Threshold growth has a limited role in differentiating HCC from other focal hepatic lesions

Abstract

Background The role of threshold growth, as one of the major features (MFs) of HCC in the Liver Imaging Reporting and Data System (LI-RADS) is inconsistent. This study evaluated the LI-RADS diagnostic performance for HCC when threshold growth was removed or replaced by highly frequent ancillary features (AFs). Methods This was a retrospective institutional review board-approved study including patients with a high HCC risk with gadoxetic acid-enhanced MRI findings consistent with pathologically proven focal hepatic observations. With pathological results used as the gold standard reference, the observations were divided into three groups: HCC, non-HCC malignancy and benign lesion. The sizes of the lesions with and without threshold growth were compared. The MFs and AFs of each observation were evaluated and compared among the three groups to select the most highly frequent AFs of HCC and was used to replace threshold growth. The LI-RADS categories of observations were categorized as follows: Scheme A, using all MFs and AFs according to LI-RADS v2018; scheme B, using all MFs except threshold growth, with threshold growth treated as an AF favouring malignancy; and scheme C, using the highly frequent AFs inplace of threshold growth as new MFs. The LR-5 (the category of definitely HCC) diagnostic performance for HCC among the 3 schemes was compared. Results A total of 379 patients and 426 observations were included. There was no statistically significant difference in the frequency of threshold growth between HCCs and non-HCC malignancies (p = 0.560). Whether HCCs, non-HCC malignancies, or benign lesions, the mean size with threshold growth was smaller than that without threshold growth (all p < 0.05). The nodule-in-nodule feature was a highly frequent AF (p < 0.05) and was used to replace threshold growth as a new MF in scheme C. The LR-5 diagnostic performance values for HCC with schemes A, B, and C were respectively as follows: a sensitivity of 74.4%, 74.0% and 75.6%; specificity of 88.6%, 88.6% and 88.6%; and accuracy of 80.3%, 80.0% and 81.0%. There was no statistically significant difference in diagnostic performance between schemes A and B or between schemes A and C (all p > 0.05). Conclusion The threshold growth is removed or replaced by the nodule-in-nodule feature