Abstract
Background: Acute respiratory distress syndrome (ARDS) is a critical disease in the intensive care unit (ICU) with high morbidity and mortality. The accuracy for predicting outcomes of ARDS patients with mechanical ventilation is limited, and most predicting model are based on clinical information. This study aimed to explore the prognostic value of chest CT images in ARDS patients with mechanical ventilation and develop an outperformed predictive model.Methods: The patients diagnosed with ARDS between January 2014 and June 2019 were retrospectively recruited. Clinical information, ventilation parameters, primary causes, illness severity, and chest CT images were collected. Radiomics features were extracted from the levels of the upper, middle, and lower lungs, and were further screened according to the primary outcome (28-day mortality after ARDS onset). Radiomics Scores for each level were computed. The univariate and multivariate logistic regression analyses were applied to figure out risk factors. Various predictive models were constructed and compared.Results: Of 366 ARDS patients recruited in this study, 276 (median age, 64 years [interquartile range, 54–75 years]; 208 male) survive on the Day 28. Among all factors, the APACHE Ⅱ Score (OR, 2.607, 95% CI: 1.896-3.584, P < 0.001), the Radiomics_Score of the middle lung (OR, 2.230, 95% CI: 1.387-3.583, P = 0.01), the Radiomics_Score of the lower lung (OR, 1.633, 95% CI: 1.143-2.333, P = 0.01) was associated with the 28-day mortality. The clinical_radiomics predictive model (AUC, 0.813, 95% CI: 0.767-0.850) show the best performance compared with the clinical model (AUC, 0.758, 95% CI: 0.710-0.802), the radiomics model (AUC, 0.692, 95% CI: 0.641-0.739) and the various ventilator parameter-based models (highest AUC, 0.773, 95% CI: 0.726-0.815).Conclusions: The radiomics features of chest CT images have incremental values in predicting the 28-day mortality in ARDS patients with mechanical ventilation. These results help to build a personalized prognostic prediction model and to stratify high-risk patients.