-VOLUME SPIROMETRY AND COMBINED NEURAL NETWORKS

In this work, detection of pulmonary abnormalities carried out using flow-volume spirometry and combined neural network is presented. The respiratory data (N = 225) were obtained from volunteers using a commercially available spirometer recorded by a standard acquisition protocol. The spirometric data were used for classification of normal, restrictive, and obstructive abnormalities using combined neural networks. The first level of the networks was implemented to assess the primary abnormality, whereas the second level of the networks further classified the degree of abnormality. The results were validated with the measured values of accuracy, sensitivity, specificity, and adjusted accuracy. Combined neural networks using a radial basis algorithm were found to be effective in classifying various degrees of abnormalities as normal, restrictive, or obstructive. Furthermore, combined neural networks achieved significant improvement in accuracy compared to stand-alone neural networks. It appears that this method of integrated assessment is useful in understanding pulmonary function dynamics with incomplete data and/or data with poor recording.

NEURAL NETWORK–BASED CLASSIFICATION OF NORMAL AND ABNORMAL PULMONARY FUNCTION USING SPIROMETRIC MEASUREMENTS

In this work, the classification of pulmonary function into normal and abnormal conditions is attempted using neural networks and spirometric measurements. The pulmonary function data (N = 229) for this study were obtained from volunteers using commercially available spirometers by adopting standard data acquisition protocol and recording conditions. The data were then subjected to neural network–based training (N = 159) and analysis (N = 70). The classification was carried out using standard feed-forward neural network and backpropagation algorithm. The outputs were then validated through sensitivity and specificity measurements together with clinical observation. The results confirmed the effectiveness of the neural network–based classification of spirometric data into normal and abnormal conditions. The sensitivity and specificity were found to be 89.25% and 82.25%, respectively. Furthermore, it seems that this method is useful in assessing the pulmonary function dynamics in cases with incomplete data and data with poor recordings. In this paper, the methodology, data collection procedure, and neural network–based analysis and results are described in detail.

Automatic identification of Chronic Obstructive Pulmonary Disease Based on forced oscillation measurements and artificial neural networks

The purpose of this study is to develop an automatic classifier based on Artificial Neural Networks (ANNs) to help the diagnostic of Chronic Obstructive Pulmonary Disease (COPD) using forced oscillation measurements (FOT). The classifier inputs are the parameters provided by the FOT and the output is the indication if the parameters indicate COPD or not. The available dataset consists of 7 possible input features (FOT parameters) of 90 measurements made in 30 volunteers. Two feature selection methods (the analysis of the linear correlation and forward search) were used in order to identify a reduced set of the most relevant parameters. Two different training strategies for the ANNs were used and the performance of resulting networks were evaluated by the determination of accuracy, sensitivity (Se), specificity (Sp) and AUC. The ANN classifiers presented high accuracy (Se > 0.9, Se > 0.9 and AUC > 0.9) both in the complete and the reduce sets of FOT parameters. This indicates that ANNs classifiers may contribute to easy the diagnostic of COPD using forced oscillation measurements.

Prediction of forced expiratory volume in pulmonary function test using radial basis neural networks and k-means clustering

In this work, prediction of forced expiratory volume in pulmonary function test, carried out using spirometry and neural networks is presented. The pulmonary function data were recorded from volunteers using commercial available flow volume spirometer in standard acquisition protocol. The Radial Basis Function neural networks were used to predict forced expiratory volume in 1 s (FEV1) from the recorded flow volume curves. The optimal centres of the hidden layer of radial basis function were determined by k-means clustering algorithm. The performance of the neural network model was evaluated by computing their prediction error statistics of average value, standard deviation, root mean square and their correlation with the true data for normal, restrictive and obstructive cases. Results show that the adopted neural networks are capable of predicting FEV1 in both normal and abnormal cases. Prediction accuracy was more in obstructive abnormality when compared to restrictive cases. It appears that this method of assessment is useful in diagnosing the pulmonary abnormalities with incomplete data and data with poor recording.

Evaluation of flow-volume spirometric test using neural network based prediction and principal component analysis

In this work, an attempt has been made to enhance the diagnostic relevance of spirometric pulmonary function test using neural networks and Principal Component Analysis (PCA). For this study, flow-volume curves (N = 175) using spirometers were generated under standard recording protocol. A method based on neural network is used to predict the most significant parameter, FEV(1). Further, PCA is used to analyze the interdependency of the parameters in the measured and predicted datasets. Results show that the back propagation neural network is able to predict FEV(1) both in normal and abnormal cases. The variation in the magnitude and direction of parameters in the contribution of the principal components shows that FEV(1) is a significant discriminator of normal and abnormal datasets and is further confirmed by the percentage variance in the first few principal components. It appears that this method of prediction and principal component analysis on the measured and predicted datasets could be useful for spirometric pulmonary function test with incomplete data.

