The following is a summary of ‘’Machine learning predicts lung recruitment in acute respiratory distress syndrome using single lung CT Scan,” published in the July 2023 issue of Critical Care by Pennati et al.
Researchers performed a retrospective study to develop and validate classifier models to identify patients with a high percentage of potentially recruitable lungs based on readily available clinical data and a single CT scan analysis at intensive care unit admission. The study involved 221 patients with acute respiratory distress syndrome (ARDS) who were mechanically ventilated, sedated, and paralyzed. A positive end-expiratory pressure (PEEP) trial was conducted at two levels: 5 cmH2O and 15 cmH2O. Patients underwent two lung CT scans, one at 5 cmH2O and another at 45 cmH2O of airway pressure. Lung recruitability was defined based on two criteria: the percent change in non-aerated tissue between 5 and 45 cmH2O (radiologically defined; recruiters: Δ45-5 non-aerated tissue > 15%), and the change in PaO2 between 5 and 15 cmH2O (gas exchange-defined; recruiters: Δ15-5PaO2 > 24 mmHg). Four ML algorithms were tested as classifiers for identifying lung recruiters based on radiological and gas exchange criteria, using different models with varied combinations of lung mechanics, gas exchange, and CT data variables.
ML algorithms using CT scan data at 5 cmH2O performed similarly to those combining multiple variables in classifying radiological recruiters. However, the CT-based algorithm achieved the highest area under the curve (AUC) in classifying gas exchange-defined recruiters.
The study concluded that a single CT data at 5 cmH2O, ML proved to be a convenient tool for classifying ARDS patients as recruiters or non-recruiters based on both radiological and gas exchange criteria within the initial 48 hours of mechanical ventilation.
Source: annalsofintensivecare.springeropen.com/articles/10.1186/s13613-023-01154-5