201 - Baseline PSMA PET-Derived SUVmean as a Prognostic Factor for Lesion-Level Metabolic Response to [177Lu]Lu-PSMA-617 In Metastatic Castration-Resistant Prostate Cancer

11:25am - 11:35am ET

Room 205

Presenter(s)

Nikolai Strusberg, MD - Baptist Health Herbert Wertheim Cancer Institute, Miami, FL

N. Strusberg^1,2, A. Gutierrez^1,2, M. P. Mehta^1,2, R. P. Tolakanahalli^1,2, S. C. George^1,2, L. Hodgson^2,3, M. A. Fagundes^1,2, Y. Weiss^1,2, M. A. M. Rodrigues^1,2, K. Mamanna^1,4, and A. Kaiser^1,2; ¹Department of Radiation Oncology, Miami Cancer Institute, Baptist Health South Florida, Miami, FL, ²Department of Oncological Sciences, Herbert Wertheim College of Medicine, Florida International University, Miami, FL, ³Department of Biostatistics, Miami Cancer Institute, Miami, FL, ⁴Florida International University, Miami, FL

Purpose/Objective(s): Lesion-level response to [177Lu]Lu PSMA 617 radiopharmaceutical therapy (Lu-RPT) in metastatic castration-resistant prostate cancer (mCRPC) is heterogeneous, and predictive variables of response are insufficiently defined. Conventional imaging and biomarker endpoints may underestimate response heterogeneity across individual metastases. We performed an artificial intelligence (AI)-based paired PSMA PET/CT analysis to characterize lesion-level response patterns and test whether baseline PSMA PET/CT features are associated with subsequent lesion-level response.

Materials/Methods: Patients with mCRPC who underwent Lu-RPT with baseline and 3-month post-treatment PSMA PET/CT scans were retrospectively analyzed. Lesions were identified, tracked, and quantified using a commercially available AI segmentation and tracking platform. SUVmean, SUVmax, SUVtotal (SUVmean x volume), volume (cc), and tumor location (osseous, visceral, other) were quantified for each lesion. Lesion-level response was classified by SUVtotal change between baseline and post-treatment imaging and categorized as complete response (CR) for SUVtotal of 0, partial response (PR) if SUVtotal decreased =30%, progressive disease (PD) if SUVtotal increased =30%, and stable disease (SD) if SUVtotal remained within ±30%. Multinomial logistic regression evaluated associations between PET/CT parameters and lesion-level response, with patient included as a fixed-effect covariate.

Results: Between 2023–25, we treated 33 patients (1247 evaluable lesions) with a median of 4 Lu-RPT cycles (IQR, 2–6). Of these, 443 lesions were baseline-present and 804 were newly detected on post-treatment scans; analyses were restricted to baseline lesions. Among the 443 baseline lesions, 84% were osseous. In multinomial logistic regression (PD reference), SUVmean unadjusted odds ratios (OR) for SD vs PD and PR vs PD were 1.39 and 1.73 (p<0.001), with patient-adjusted odds ratios (aOR) of 1.43 and 1.66 (p<0.001), respectively. SUVmean was not significant for CR vs PD after adjustment (aOR 1.04, p=0.68). Larger baseline lesion volume was inversely associated with CR vs PD (OR 0.95, p=0.006; aOR 0.89, p<0.001). SUVmax demonstrated weaker associations for SD vs PD (OR 1.09, p<0.001; aOR 1.09, p<0.001), PR vs PD (OR 1.14, p<0.001; aOR 1.13, p<0.001), and CR vs PD (OR 1.06, p=0.003; aOR 0.989, p=0.66).

Conclusion: AI-derived lesion-level PSMA PET/CT analysis in mCRPC treated with Lu-RPT reveals a strong association between baseline SUVmean and lesion-level outcomes. Higher SUVmean was associated with increased odds of PR and SD relative to PD, whereas larger baseline lesion volume was inversely associated with CR vs PD. These findings support quantitative lesion-level PSMA PET as a potential tool for risk- or response-adaptive RPT strategies.