CT-based liver surface nodularity for the detection of clinically significant portal hypertension: defining measurement quality criteria

CT-based liver surface nodularity for the detection of clinically significant portal hypertension: defining measurement quality criteria:

Abstract

Purpose

To establish measurement quality criteria for the noninvasive assessment of clinically significant portal hypertension (CSPH) in patients with cirrhosis using CT-based liver surface nodularity (LSN) measurements.

Methods

Seventy-four consecutive patients with cirrhosis (mean 62 ± 13 years), including 30 with CSPH (41%), underwent CT and hepatic venous pressure gradient measurements. Three independent readers performed 15 LSN measurements/patient using dedicated software. LSN was computed based on the median and means of one to 15 measurements. Accuracy for diagnosing CSPH was assessed using receiver operating characteristic (ROC) curve analysis. Variability was assessed by the intra-class correlation coefficient (ICC) and the Bland–Altman plot (BA). Quality criteria were identified to maximize the accuracy of LSN and minimize variability.

Results

The area under the (AU) ROCs of mean and median LSN measurements based on one to 15 measurements ranged from 0.79 ± 0.05 to 0.91 ± 0.04 and 0.86 ± 0.04 to 0.91 ± 0.03, respectively, with no difference on pair-wise comparisons (all p > 0.05). AUROCs of LSN increased from one to eight and leveled off between eight and 15 measurements. Inter- and intra-reader variability decreased from one to 15 measurements, with only slight improvement after more than eight measurements. Intra- and inter-observer agreements were excellent with eight measurements (ICC = 0.90 [95%CI 0.84–0.94], and ICC = 0.93 [95%CI 0.89–0.95], respectively), and variability for intra-observer and inter-observer agreement was low (BA bias 4.2% (95% limits of agreement [LoA] [− 15.3; + 23.7%]) and 4.8% LoA [ − 17.5; + 27.1%], respectively).

Conclusions

CT-based LSN measurement is highly reproducible and accurate. We suggest using at least 8 valid measurements to determine the mean LSN value for the detection of CSPH.