A Nonlinear Hysteretic Model for Automated Prediction of Lung Mechanics during Mechanical Ventilation. Issue 5 (2020)

• Record Type: Journal Article
• Authors: Zhou, Cong; Chase, J. Geoffrey; Sun, Qianhui; Knopp, Jennifer
• Is Part Of: IFAC-PapersOnLine. Volume 53:Issue 5(2020)
• Page Start: 817
• Page End: 822
• Publication Date: 2020
• Subjects: Nonlinear Modelling -- Hysteresis Model -- Hysteresis loop analysis -- Virtual Patient -- Mechanical Ventilation -- Lung mechanics; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.04.177
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Mechanical ventilation (MV) is core intensive care unit (ICU) therapy during the Covid-19 pandemic. Optimising MV care to a specific patient with respiratory failure is difficult due to inter- and intra- patient variability in lung mechanics and condition. The ability to accurately predict patient-specific lung response to a change in MV settings would enable semi-automated care and significantly improve the efficiency of MV monitoring and care. It has particular emphasis when considering MV care required to treat Covid-19 patients, who require longer MV care, where patient-specific care can reduce the time on MV required. This study develops a nonlinear smooth hysteresis loop model (HLM) able to capture the essential lung dynamics in a patient-specific fashion from measured ventilator data, particularly for changes of compliance and infection points of the pressure-volume loop. The automated (no human input) hysteresis loop analysis (HLA) method is applied to identify HLM model parameters, enabling automated digital cloning to create a virtual patient model to accurately predict lung response at a specified positive end expiratory pressure (PEEP) level, as well as in response to the changes of PEEP. The performance of this automated digital cloning approach is assessed using clinical data from 4 patients and 8 recruitment maneuver (RM) arms. Validation results show the HLM-based hysteresis loops identified using HLA match clinical pressure-volume loops very well with root-mean-square (RMS) errors less than 2% for all 8 data sets over 4 patients, validating the accuracy of the developed HLM in capturing the essential lung physiology and respiratory behaviours at different patient conditions. More importantly, the patient-specific digital clones at lower PEEP levels accurately predict lung response at higher PEEP levels with predicted peak inspiratory pressure (PIP) errors less than 2% in average. In addition, the resulted additional lung volume Vfrc obtained with PEEP changes are predicted with average absolute difference of 0.025L. The overall results validate the versatility and potential of the developed HLM for delineating changes of nonlinear lung dynamics, and its capability to create a predictive virtual patient with use of HLA for future treatment personalization and optimisation in MV therapy.