Radiomics-enhanced modelling approach for predicting the need for ECMO in ARDS patients: a retrospective cohort study

Area of Science:

• Radiology and Imaging
• Critical Care Medicine
• Machine Learning in Healthcare

Background:

• Clinical decisions for vv-ECMO in ARDS often lack precision due to insufficient reflection of lung injury severity.
• Existing methods may not adequately capture the heterogeneity of lung damage in ARDS patients.

Purpose of the Study:

• To develop a predictive model integrating machine learning (ML)-derived CT features with clinical data for early vv-ECMO decision-making in ARDS.
• To enhance individualized patient management and optimize vv-ECMO initiation.

Main Methods:

• Retrospective cohort study of 375 COVID-19 ARDS patients.
• Utilized CNN for lung segmentation and extracted 592 radiomic features.
• Developed and compared Imaging, Clinical, and Combined logistic regression models.

Main Results:

• The Combined Model demonstrated the highest predictive performance in the validation cohort (AUROC 0.705), outperforming Clinical (0.674) and Imaging (0.639) models.
• Kaplan-Meier analysis showed a significantly higher incidence of ECMO therapy in high-risk predicted patients (p < 0.001).

Conclusions:

• Quantitative CT features and radiomic markers, when combined with clinical data, significantly enhance prognostic accuracy for vv-ECMO need in ARDS.
• The developed Combined Model offers a data-driven approach to support timely, individualized vv-ECMO decisions in critical care settings.