Identifying severe community-acquired pneumonia using radiomics and clinical data: a machine learning approach

Area of Science:

• Medical Imaging
• Machine Learning
• Pulmonology

Background:

• Community-Acquired Pneumonia (CAP) diagnosis requires accurate severity assessment for effective treatment.
• Distinguishing Severe Community-Acquired Pneumonia (SCAP) early is critical for patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning model for rapid and accurate SCAP identification.
• To evaluate the efficacy of radiomic and clinical features, individually and combined, in SCAP detection.

Main Methods:

• Radiomic features were extracted from chest CT scans of 174 CAP patients (64 SCAP).
• Clinical indicators were screened alongside radiomic features to create distinct feature sets.
• Eight machine learning models were trained and evaluated for SCAP identification, including interpretability analysis.

Main Results:

• 15 radiomic features and 2 clinical features (Lymphocyte, Albumin) were identified as significant predictors.
• Models using only radiomic features achieved an AUC of 0.85; clinical features yielded an AUC of 0.82.
• The combined radiomic and clinical feature set, using Ada Boost, resulted in the highest AUC of 0.89.

Conclusions:

• Combining radiomics and clinical data significantly improves the accuracy of SCAP identification.
• Machine learning models offer a promising approach for objective and efficient SCAP assessment.
• This integrated approach aids in timely and appropriate clinical decision-making for CAP patients.