Towards Automated COVID-19 Presence and Severity Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Accurate COVID-19 classification and severity prediction from 3D thorax CT scans are critical for healthcare management.
• Predicting patient severity is vital for intensive care unit (ICU) capacity planning.

Purpose of the Study:

• To develop and evaluate a deep learning model for COVID-19 presence classification and severity prediction using 3D CT scans.
• To aid medical professionals in patient management and resource allocation.

Main Methods:

• An ensemble learning strategy using 5-fold cross-validation.
• Transfer learning combining pre-trained 3D ResNet34 and DenseNet121 models.
• Domain-specific preprocessing and integration of clinical data (infection-lung-ratio, age, sex).

Main Results:

• Achieved an Area Under the Curve (AUC) of 79.0% for COVID-19 severity prediction.
• Achieved an AUC of 83.7% for COVID-19 presence classification.
• Performance is comparable to existing state-of-the-art methods.

Conclusions:

• The proposed model demonstrates robust performance in classifying COVID-19 presence and predicting severity.
• The approach, implemented in the AUCMEDI framework, ensures reproducibility and can support clinical decision-making.
• Integration of imaging and clinical data enhances predictive capabilities.