The following is a summary of “Phenotyping COVID-19 respiratory failure in spontaneously breathing patients with AI on lung CT-scan,” published in the August 2024 issue of Critical Care by Rezoagli et al.
Automated analysis of lung computed tomography (CT) scans could help uncover distinct subphenotypes of acute respiratory conditions.
Researchers conducted a retrospective study combining lung CT features from deep learning with clinical and lab data to better identify COVID-19 subphenotypes in spontaneously breathing patients (SBP).
They focused on SBP with COVID-19 respiratory failure who underwent early lung CT within 7 days of admission. Deep learning techniques were utilized to perform both quantitative and qualitative analyses of lung CT images. Additionally, latent class analysis (LCA) was conducted using a combination of clinical, laboratory, and lung CT variables. The regional differences were also examined among subphenotypes through 3D spatial trajectory analysis.
The result showed complete datasets for 559 patients, and LCA identified 2 distinct subphenotypes, including subphenotype 1 (n = 156) and subphenotype 2 (n = 403). Compared to subphenotype 2, patients in subphenotype 1 were older, exhibited higher inflammatory biomarkers, and had greater hypoxemia. Subphenotype 1 also had a higher density gravitational gradient and more consolidated lung regions, while subphenotype 2 showed a greater submantellar hilar density gradient and more ground-glass opacities. Additionally, subphenotype 1 had a higher prevalence of comorbidities linked to endothelial dysfunction and a higher 90-day mortality rate, even after adjusting for clinically relevant variables.
Investigators concluded that combining lung CT data with LCA revealed 2 distinct COVID-19 subphenotypes, highlighting the potential of machine learning in automating patient subphenotyping in respiratory failure.
Source: ccforum.biomedcentral.com/articles/10.1186/s13054-024-05046-3
The post Machine Learning-Based Subphenotyping of COVID-19 Using Lung CT first appeared on Physician's Weekly.