Interpretive consistency of a qualitative ⁶⁸Ga-PSMA PET framework for treatment response assessment in hepatocellular carcinoma: a head-to-head comparison with cross-sectional imaging.

Abstract

Accurate treatment response assessment in HCC is crucial but challenged by interpretive variability on CT/MRI. PSMA has emerged as a theranostic biomarker in HCC; however, interpretive reliability is a prerequisite for its validation against oncologic outcomes. Therefore, we evaluated interpretive consistency of a qualitative ⁶⁸Ga-PSMA PET framework for treatment response assessment in HCC, benchmarking it against CT/MRI criteria. Patients with treatment-naïve HCC were prospectively enrolled for ⁶⁸Ga-PSMA PET under an FDA IND application. Eligibility for analyses necessitated a baseline ⁶⁸Ga-PSMA-avid target lesion with post-treatment ⁶⁸Ga-PSMA PET and CT/MRI for response assessment following either locoregional therapy (LRT) or immune checkpoint inhibitor therapy (ICI). Three radiologists independently assessed response on ⁶⁸Ga-PSMA PET using a 4-point scale, subsequently dichotomized to viable (uptake > liver) versus non-viable status (uptake ≤ liver). Reliability was benchmarked against binary mRECIST, defined by arterial phase hyperenhancement (APHE), ordinal and binary RECIST 1.1, and binary LR-TR criteria. Inter- and intra-reader agreement were quantified using Gwet's agreement coefficients (AC1/AC2), with differences between dichotomized methods compared using non-parametric bootstrapping. From 88 patients with baseline ⁶⁸Ga-PSMA PET, 13 were eligible: 7 with 10 lesions received LRT, and 6 with 19 lesions received ICI. Overall inter-reader agreement on ⁶⁸Ga-PSMA PET was almost perfect for both the 4-point qualitative scale (Gwet's AC2: 0.86, 95% CI: 0.77-0.96) and the binary viable/non-viable status (AC1: 0.90, 95% CI: 0.77-1.00). Intra-reader agreement was also almost perfect (AC1 range: 0.86-0.93). In contrast, inter-reader agreement for mRECIST was moderate (AC1: 0.54, 95% CI: 0.29-0.77), though its intra-reader agreement was more robust (AC1 range: 0.72-0.86). In the ICI subgroup, inter-reader agreement using ⁶⁸Ga-PSMA PET (AC1: 0.93) was significantly higher than for mRECIST (AC1: 0.37; mean ΔAC1 = 0.53, 95% CI for difference: 0.19-0.85, P = 0.006) and dichotomized RECIST 1.1 (AC1: 0.39; mean ΔAC1 = 0.52, 0.25-0.83, P < 0.001). In the LRT subgroup, inter-reader agreement using ⁶⁸Ga-PSMA PET (AC1: 0.89) was also almost perfect, but not significantly different from mRECIST (AC1: 0.87) or dichotomized LR-TR (AC1: 0.88) (P > 0.99 for both), with similarly high and comparable intra-reader agreement across all methods in both subgroups (P = > 0.99). A qualitative PSMA PET framework for HCC treatment response assessment yields nearly perfect interpretive consistency, outperforming CT/MRI-based criteria in post-ICI settings. These findings provide the foundation to validate PSMA-based response against oncologic outcomes in future clinical trials, a critical step toward guiding patient management and accelerating therapeutic innovation in HCC.