Predicting cancer organ tropism in mCRPC patients using triple-tracer PET imaging radiomics clustering from the 3TMPO cohort study.

Abstract

29 Background: The 3TMPO study has characterized the prevalence of intrapatient intermetastatic heterogeneity (IIH) in mCRPC using a Triple-Tracer PET Imaging approach (PMID 39327017). Artificial intelligence based radiomics clustering can help to further characterize both IIH and interpatient cancer heterogeneity. In this study we have exploited the triple-tracer radiomics signatures to cluster patients and predict cancer organ tropism. Methods: 3TMPO is a multisite prospective observational study in which 41 mCRPC patients underwent 18 F-FDG, 68 Ga- PSMA–617 and 68 Ga-DOTATATE PET/CT scans. Calculations for hierarchic clustering used 9 vectors (SUVmax 18 F-FDG, SUVmean 18 F-FDG, SUVpeak 18 F-FDG, SUVmax 68 Ga- PSMA–617, SUVmean 68 Ga- PSMA–617, SUVpeak 68 Ga- PSMA–617, SUVmax 68 Ga-DOTATATE, SUVmean 68 Ga-DOTATATE, SUVpeak 68 Ga-DOTATATE on a total of 355 (35% nodal, 34% bone, 31% visceral) unique lesions. The model was trained on 284 (80.0%) lesions and used the remaining 71 (20%) for internal validation. Results: Based on triple tracer radiomics in each metastasis, 5 clusters could be identified with a total variance of 69.4%. Cluster 2 (C2) showed a triple-tracer signature strongly suggestive (specificity: 79%) of visceral metastasis, as did cluster 1 (C1) (specificity: 66%). Lesions in cluster 4 (C4) had a specificity of 74% for nodal metastasis. Nodal lesions in C2 showed, in contrast to C4, a vector activity more resemblant of liver lesions, putting forward the hypothesis that nodal disease with a C4 signature could be biologically different from lymph nodes with a C2 signature. Patients with lesions who expressed a C2 signature had a decrease in overall survival (OS) (3.1 months, n=4) compared to patients with lesions in other (C1, C3, C4 & C5) clusters (7.3 months, n = 37) (p-value = 0.06). Conclusions: An internally validated model using the 3TMPO 18 F-FDG, 68 Ga- PSMA–617 and 68 Ga-DOTATATE PET/CT imaging vectors to identify organ tropism clusters is presented. Different triple tracer signatures could be identified, some of which are associated with a shorter OS. Signatures of lymphatic lesions in M1c disease differ from M1a/b cases, suggesting that triple-tracer imaging can highlight differences in tumour biology. These findings favour our hypothesis that radiomic clusters could be used to predict organ tropisms and prognosis to guide precision medicine treatment intensification.