An AI-based novel system for predicting respiratory support in COVID-19 patients through CT imaging analysis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• COVID-19 significantly impacts respiratory health, necessitating accurate prognostication.
• Determining the appropriate level of respiratory support is critical for patient management.
• Computed tomography (CT) imaging offers detailed visualization of COVID-19 lung lesions.

Purpose of the Study:

• To develop and validate an AI-based system for predicting respiratory support requirements in COVID-19 patients.
• To correlate COVID-19 lesion characteristics on CT scans with the level of respiratory support needed.
• To differentiate between varying severity levels of COVID-19 respiratory compromise.

Main Methods:

• Utilizing CT imaging to analyze COVID-19 lesions.
• Employing a 2D, rotation-invariant, Markov-Gibbs random field (MGRF) model for lesion appearance modeling.
• Developing three MGRF-based models, each corresponding to a respiratory support level (0, 1, or 2).
• Implementing a neural network-based fusion system to integrate MGRF model outputs for final prediction.

Main Results:

• The AI system achieved high prediction accuracy, sensitivity, and specificity in assessing 307 COVID-19 patients.
• The system demonstrated the ability to differentiate between minimum, non-invasive, and invasive respiratory support levels.
• Analysis of CT-derived lesion features correlated significantly with respiratory support needs.

Conclusions:

• The proposed AI system effectively predicts the necessary respiratory support for COVID-19 patients based on CT imaging.
• This AI tool has the potential to aid clinicians in timely and appropriate respiratory management decisions.
• The study highlights the utility of AI and advanced imaging analysis in managing severe respiratory illnesses like COVID-19.