Hepatocellular carcinoma likelihood on MRI exams: evaluation of a standardized categorization system

Abstract

Purpose: Evaluate the reliability and validity of a standardized reporting system designed to improve communication between the clinician and radiologist regarding likelihood of hepatocellular carcinoma (HCC).

Materials and methods: The system assigns liver lesions into 1 of 5 categories of estimated likelihood of HCC: 1, <5%; 2, 5%-20%; 3, 21%-70%; 4, 71%-95%; 5, >95%. Six American Board of Radiology-certified radiologists reviewed 100 abdominal MRI studies (performed between September 2009 and June 2010 for HCC surveillance) blinded to the official reports and clinical information. Each reader recorded the highest category (1-5) assigned to any lesion per study. Reliability between readers was calculated by the Shrout-Fliess random sets intraclass correlation (ICC). To examine validity, original pretransplant reports from January 2009 to December 2010 were compared to pathology reports on liver explants. Sensitivities, specificities, predictive values, and receiver operating characteristic (ROC) curves were then produced.

Results: The ICC for retrospective readings was 0.80, indicating very good reliability. Of 45 pathologically proven cases, 16 category 1 or 2 cases were all free of HCC (negative predictive value 100%). Five of nine category 3 cases contained HCC. Six of eight category 4 cases contained HCC (PPV 75%). All 12 category 5 cases contained HCC (positive predictive value 100%). The area underneath the ROC curve was 0.949. If categories 1 and 2 are considered negative and categories 3-5 considered positive, this achieves 100% sensitivity with 73% specificity.

Conclusion: This standardized system for reporting likelihood of HCC, which is a forerunner of the recently introduced Liver Imaging Reporting and Data System, produces strong reliability and validity, while aiming to improve the clarity of clinical magnetic resonance imaging reports.