CT-Based Radiomic Signatures Associated with Serum CEA Status in Colon Cancer.

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Carcinoembryonic antigen (CEA) is a biomarker for colon cancer management, but its accuracy is limited by variability and potential for false results.
• Radiomics offers quantitative assessment of tumor heterogeneity from medical images, providing objective tumor characteristics.

Purpose of the Study:

• To evaluate the potential of computed tomography (CT)-based radiomic features to distinguish between CEA-positive and CEA-negative colon cancer patients.
• To explore the association between radiomic features and serum CEA status in colon cancer.

Main Methods:

• Retrospective analysis of 109 colon cancer patients' preoperative contrast-enhanced CT images.
• Extraction and selection of 107 radiomic features, followed by machine learning classification (k-NN, SVM, NN) using 5-fold cross-validation.
• Performance evaluation using accuracy, recall, specificity, F1-score, and ROC-AUC.

Main Results:

• A subset of 41 radiomic features effectively differentiated CEA-positive from CEA-negative patients.
• The k-Nearest Neighbor (k-NN) classifier achieved the highest accuracy (77.4%) and ROC-AUC (0.8523).
• Support Vector Machine (SVM) and Neural Network (NN) classifiers demonstrated high recall (83.0%).

Conclusions:

• CT-based radiomics combined with machine learning shows significant potential for differentiating colon cancer patients by serum CEA status.
• Radiomic features can provide imaging-based insights correlating with serum biomarkers like CEA.
• This approach may enhance tumor characterization and support personalized treatment decisions in colon cancer management.