Multimodal CT radiomics combined with machine learning algorithms to differentiate benign from malignant pulmonary...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Radiologist interpretation of pulmonary nodules is subjective, leading to misdiagnosis.
• Current methods for pulmonary nodule evaluation lack consistent accuracy.

Purpose of the Study:

• To compare the diagnostic performance of non-contrast-enhanced computed tomography (NCECT) and contrast-enhanced computed tomography (CECT) for pulmonary nodule differentiation.
• To evaluate multi-regional radiomics and machine learning for improved nodule characterization.

Main Methods:

• Retrospective analysis of 194 patients with NCECT and CECT scans.
• Radiomics features extracted from intra-nodular and peri-nodular regions.
• Machine learning models developed and validated using AUC, calibration, and decision curve analysis.

Main Results:

• A logistic regression model demonstrated the best stability.
• Multi-regional analysis outperformed single-region analysis for both NCECT and CECT.
• NCECT combined regions achieved an AUC of 0.901, comparable to CECT.
• External validation of the NCECT model showed robustness with an AUC of 0.863.

Conclusions:

• Multi-regional radiomics models effectively differentiate benign from malignant pulmonary nodules.
• NCECT-based models offer diagnostic efficacy comparable to CECT, suggesting potential for reduced contrast administration.