Incremental predictive value of a CT-based deep learning radiomics model for differentiating benign and malignant...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Pleural effusion diagnosis relies on invasive procedures and biochemical markers with limitations.
• Differentiating malignant pleural effusion (MPE) from benign pleural effusion (BPE) is critical for patient management.

Purpose of the Study:

• To develop and validate a diagnostic model for MPE and BPE using non-contrast chest CT deep learning (DL) and radiomics features.
• To assess the incremental diagnostic value of this model when combined with conventional biochemical biomarkers.

Main Methods:

• Retrospective analysis of 208 patients for training/testing and 52 for temporal validation.
• Extraction of radiomics and DL features from non-contrast CTs to create a radiomics score (Radscore).
• Multivariable logistic regression, Net Reclassification Improvement (NRI), and Integrated Discrimination Improvement (IDI) were used for evaluation.

Main Results:

• The Radscore, built on four radiomics features, independently predicted MPE (OR: 2.718-2.776, P < 0.05) with good performance (AUC = 0.805-0.864).
• The Radscore showed weak correlation with pleural effusion carcinoembryonic antigen (pCEA; r = 0.270, P = 0.001).
• Integration of the Radscore with biochemical markers (pCEA, sCEA, pADA, sCA125) improved risk estimation for 75.0% of MPE and 66.7% of BPE patients.

Conclusions:

• The DL-radiomics-based Radscore is a promising quantitative biomarker for differentiating MPE from BPE.
• It functions independently of conventional biochemical metrics.
• The Radscore provides meaningful incremental value, refining risk probability estimation in pleural effusion patients.