CT Radiomics Features of Abdominal Adipose and Muscle Tissues Can Predict the Postoperative Early Recurrence of Hepatocellular Carcinoma.

RATIONALE AND OBJECTIVES

To investigate the potential of computed tomography radiomics features extracted from abdominal adipose and muscle in predicting early recurrence (ER) of hepatocellular carcinoma (HCC) after surgery.

MATERIALS AND METHODS

This retrospective study enrolled 252 patients with HCC who underwent curative resection from two independent institutions. In the training cohort of 178 patients from institution A, radiomics signatures extracted from abdominal visceral adipose, subcutaneous adipose, and muscle were applied to establish the radiomics score using the least absolute shrinkage and selection operator regression. Using multivariable Cox regression analysis, two models were developed: one incorporated preoperative variables, and the other incorporated both pre- and postoperative variables. The external validation of the two models was conducted at institution B with 74 patients.

RESULTS

The preoperative model incorporated tumor size, alpha-fetoprotein, body mass index, and radiomics score, whereas the postoperative model additionally integrated Edmondson-Steiner grade on the basis of the aforementioned parameters. In both cohorts, both models demonstrated superior performance to traditional staging systems and corresponding clinical models (P < 0.01), with time-dependent area under the curve exceeding 0.81 and concordance indices exceeding 0.72. Furthermore, the two models exhibited lower prediction errors (integrated Brier score < 0.19), well-calibrated calibration curves, and greater net clinical benefits. Finally, the two radiomics-based models facilitated risk stratification by accurately distinguishing the high-, intermediate-, and low-risk groups for ER (P < 0.01).

CONCLUSION

Statistical models integrating the radiomics data of abdominal adipose and muscle can provide accurate and reliable predictions of postoperative ER for patients with HCC.