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

This study was approved by the Ethics Committee of Tianjin Third Central Hospital. Informed consent was waived due to the retrospective study design. Using electronic medical records, data from patients with a high risk of HCC [8] between June 2016 and June 2021 were retrospectively collected. The inclusion criteria were as follows: 1) ≥ 18 years old; 2) underwent EOB-MRI scans; 3) nodule number ≤ 3;and 4) underwent surgery or pathological biopsy. Patients meeting the following criteria were excluded from the study: (1) patients who had cirrhosis due to congenital hepatic fibrosis or vascular disorder; (2) interval of more than 1 month between pathological diagnosis and EOB-MRI; (3) diffuse hepatic lesions; (4) treatment prior to the EOB-MRI scans; (5) liver function was Child-Pugh C; (6) suboptimal image quality; or (7) LR-TIV (Fig. 1).

Patient clinicopathologic characteristics, including the age, sex, number of patients with cirrhosis, and the causes of liver disease were recorded through the patient records system. A board-certified radiologist with 4 years (J.W.) of experience in abdominal MRI reviewed all images and recorded the number of lesions.

For all examinations, studies were carried out by using a 3.0T MR system (Magnetom Verio, Siemens Healthcare, Erlangen, Germany) and an 8-channel phased-array torso coil. The liver MR imaging protocol consisted of in- and out-of-phase T1-weighted imaging (T1WI) acquired with a gradient recalled echo (GRE) sequence, respiratory-triggered axial T2-weighted turbo spin echo (TSE) sequence with fat suppression, free-breathing single-shot echo-planar diffusion-weighted imaging (DWI) with b values of 0, 50, 600, and 1000s/mm². The liver MR imaging protocol also includes pre- and postcontrast T1-weighted three-dimensional volumetric interpolated breath-hold examination (VIBE) sequences acquired with a GRE sequence in the late arterial phase (AP) (30–35 s after aortic enhancement using the bolus tracking method), portal venous phase (PVP) (46–60 s), transitional phase (TP) (150–180 s), and hepatobiliary phase (HBP) (20 min after the contrast injection). Contrast-enhanced dynamic MR images of the liver were obtained after intravenous administration of gadoxetic acid (Primovist; Bayer Healthcare, Berlin, Germany) at 0.025 mmol/kg of body weight and a rate of 1.0 mL/s via a power injector, followed by 25 mL of 0.9% saline chaser at the same rate. The detailed parameters of each acquisition sequence are shown in Table 1.

All MRI scans were independently reviewed by two board-certified radiologists with 10 years (W.J.H.) and 6 years (D.W.) of experience in abdominal MRI, respectively. All MFs and AFs for each liver observation according to the LI-RADS v2018 were reviewed. All readers were blinded to the pathologic results. Discrepancies between the two readers were resolved by a third radiologist (R.L. with 17 years of experience in abdominal MRI) to reach a final consensus reading. For observations with documented threshold growth, readers determined the qualifying threshold growth criterion (≥ 50% diameter increase in ≤ 6 months) by retrospectively reviewing, measuring and comparing the observation size between current and prior exams results, with prerequisite that no treatment is performed between the two exams. The imaging exams, including EOB-MRI, contrast-enhanced MRI or CT (CE-MRI or CECT). The TP or HBP image from EOB-MRI and the PVP or delayed phase (DP) image from CE-MRI or CECT were selected to measure the observation size (the longest diameter on the plane with the largest cross-sectional area of the lesion), and the final lesion size was the average of the measurement results of 2 readers.

Logistic regression analyses were performed to identify independent significant AFs for the diagnosis of HCC, which were then used as new MFs to replace the threshold growth feature. The sizes of observations with and without threshold growth were compared in the HCC, non-HCC malignancy and benign lesion groups.

Every observation was initially assigned to a LI-RADS category according to the LI-RADS v2018 algorithm (first classified according to the MFs, then adjusted according to the AFs as follows: for ≥ 1 AF favouring malignancy, the LI-RADS category was upgraded by 1 up to LR-4; for ≥ 1 AF favouring benignity, the LI-RADS category was downgraded by 1; and for ≥ 1 AF favouring malignancy and ≥ 1 AF favouring benignity, the LI-RADS category remained unchanged).

Then, every observation was reassigned a category according to the following two schemes: Scheme A: the categories were assigned using all MFs except threshold growth, with only the threshold growth feature downgraded and treated as an AF favouring malignancy (not HCC in particular). Scheme B: the threshold growth feature was replaced with independently significant AFs and the latter were upgraded and treated as new MFs. LR-5 was considered as 100% definite HCC by LI-RADS classification, so the pathological results of LR-5 were taken as the gold standard reference. The diagnostic performance of schemes A and B for HCC diagnosis was calculated and compared with that of the unaltered LI-RADS v2018.

Categorical variables were summarized as counts and percentages. Continuous variables were summarized as the means and standard deviations and were compared using Student’s t-test. The median (interquartile range) was used for nonnormally distributed data. To determine the AFs performance of HCC prediction, univariate and multivariate logistic regression analyses were performed. Variables with a p value < 0.1 in the univariable analysis were entered into the multivariable analysis to identify the independently significant AFs for HCC diagnosis. For the multivariable analysis, a stepwise backwards elimination method was used. Diagnostic performance was reported as the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and Youden index, and comparisons between grading protocols were made using McNemar’s test. Unless otherwise indicated, all statistical tests were two-tailed and conducted at the statistically significant level of 0.05 with p values reported. Statistical analyses were performed using SPSS Statistics version 25.0 (IBMCorp, Armonk, NY, USA).