Using optimized CT type to predict histological classifications of thymic epithelial tumors: a radiomics integrated...

Area of Science:

• Medical Imaging and Radiology
• Oncology
• Artificial Intelligence in Medicine

Background:

• Thymic epithelial tumors (TETs) exhibit diverse histological classifications, impacting clinical management and prognosis.
• Accurate pre-operative histological classification of TETs is crucial for treatment planning.
• Current diagnostic methods often rely on invasive procedures; non-invasive imaging biomarkers are needed.

Purpose of the Study:

• To develop and externally validate an integrated model for predicting TET histological classifications.
• To utilize optimized radiomics features from non-contrast-enhanced CT (NE-CT) and contrast-enhanced CT (CE-CT).
• To combine radiomics, morphological, and clinical risk factors for enhanced predictive performance.

Main Methods:

• A cohort of 182 patients with TET was divided into training (N=122) and external validation (N=60) sets.
• Radiomics features were extracted from NE-CT and CE-CT, followed by rigorous feature selection to create Rad-scores.
• An integrated model was built using multivariate logistic regression, combining optimal Rad-scores, morphological features, and clinical risk factors.

Main Results:

• Radiomics features from CE-CT demonstrated superior performance compared to NE-CT in predicting TET classification (AUCs 0.783 vs 0.749 in training, 0.775 vs 0.723 in validation).
• The integrated model, incorporating CE-CT radiomics and five morphological features, achieved high predictive accuracy (AUCs 0.814 in training, 0.802 in validation).
• The integrated model showed impressive predictive capability in distinguishing TET histological classifications upon external validation.

Conclusions:

• Radiomics features from CT effectively capture TET heterogeneity, outperforming morphological features alone.
• Contrast-enhanced CT-based radiomics provides a more robust basis for predicting TET histological classifications than NE-CT.
• The developed integrated model offers a promising non-invasive tool for identifying TET histological classifications, aiding clinical decision-making.