Deep Learning-Based Computer-Aided Detection System for Preoperative Chest Radiographs to Predict Postoperative...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pulmonary Medicine

Background:

• The predictive value of preoperative chest radiography (CR) for postoperative pneumonia is not well-established.
• Accurate prediction of postoperative pneumonia is crucial for patient management and outcomes.

Purpose of the Study:

• To develop and validate a prediction model for postoperative pneumonia.
• To incorporate findings from preoperative CRs analyzed by a deep learning-based computer-aided detection (DL-CAD) system.

Main Methods:

• Retrospective analysis of consecutive surgical patients (development and validation cohorts).
• Preoperative CRs analyzed using a commercial DL-CAD system to identify 10 abnormalities.
• Logistic regression models (clinical model vs. DL-CAD model) developed to predict postoperative pneumonia.

Main Results:

• Pulmonary nodules, consolidation, and cardiomegaly on CRs, identified by DL-CAD, were independent predictors of postoperative pneumonia in the development cohort.
• The DL-CAD model demonstrated superior discriminative performance (AUROC 0.843) compared to the clinical model (AUROC 0.815) in the validation cohort.

Conclusions:

• Abnormalities detected by DL-CAD in preoperative CRs are significant risk factors for postoperative pneumonia.
• Integrating DL-CAD analysis of preoperative CRs enhances the prediction of postoperative pneumonia.