E2836. LIRADS-TRA for Predicting Residual HCC After Y90 Therapy: Efficacy of Routine Clinical Versus Expert Reads with Histologic Correlation
Authors
Matthew Harwood;
Mayo Clinic
Hilary Ho;
Mayo Clinic
Maria Zulfiqar;
Mayo Clinic
Cameron Adler;
Mayo Clinic
Charis Wang;
Mayo Clinic
Parth Parikh;
Mayo Clinic
Alvin Silva;
Mayo Clinic
Objective:
The Liver Imaging Reporting and Data System treatment response algorithm (LIRADS-TRA) is a standardized assessment tool for assessing hepatocellular cancer (HCC) response to locoregional therapy. As transarterial radioembolization (TARE) with yttrium-90 has unique post-treatment features that can confound assessment, this study aims to compare expert readers’ blinded interpretation applying LIRADS-TRA against MRI/CT interpretations obtained during routine clinical care for predicting HCC histopathologic viability.
Materials and Methods:
Following IRB exemption and using institutional database search tool, we identified 96 patients (age: 37-72) who underwent liver transplant after Y90 treatment for HCC between 10/2015 and 10/2021. 27 were excluded because at least 90 days had not elapsed between treatment and MRI/CT, or because multiple treatments employed. Ultimately, 25 patients with at least 1 treated HCC comprised the cohort. The most common cirrhotic etiologies were hepatitis C (40%), alcohol (24%), and NASH (20%). Mean time between treatment and MRI (23/25) or CT (2/25) was 164 ±89 days (mean ±SD). Three fellowship-trained, abdominal radiologists (1, 8, 20 years’ experience) blinded to pathology results independently reviewed pre- and post-treatment exams, strictly adhering to LIRADS-TRA criteria. In comparison, only 6 of 25 (24%) clinical reports explicitly used LIRADS-TRA. LR-TRA viable & equivocal were considered positive. LR-TRA nonviable was considered negative. Lesions were characterized histologically as viable (<100% tumor necrosis) or nonviable (100%). Performance characteristics and predictive values for LR-TRA categories were calculated for each reader, and Interreader association calculated using Intraclass Correlation Coefficient (ICC).
Results:
For predicting 100% pathologic necrosis, expert reads’ LR-TRA nonviable accuracy and positive predictive value ranged up to 80% and 100% respectively, as compared to 60% and 80% for Clinical reads. However, the range was 48 - 80% and 50 - 100% respectively for the three Expert readers, with ICC for LR-TRA: 0.24 (95%: CI 0.07 - 0.41). Similar sensitivity/specificity/positive likelihood ratio/negative likelihood ratio for Clinical (31%, 92%, 3.7, 0.76) as compared to Expert readers (46%/, 81%/, 1.7, 0.57).
Conclusion:
Strict adherence to LIRADS-TRA did not improve accuracy or test performance as expected. Poor inter-reader variability is attributed to considerable heterogenous non-target/background liver radiation effects and prolonged expected treatment-related intralesional enhancement that may persist for months. Previous authors evaluating liver resection pathology after Y90 showed higher accuracy. However, their post-treatment imaging time frame was not defined, whereas this study focused on the 6-month post-treatment exam. Assessment of later post-treatment dates, and perhaps ancillary features may help predictions. Additionally, texture analysis and machine learning may eventually lead to better predictions. Further prospective studies with a large multicenter multireader setting will be valuable.
