A hybrid machine learning/deep learning COVID-19 severity predictive model from CT images and clinical data

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Computational Pathology

Background:

• COVID-19 exhibits diverse clinical presentations and prognoses, necessitating accurate outcome prediction.
• Distinguishing between non-intensive care unit (non-ICU) and intensive care unit (ICU) outcomes is crucial for patient management.
• Early identification of patients at risk for severe COVID-19 is essential for resource allocation and timely intervention.

Purpose of the Study:

• To develop and validate a hybrid machine learning and deep learning model for classifying COVID-19 patient outcomes.
• To predict the likelihood of ICU admission or death in COVID-19 patients.
• To provide clinical decision support through probabilistic outcome scores and interpretable feature importance.

Main Methods:

• A hybrid model combining a 3D Convolutional Neural Network (CNN) for feature extraction from CT images with machine learning algorithms (Boruta, CatBoost) was developed.
• Patient data included baseline CT scans, laboratory results, and clinical information from 558 patients.
• SHAP (SHapley Additive exPlanations) values were utilized for feature selection and model interpretability.

Main Results:

• The hybrid model achieved a high Area Under the Curve (AUC) of 0.949 on the holdout test set.
• The model successfully classified patients into non-ICU and ICU (intensive care admission or death) outcome categories.
• Case-based SHAP interpretation provided insights into the importance of specific features for outcome prediction.

Conclusions:

• The developed hybrid AI model demonstrates significant potential for predicting COVID-19 patient outcomes.
• This tool can assist clinicians in making informed decisions regarding patient care and resource allocation.
• The integration of deep learning for image analysis and interpretable AI methods enhances clinical utility.