PET Radiomics Discriminate Among Different Lymphoma Sub-Types

Le résumé fourni par la source

Despite numerous studies, the predictive role of radiomics extracted from 18F-fluoro-deoxy-glucose positron emission tomography/computed tomography (PET/CT) in lymphoma remains uncertain. In several solid cancers radiomics, and in particular the Haralick heterogeneity metrics (HM), have shown the ability to describe the phenotypes of tumors. This study investigates whether HM can differentiate between various lymphoma subtypes. We conducted an extensive analysis of a large cohorts of patients with diffuse large B-cell lymphoma (DLBCL) from the GOYA (NCT01287741), Hodgkin lymphoma (HL) from the HD0607 (NCT00795613), and follicular lymphoma (FL) from the FOLL12 (NCT02063685) clinical trials in which baseline PET was acquired and analysed per-protocol. Metabolic tumor volumes (MTVs) were semi-automatically delineated using a fixed SUV threshold of 4, as recommended in 2023 Menton meeting. Only MTVs larger than 50 ml, corresponding to a lesion radius of approximately 2 cm, were included in the analysis. HM were automatically extracted in Cuneo corelab. HM with a correlation coefficient lower than 0.90 and a variance higher than 0.95 were included in the analysis. Different artificial intelligence (AI) models were applied to training and test populations (70-30%). Data were available for 746 DLBCL, 285 HL, and 276 FL. The total number of lesions analysed were 1,918 for DLBCL, 607 for HL, and for 804 FL. The median MTV were 252, 81, and 90 ml, for DLBCL, HL, and FL, respectively. Among the AI, Random Forest (RF) showed the highest accuracy, 0.79 (95% CI: 0.76-0.81), on the test dataset. Sensitivity in identifying DLBCL, HL and FL were 0.93, 0.38 and 0.78. Specificity in identifying DLBCL, HL and FL were 0.86, 0.86 and 0.93. Radiomics has the potential to classify different lymphoma subtypes, when calculated on lesions large enough to mitigate the technical limitations of PET scanners. The discrimination is particularly high for DLBCL and FL. We anticipate that PET images acquired using new digital PET technology with dedicated reconstruction algorithms could further enhance the classification of lymphoma subtypes, thereby paving the way for their use also in patients prognosis definition.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.