Background: The diaphragm is poorly monitored in the Intensive care unit (ICU), despite its evident importance in respiration. Ultrasound (US) is frequently employed to evaluate diaphragm thickness (DT) and diaphragm thickening factor (DTF) but requires expertise and only covers a small diaphragm area. Therefore, advanced US (post-)processing techniques are being investigated for diaphragm applications, including speckle tracking methods. Objective: The primary aim of this pilot study was to develop an algorithm enabling DT quantification. Second, both an existing Fourier-based (FBST) and intensity-based speckle tracking (IBST) algorithm were modified to determine their feasibility in diaphragm strain quantification. Results: Minimal (DTmin, mm) and maximal DT (DTmax, mm), and DTF (%) values were 1.9 ± 0.4, 2.3 ± 0.5, and 22.6 ± 10.9 in the MV patient group, and 2.2 ± 0.4, 3.7 ± 1.5 and 66.5 ± 45.0 in the volunteer group (all p > 0.05). Correlation of DT algorithm variables and manual expert grading were allsignificant, with a good correlation in DTmax (ICC 0.9, r or ρ 0.7, p < 0.001) and a moderate correlation in DTmin and DTF (ICC 0.7, r or ρ 0.7, p < 0.001 and p = 0.001, respectively). Inter- and intra-rater reproducibility of manual DT assessment was poor (ICC < 0.4 and r or ρ < 0.2). GLS (%) and GLSR (%/s) values were -32.0 ± 19.8 and -6.6 ± 3.8 in the patient group and -40.3 ± 17.3 and -10.4 ± 6.8 in the volunteer group, respectively (all p > 0.05). IBST scores were 26.2 ± 17.6 in the patient group and 68.1 ± 32.5 in the volunteer group (p = 0.009). IBST score values showed a good correlation with DTF (r or ρ 0.7, p = 0.003), and a moderate, negative correlation with DTmin (r or ρ -0.5, p = 0.043). No significant correlations were seen between the remaining manual expert DT assessment and algorithm-derived FBST and IBST variables.Conclusion: DT quantification using the algorithm developed during this study correlated to conventional US expert assessment. Poor reproducibility of current diaphragm function quantification supports the need for such an automated assessment. The clinical value of diaphragm strain assessment using the modified FBST and IBST algorithms remains unclear. Further research involving a widely defined gold standard technique in diaphragm function quantification is warranted.

Critically ill, mechanically ventilated patients require a substantial amount of energy to meet their basic metabolic demands, as well as for their recovery, spontaneous breathing, gaining of strength, and weaning from the ventilator. It is important for patients to use their energy as efficient as possible at all stages of the disease and recovery. Physicians need to estimate and match the energy cost of the treatment regimen with the patient’s ability for energy expenditure. If the demands of the treatment regimen, such as weaning from the ventilator, are too high, it will lead to exertion and fatigue. On the other hand, if the patient is not challenged enough, it could lead to a slower recovery and prolonged admission to the Intensive Care Unit (ICU). For example, it is known that a too slow reduction in ventilator support can lead to atrophy of the diaphragm, resulting in prolonged ventilation time. Predicting successful extubation is crucial for mechanically ventilated ICU patients. Prolonged intubation leads to prolonged mechanical ventilation, which is known to have adverse effects. Up to 20 % of all ICU patients fail to wean from mechanical ventilation. Therefore, it is crucial to investigate different parameters of effort. This will be elaborated on in Chapter 1 of the thesis. The oxygen consumption seems to be a good new parameter to add in standard of care. To measure oxygen consumption (VO2) breath-by-breath, it is necessary to synchronize flow and oxygen concentration data. To create this algorithm, it was necessary to design an in vitro test setup that correctly mimics the mechanically ventilated ICU patient. This setup and the plan of requirements is described in Chapter 2. In the test setup a cylinder was used with N2 and CO2 to simulate oxygen consumption and CO2 production. The algorithm developed using the data generated with the test setup using the Hamilton Mechanical Ventilator C6 flow data and the Masimo ISA OR+ oxygen concentration data, is used to synchronize the data from the two devices and to determine the VO2. As the rise time of the oxygen sensor in the Masimo was slow, this signal needed some adaptions. Furthermore, the time stamps were not the same, as with the sample frequency. After the synchronization, the VO2 was derived by taking the integral of the two synchronized signals per breath. These values were compared to predicted values, derived from breath-by-breath parameter data of the two devices. All the preprocessing, synchronization and eventual oxygen consumption measurements are described in Chapter 3. As there were too many assumptions in the design of the in vitro test setup, it was necessary to create a new setup with minimized number of assumptions. This new test setup had two extra flow sensors in order to monitor all leaks and flows. Next to the more reliable data from the setup, some simulation data was created using Python. Both data were used to validate the algorithm. These steps were explained in Chapter 4. In Chapter 5 some perspectives for the future were explained. This includes the integration of the VO2 in clinical practice and the steps that still need to be taken to come that far. These include testing on patient data and performing more clinical trials.

