LI-RADS Nonradiation Treatment Response Algorithm Version 2024: Diagnostic Performance and Impact of Ancillary Features.

Abstract

<b>BACKGROUND.</b> LI-RADS Treatment Response Algorithm (TRA) version 2024 (v2024) introduced separate algorithms for detecting hepatocellular carcinoma (HCC) viability after radiation and nonradiation locoregional therapies (LRTs). The nonradiation algorithm incorporated MRI-based ancillary features to optionally upgrade lesions from LR-TR Equivocal to LR-TR Viable. <b>OBJECTIVE.</b> The purpose of this study was to compare the diagnostic performance of LI-RADS Nonradiation TRA v2024 with that of LI-RADS TRA version 2017 (v2017) and modified RECIST (mRECIST) for evaluating HCC response to LRT on MRI, with attention given to the impact of ancillary features. <b>METHODS.</b> This retrospective study included 231 patients (198 men and 33 women; median age, 56 years) who underwent LRT for HCC followed by liver resection or transplant between January 2017 and December 2022. Two radiologists (reader 1 and reader 2) independently evaluated treated lesions (<i>n</i> = 306) using LI-RADS Nonradiation TRA v2024, LI-RADS TRA v2017, and mRECIST. Lesions were classified as showing pathologic viability (<i>n</i> = 249) or complete pathologic necrosis (<i>n</i> = 57) based on curative surgery pathology. The diagnostic performance for pathologic viability was compared using Bonferroni-adjusted McNemar tests, with LR-TR Equivocal assessments classified as test negative. <b>RESULTS.</b> The sensitivity, specificity, and accuracy for LI-RADS Nonradiation TRA v2024 with ancillary features were 85.5%, 75.4%, and 83.7%, respectively, for reader 1 and 87.2%, 63.2%, and 82.7%, respectively, for reader 2; for LI-RADS Nonradiation TRA v2024 without ancillary features, they were 81.1%, 78.9%, and 80.7%, respectively, for reader 1 and 80.3%, 78.9%, and 80.1%, respectively, for reader 2; for LI-RADS TRA v2017, they were 79.9%, 82.5%, and 80.4%, respectively, for reader 1 and 79.1%, 79.0%, and 79.1%, respectively, for reader 2; and for mRECIST, they were 83.9%, 54.4%, and 78.4%, respectively, for reader 1 and 87.2%, 40.4%, and 78.4%, respectively, for reader 2. LI-RADS Nonradiation TRA v2024 with ancillary features showed higher sensitivity and accuracy than LI-RADS Nonradiation v2024 without ancillary features (both readers), higher sensitivity than LI-RADS TRA v2017 (both readers), higher specificity than mRECIST (both readers), and higher accuracy than LI-RADS TRA v2017 (reader 2) (<i>p</i> < .008); remaining comparisons between LI-RADS Nonradiation TRA v2024 with ancillary features and other systems were not significant (<i>p</i> > .008). <b>CONCLUSION.</b> LI-RADS Nonradiation TRA v2024 showed good diagnostic performance in detecting pathologic viability. Ancillary features yielded improved sensitivity and accuracy without a significant change in specificity. <b>CLINICAL IMPACT.</b> Use of LI-RADS Nonradiation TRA v2024 with ancillary features is recommended for guiding prognostic assessments and treatment decisions after LRT.