FDG‐PET IMAGING AND RADIOMICS IN RESPONSE ASSESSMENT OF LYMPHOMA PATIENTS UNDERGOING CAR T‐CELL THERAPY

Background: Radiomics involves the extraction of quantitative features from medical images, such as positron emission tomography (PET), representing potential surrogate markers of the lymphoma phenotypes. Chimeric antigen receptor-T (CAR-T) cell therapy has revolutionized the treatment of large B cell lymphomas. Unfortunately, the relapse rate is around 30–60% and almost 20% of patients develop grade >3 Cytokine Release Syndrome (CRS). In this setting, it is of critical importance the early identification of relapsed/refractory patients and of whom can develop CRS. Primary aim was to early predict the response to CAR-T based on metabolic features extracted from the clinical and baseline FDG-PET. Secondary aim was to determine the presence of CRS from the clinical and image-based features. Methods: From September 2019 to December 2020, 20 patients were treated with CAR T-cell therapy. All patients underwent FDG-PET/CT evaluation at baseline (PET_0) and 1 month after (PET_1) CAR T-cell infusion. PET_0 semi-quantitative parameters, namely SUVmax, metabolic tumor volume (MTV), total lesion glycolysis (TLG), were calculated. Pyradiomics library was used for the extraction of 105 radiomics features from each image. Univariate analysis on both clinical and radiomics features was performed to evaluate the correlation with the outcome. A generalized linear model (GLM) was trained to predict the outcome and ROC analysis was used to assess the prediction capability. Results: Patients had a wide range of baseline disease burden, with a median MTV of 129 ml (range: 5-5916) and a median TLG of 971 Bq (range: 21-32241). Ten (53%) patients achieved a complete (CR) and 9 (47%) a partial response (PR). Of these 9 patients, 6 underwent re-evaluation at 3 months: 1 converted to CR, 4 had a progression to PD and 1 patient maintained the PR. No correlation was found between baseline MTV and TLG and tumor response at PET_1, while they were significantly associated with the severity of CRS (p < 0.5) (AUC: 0.95; 95%CI: 0.87-1 for MTV and AUC: 0.92; 95%CI: 0.79-1 for TLG). Two radiomics features, Kurtosis and Median, were statistically significantly correlated with response at PET_1, while the surface area was statistically significantly correlated with moderate/severe CRS. In the GLM, only Median was a prognostic factor of response, with an AUC of 0.81 (p < 0.002) while surface area was a prognostic factor of moderate/severe CRS with an AUC of 0.81 (p < 0.009). Conclusions: Baseline FDG-PET radiomics features were able to differentiate between early responder and non-responder patients treated with CAR-T and between patients with/without moderate/severe CRS; MTV and TLG were also associated with CRS. Further correlation between FDG-PET_0 radiomics features/clinical baseline characteristics and patients’ outcome (including progression free survival) will be investigated with larger cohort and longer follow-up. Keywords: Diagnostic and Prognostic Biomarkers, PET-CT, Cellular therapies No conflicts of interests pertinent to the abstract.