Introduction: In intensive care units (ICU), the most significant life support technology for patients with acute respiratory failure is mechanical ventilation. A mismatch between ventilatory support and patient demand is referred to as patient-ventilator asynchrony (PVA), and it is associated with a series of adverse clinical outcomes. Although the use of a reference signal for patient effort is critical in recognition of PVA, existing detection algorithms are frequently solely based on the ventilator’s airway pressure (Paw) and flow-time signals. The aim of this study was to develop an automated detection algorithm for PVA using the ventilator’s Paw, flow-time and esophageal pressure (Pes) signals. Methods: We proposed a two-dimensional convolutional neural network (2DCNN) to detect two types of PVA (reverse triggering (RT) and premature cycling) using a dataset of respiratory cycles recorded from 11 patients. Mechanical ventilation experts with access to the Pes signal annotated 12.337 respiratory cycles to create a gold standard dataset. Several techniques for a potential class imbalance problem, as well as several changes to the initial model architecture, were investigated. A leave-one-patient-out cross-validation technique was used to evaluate model performance. The proposed Pes-based 2DCNN (Pes_2DCNN) was compared to a similar model based solely on the ventilator Paw and flow-time signals (PF_2DCNN). Results: The proposed Pes_2DCNN exhibited superior performance in detecting RT as compared to PF_2DCNN in terms of area under the receiver operating characteristic (AUROC) (0.80 ± 0.07 vs. 0.75 ± 0.13, respectively; p < 0.01). Furthermore, the results indicate that the class imbalance solutions did not improve the performance for detection of RT. For detection of premature cycling, Pes_2DCNN also outperformed PF_2DCNN in terms of AUROC (0.88 ± 0.09 vs. 0.71 ± 0.24, respectively; p < 0.01). Conclusion: The findings of this study suggest the added value of the Pes signal in detection of RT and premature cycling. However, because this is a preliminary study, more research is required to further investigate the importance of the Pes signal in PVA detection.

Introduction Patient-ventilator asynchrony (PVA) poses a significant challenge in the management of mechanically ventilated patients, contributing to adverse clinical outcomes. Current methods of detecting PVA rely on visual assessment by clinicians, leading to subjectivity and inconsistency. Therefore, there is a need for automated techniques to identify PVA accurately and efficiently. In this study, we explore the application of supervised and unsupervised machine learning algorithms to develop an automatic detection system for PVA. Methods This study was conducted at the ICU of the LUMC in Leiden, the Netherlands. Patients eligible for inclusion were mechanically ventilated with an esophageal balloon inserted. Data collected included flow, Paw, and Pes curves, which were labelled using an open-source data labeling platform and processed in Python. Supervised CNN models were trained for different ventilation modes, while unsupervised techniques, utilizing Mahalanobis distance, were explored for data pre-labeling. The discriminative capability of the models was assessed using AUROC values. Results 25 patients were included in this study and labelled by clinicians. Using an unsupervised machine learning technique based on the Mahalanobis distance for data pre-labeling, a threshold of 3.5 was selected, resulting in a 95% accuracy in correctly identifying normal breaths. Creating different CNN models for automating the detection of PVA the results demonstrate the discriminative capability of the various models across all ventilation modes, PSV and PCV ventilation. They can differentiate between normal and abnormal breaths, as indicated by the AUROC values of 0.85(±0.08), 0.83 (±0.12), and 0.80 (±0.28) respectively. Discussion This study investigated the application of machine learning techniques to analyse ventilation data in critical care settings. Through a combination of unsupervised and supervised learning methods, we have explored the automation of the labelling process and the development of predictive models for identifying patient-ventilator asynchronies (PVAs).