18F-FDG PET/CT Imaging Biomarkers for CAR-T Cell Therapy Outcomes in Large-B-Cell Lymphoma

Introduction: CAR-T cell therapy has transformed the care of patients with large-B-cell lymphoma (LBCL), but disease relapse and toxicities remain clinical challenges, while effective quantitative biomarkers are lacking. Our aim was to determine the role of 18F-FDG PET/CT imaging features as biomarkers in LBCL treated with CD19 CAR-T cell therapy.

Methods: 18F-FDG PET/CT scans performed before CAR-T for LBCL were analyzed retrospectively by a board certified radiologist. Metabolic tumor volume (MTV), SUVmax, total lesion glycolysis (TLG), presence of bulky disease, and 106 radiomic features per lesion were recorded using the PET/CT and PyRadiomics plugins for FIJI. Laboratory values including IL-6, IL-10, TNF-α, LDH, ferritin, fibrinogen, and D-dimer pre CAR-T infusion were recorded. Response to CAR-T cell therapy was defined according to Lugano criteria. Cytokine release syndrome (CRS) and neurotoxicity were graded according to ASTCT guidelines. Progression-free survival (PFS) and overall survival (OS) were measured from time of CAR-T cell infusion; PFS events were death, relapse, and disease progression. Association between PET features, markers of inflammation and outcomes were studied with Spearman rank correlations, Cox, and logistic regression models. Multivariable generalized estimating equation models were applied to evaluate differential expression of radiomic features between responders and non-responders, accounting for the dependency structure among ROIs from the same individual.

Results: A total of 180 LBCL patients (median age 66 years) treated with autologous CD19 CAR-T (93 axi-cel, 52 tisa-cel, 35 liso-cel) were included. Two time points were considered: 1) pre-apheresis scans (n=161; median SUVmax 19, MTV 56 mL, TLG 427 SUV*mL), and 2) pre-infusion, i.e., latest disease evaluation before CAR-T infusion including post-apheresis/post-bridging, as well as pre-apheresis scans for patients without additional imaging afterwards (n=180; median SUVmax 15, MTV 44 mL, TLG 338 SUV*mL), with a total of 341 PET/CT scans evaluated. At lymphodepletion, MTV was significantly correlated with multiple elevated inflammatory laboratory markers; at latest assessment, all PET and inflammatory markers showed significant correlations. MTV at last assessment before infusion was associated with increased risk of grade ≥ 2 CRS (OR 1.08 [95%CI 1.01-1.20], P=0.031), but not ICANS, in a multivariable logistic regression model adjusted for age, pre-lymphodepletion LDH, CAR-T product, bridging and SUVmax. A higher SUVmax at latest assessment was associated with a lower likelihood of non-complete response (CR) (OR 0.58 [0.41-0.80], P<0.001), while MTV lacked such an association after multivariable model. Increased MTV both before apheresis and at last disease assessment before infusion, but not SUVmax, were associated with decreased PFS (HR 1.11 [1.05-1.17] and HR 1.04 [1.02-1.07], respectively), in a multivariable adjustment. A similar pattern was observed for OS. Notably, the combination of MTV and LDH enabled clear separation of high and low PFS risk groups (P<0.001, Figure 1). Finally, when considering over 100 granular pre-infusion radiomic features (Figure 2), which are not routinely evaluated in CAR-T patients, multiple novel determinants of CR were discovered, including gray level dependence, run length matrix, and shape-based features (P<0.05).

Conclusions: To the best of our knowledge, this represents the most extensive analysis of quantitative 18F-FDG PET/CT features within an LBCL cohort undergoing CAR-T treatment. Our findings indicate that standard PET features (MTV and SUVmax) predict both toxicity and efficacy of CAR-T cell therapy, while also showing correlations with systemic inflammation. Additionally, we present, for the first time, the relevance of granular radiomic features in assessing the probability of treatment success. If confirmed, PET features could serve as effective biomarkers in CAR-T patients and have the potential to guide interventions in high-risk populations.