Machine Learning Methods Based on Chest CT for Predicting the Risk of COVID-19-Associated Pulmonary Aspergillosis

Area of Science:

• Medical Imaging and Machine Learning
• Infectious Diseases
• Pulmonology

Background:

• Hospitalized COVID-19 patients are at risk for secondary infections.
• Aspergillus infection is a serious complication in these patients.
• Accurate prediction of aspergillosis is crucial for timely intervention.

Purpose of the Study:

• To develop and validate a machine learning model for predicting secondary aspergillus infection.
• The model integrates chest CT radiomics and clinical risk factors.
• To assess the model's predictive performance and clinical utility.

Main Methods:

• Retrospective study of 291 COVID-19 patients.
• Utilized least absolute shrinkage and selection operator regression for feature selection from chest CT.
• Developed a multifactorial logistic regression model combining CT features and clinical data.
• Validated model performance using receiver operating characteristic curves (AUC) and decision curve analysis (DCA).

Main Results:

• A multifactorial model incorporating 11 radiomics features and 7 clinical factors showed high predictive performance.
• Achieved AUC values of 0.98 (training), 0.98 (internal validation), and 0.87 (external validation).
• The multifactorial model significantly outperformed models using only CT or clinical factors alone.

Conclusions:

• The developed multifactorial model is a reliable tool for predicting COVID-19-associated pulmonary aspergillosis.
• This model can aid in early identification and management of secondary aspergillus infections.
• Integration of imaging and clinical data enhances predictive accuracy for opportunistic infections.