Performance of radiomics features in patients with peripheral T-cell lymphoma: Data from international prospective T-cell project 2.0

Abstract Introduction: Peripheral T-cell lymphomas (PTCLs) are a rare, heterogeneous group of hematological malignancies with extremely poor prognosis for nearly all subtypes. FDG-PET using specific radiomics indexes have showed to be a possible tool for prognostication, treatment response assessment and to extract quantitative features that correlate with the disease's biological characteristics. However, its usefulness in routine clinical practice for PTCLs is still challenging due to the various subtypes, heterogeneity, and the absence of large prospective trials that include FDG-PET assessment. To address this question, we conducted PET assessment as an ancillary sub-study of the prospective real-world T-Cell Project 2.0 (NCT03964480).Patients and methods: 138 PTCLs patients from 8 international centers with available PET0 AND/OR a EOT-PET were enrolled in this study. For the further analysis were confirmed eligible 104 patients (PTCL-NOS 35, AITL 17, ALCL ALKneg 19, ALKpos 9, NKTCL 13, ATLL 2, others 9). Anonymized scans underwent a central review on WIDEN platform by a pool of nuclear medicine physicians. Tumor segmentation was performed at baseline with a percentage threshold of SUVmax of 41% in each lesion. Quantitative metrics including SUVmax, SUVpeak, total lesion glycolysis (TLG) and MTV were defined for each lesion. TMTV and TTLG were defined as the sum of MTV/TLG through all the lesion while SUVmax and SUVpeak were identified in the lesion with the highest uptake at baseline. EOT-PET were analyzed using Lugano classification with the 5-point Deauville scale.Overall survival (OS) and progression free survival (PFS) were estimated by Kalan-Meier method and the prognostic value was assessed in univariate analysis using log-rank test and estimating the effect as hazard ratio (HR), from Cox PH regression model. All reported p-value were two-sided.Results: Median follow-up time was 39 months (95%CI 32-45). OS and PFS at 3 years were 55 vs 46%, respectively. Median TMTV was 98 ml (5%-95% 3-1213). A TMTV of 200 ml was chosen as threshold between high and low risk group for OS and PFS. The 3-yr PFS for TMTV41% 230 was 56% vs 31% for 230 (0.020), TMTV41% 130 w/o SL was 56% vs 36% for 130 and TMTV41% 130 w/o SL (p=0.020) and TMTV41% 130 w/o SL and diff. spleen was 62% vs 39% (p=0.010). In univariate analysis, age60 yo (p=0.001), LDHUNL (p=0.014), PS1 (p=0.008), Hb12g/dL (p 0.042), albumin3.5g/dL (p=0.01), PLT200x109/L (p=0.047), ALC1x109/L (p=0.031), PIT intermediate (p=0.002) and high (p0.001), PMR/PMD (p0.001), DS4-5 (p0.001), TMTV41%200ml (p0.01) were correlated with lower OS. Adverse prognostic features were the same for PFS, except albumin3.5 g/dL (p=0.06). A DS4-5 in EoT PET demonstrated also to have a strong response assessment value with HR= 4.93 (95%CI 2.40-10.1 in PFS and HR= 3.22 (95%CI 1.48-7.00) in OS. The prognostic value for baseline TMTV was reinforced when combined to DS EOT. The 3-yr OS for DS 1-3200 ml vs DS 1-3 200 ml vs DS 4-5 was 89%, 56% and 41% (p0.001), respectively. At the same time timepoints PFS was 85%, 50% and 21% (p0,001), respectively. Analysis by subtypes delineated a robust association between histological variants and metabolic activity, with AITL displaying the highest median TMTV 41% of 486.2 ml compared with 141.6 ml for PTCL NOS, 188.8 ml for ALK +, 88 ml ALK – and 77.0 for other rare PTCL subtypes (p=0.033). Besides this empirical observation, there was no statistically significant difference between histological subtypes and metabolic activity (p=0.073).Conclusion: Our data confirm that TMTV41% appears to be a robust biomarker for different histological PTCLs subtypes for development the first-line PET-adapted approaches in PTCL. The identified threshold of TMTV41% 230ml effectively stratifies patients into high and low-risk groups and suits as a more promising tool. Moreover, high TMTV has been associated with poor outcomes independently of clinical prognostic factors. Different segmentation methods are being analysed (SUV4). Further analyses is needed to fully understand how the metabolic activity may vary between different PTCLs subtypes.