A Reproducible MRI-Based Quantitative Feature for Differentiating Dysplastic Nodules from Hepatocellular Carcinoma: A Multicenter Retrospective Study

Purpose: High percentage of well-differentiated hepatocellular carcinoma (HCC) was radiologically misdiagnosed as dysplastic nodule (DN), which may lead to delay in treatment. This study aims to develop a reproducible feature for DN/HCC differentiation. Patients and Methods: This study included patients which underwent curative hepatectomy or biopsy from four hospitals in China. The study population was divided into internal training cohort, internal test cohort, external validation cohort. Energy value was extracted from arterial and hepatobiliary phase lesion to represent intensity. Hepatobiliary-Arterial Intensity Ratio (HAIR) was calculated to demonstrate the intensity change from cirrhotic nodule to HCC. GPC3 (Glypican-3) was analyzed to validate HAIR’s relation with nodule differentiation. Supporting Vector Machine (SVM) was built as a comparative model. Area Under the Curve (AUC) was used to evaluate and compare the predictive efficacy of HAIR. Results: A total of 116 patients with 120 lesions (52 dysplastic nodules and 68 hepatocellular carcinomas) were included. The AUC of HAIR and SVM for differentiating DN and HCC in internal test cohort were 0.824 (95% CI, 0.662– 0.986, p < 0.01) and 0.733 (95% CI, 0.593– 0.872, p < 0.01), respectively. The cut-off value of the HAIR logistic regression based on training cohort was 0.46, which was subsequently validated in the external validation group. The AUC of HAIR in external validation cohort was 0.667. The correlation between HAIR and GPC3 showed statistical significance ( p < 0.01). Conclusion: HAIR based on arterial and hepatobiliary phase provides a quantitative, reproducible and interpretable tool for DN/HCC differentiation. Keywords: dysplastic nodule, hepatocellular carcinoma, magnetic resonance imaging, quantitative imaging feature, reproducibility