Prediction of respiratory support levels in COVID-19 patients using a CT-based system

The COVID-19 pandemic has underscored the urgent need for intelligent, image-based tools to assist in determining the appropriate level of respiratory support for affected patients. Accurate prediction of respiratory support requirements can enable physicians to stratify COVID-19 severity, optimize resource allocation, prevent clinical deterioration, and ensure timely intervention. This chapter presents an innovative diagnostic framework that predicts respiratory support levels by analyzing correlations between CT-based radiomic features and clinical interventions, categorized as Level 0 (minimal support), Level 1 (non-invasive, e.g., oxygen therapy), and Level 3 (invasive, e.g., mechanical ventilation). The system extracts second-order texture features from segmented COVID-19 lesion regions. Each feature is independently classified using a support vector machine (SVM) to produce localized predictions, which are then fused using a backpropagation neural network. The framework is evaluated on CT scans from 305 COVID-19 patients with varying severity levels, achieving an overall accuracy of 97.02% and a Cohen’s kappa score of 95.48%. These results confirm its potential and reliability in determining the required level of respiratory support.