Assessment and classification of mechanical strength components of human femur trabecular bone using texture analysis and neural network

In this work the mechanical strength components of human femur trabecular bone are analyzed and classified using planar radiographic images and neural network. The mechanical strength regions such as Primary Compressive, Primary Tensile, Secondary Tensile and Ward Triangle in femur trabecular bone images (N = 100) are delineated by semi-automatic image processing procedure. First and higher order texture parameters and parameters such as apparent mineralization and total area associated with the strength regions are derived for normal and abnormal images. The statistically derived significant parameters corresponding to the primary strength regions are fed to the neural network for training and validation. The classifications are carried out using feed forward network that is trained with standard back propagation algorithm. Results demonstrate that the apparent mineralization of normal samples is always high as (71%) compared to abnormal samples (64%). Entropy shows a high value (7.3) for normal samples and variation between the mean intensity and apparent mineralization for the primary strength zone is statistically significant (p < 0.0005). The classified outputs are validated by sensitivity and specificity measurements and are found to be 66.66% and 80% respectively. Further it appears that it is possible to differentiate normal and abnormal samples from the conventional radiographic images. As trabecular architecture in the human femur is an important factor contributing to bone strength, the procedure adopted here could be a useful supplement to the clinical observations for bone loss and fracture risk.

Estimation of errors in determining intrathoracic blood volume using thermal dilution in pigs with acute lung injury and haemorrhage †

BACKGROUND

Global end diastolic volume (GEDV) has a constant and predictable relationship to intrathoracic blood volume (ITBV). The present study assesses the difference between ITBV derived from GEDV and ITBV measured directly in pigs with acute lung injury (ALI) and mild haemorrhage.

METHODS

We caused ALI in 12 anaesthetized pigs by i.v. injection of oleic acid and removed 10% of estimated blood volume. EVLW, GEDV, ITBV (COLD; Pulsion Medical Systems), Pa(o(2))/Fi(o(2)), lung compliance and haemodynamic variables were measured at baseline (time 0) and at 30 and 120 min. All animals were volume-resuscitated, followed by measurements at 180 min. A linear equation estimated from the 44 pairs of ITBV and GEDV values in 11 animals was applied iteratively to the four GEDV measurements in the 12th animal, enabling 48 comparisons between measured (ITBVm) and derived ITBV (ITBVd) to be made.

RESULTS

Increase in extravascular lung water index (EVLWi) was associated with significant pulmonary hypertension, worsening of oxygenation and compliance (repeated measures ANOVA; P<0.05). There was good within-subject correlation and agreement between ITBV(m) and ITBV(d) (r=0.72, mean bias 0.8 ml; sd 32 ml). Mean error in deriving ITBV from GEDV was 4.5%. (sd 4.2%; range 0.05-19%). There were no significant differences in errors in the presence of small (up to 10%) deficits in blood volume (F=1.0; P=0.41).

CONCLUSIONS

ITBV estimated by thermodilution alone is comparable to measurements made by the thermo-dye dilution technique in the presence of pulmonary hypertension and mild deficits in total blood volume.

Estimating respiratory system compliance during mechanical ventilation using artificial neural networks

In this study we evaluated whether a technology based on artificial neural networks (ANN) could estimate the static compliance (C(RS)) of the respiratory system, even in the absence of an end-inspiratory pause, during continuous mechanical ventilation. A porcine model of acute lung injury was used to provide recordings of different respiratory mechanics conditions. Each recording consisted of 10 or more consecutive breaths in volume-controlled mechanical ventilation, followed by a breath having an end-inspiratory pause used to calculate C(RS) according to the interrupter technique (IT). The volume-pressure loop of the breath immediately preceding the one with pause was given to the ANN for the training, together with the C(RS) separately calculated by the IT. The prospective phase consisted of giving only the loops to the trained ANN and comparing the results yielded by it to the compliance separately calculated by the investigators. Determination of measurement agreement between ANN and IT methods showed an error of -0.67 +/- 1.52 mL/cm H(2)O (bias +/- SD). We could conclude that ANN, during volume-controlled mechanical ventilation, can extract C(RS) without needing to stop inspiratory flow.

IMPLICATIONS

We studied the application of artificial neural networks (ANN) to the estimation of respiratory compliance during mechanical ventilation. The study was performed on an animal model of acute lung injury, testing the performance of ANN in both healthy and diseased conditions of the lung.

Effects of positive end-expiratory pressure increments can be predicted by computer simulation based on a physiological profile in acute respiratory failure

OBJECTIVE

We examined whether computer simulation predicts airway pressures after increments of positive end-expiratory pressure (PEEP) in acute respiratory failure.

DESIGN AND SETTING

Prospective, nonrandomized comparative trial in an intensive care unit of a university hospital.

PATIENTS

Twelve consecutive acute respiratory failure patients. INTERVENTIONS. PEEP increments from 0 to 2.5, 5, 7.5, 10, and 15 cm H(2)O.

MEASUREMENTS AND RESULTS

A physiological profile comprising values for compliance, respiratory resistance and CO(2 )elimination as a function of tidal volume was established from a recording of ordinary breaths prior to increments of PEEP. Airway pressures and CO(2 )elimination were measured 30 s after resetting, pressures also after 10 min. Values from simulation of the resetting, based on the profile, were compared to measured values. The profiles indicated vast differences in physiology between the 12 subjects. Errors of simulation of airway pressures were nonsignificant or trivial up to PEEP levels of 10 cm H(2)O (95% of errors <3 cm H(2)O). After 10 min plateau pressure averaged 1.5 cm H(2)O lower than 30 s after resetting. At increments to PEEP 7.5, 10, and 15, CO(2 )elimination fell by on average 4%, 8%, and 11%, respectively. As tidal volume and respiratory rate was unchanged this was not predicted.

CONCLUSIONS

On the basis of a simple lung model, simulation predicted effects of moderate increments of PEEP on airway pressures in patients with complex physiology.