Preoperatively predicting vessels encapsulating tumor clusters in hepatocellular carcinoma: Machine learning model...

Area of Science:

• Radiology and Imaging
• Oncology
• Medical Informatics

Background:

• Vessels encapsulating tumor clusters (VETC) is a distinct vascular pattern in hepatocellular carcinoma (HCC).
• VETC is an independent risk factor for poor HCC prognosis.
• VETC facilitates tumor cluster entry into the bloodstream.

Purpose of the Study:

• To develop and validate a preoperative nomogram for predicting VETC+ in HCC.
• Utilize contrast-enhanced computed tomography (CECT) for prediction.
• Integrate radiomics and clinical-radiological features.

Main Methods:

• Retrospective analysis of 190 HCC patients with CECT and immunochemical staining.
• Radiomics analysis of intratumoral and peritumoral regions.
• Machine learning to develop a radiomics model and nomogram.
• Validation on two independent test sets using ROC and decision curve analysis.

Main Results:

• A radiomics nomogram combining clinical-radiological and radiomics features achieved AUCs of 0.859, 0.848, and 0.757 in training and test sets.
• The nomogram demonstrated superior clinical utility compared to individual feature models.
• 13 radiomics features were selected to construct the Rad-score.

Conclusions:

• A CECT-based radiomics nomogram can effectively identify VETC+ HCC.
• The nomogram integrates clinical-radiological and radiomics features for enhanced prediction.
• This tool shows potential for improving preoperative assessment and patient management in HCC.