Radiomics analysis in cancer of the gastro-oesophageal junction: 18F-FDG-PET/MRI-derived quantitative features are associated with M-stage

Aim: To identify quantitative imaging biomarkers from multiparametric 18F-FDG-PET/MRI for predicting distant metastases at staging in patients with gastro-oesophageal junction (GOJ) cancer. Materials and methods: Following IRB approval and informed consent, 24 patients with histologically proven GOJ cancer were prospectively recruited; 4 patients were excluded for technical reasons. Finally, 19 male and 1 female (68.3±9.1 years) were included. Patients were injected with 326±28 MBq FDG intravenously. Uptake time was 120 minutes. Two experienced radiologists and nuclear physicians reviewed the images in consensus. All tumours were manually segmented using ImageJ software. Quantitative imaging features were extracted using an in-house software developed in Matlab. First-order and second-order statistical texture features were computed for PET SUV and MRI T1, T2, DWI and ADC images of the whole tumour volume. k-means clustering algorithm was used to assess the correlation of feature-pairs with the presence of distant metastases. Multivariate analysis of variance (MANOVA) was performed to assess the statistical separability of the groups identified by feature-pairs. Sensitivity (SE), specificity (SP), positive predictive value (PPV), negative predictive value (NPV) and accuracy (ACC) were calculated to quantify the discrimination ability of features in comparison with maximum standardized uptake value (SUVmax) and tumour size. Results: Second-order entropy derived from ADC and maximum probability derived from SUV, linked to texture irregularity and homogeneity respectively, were the best feature-pair for discrimination of patients with and without metastatic disease (SE=80%, SP=80%, PPV=80%, NPV=80%, ACC=80%). MANOVA confirmed that the means of these features for the two groups of patients differed significantly, with a p-value<0.001. SUVmax (SE=80%, SP=30%, PPV=53%, NPV=60%, ACC=55%) and tumour size (SE=90%, SP=10%, PPV=50%, NPV=50%, ACC=50%) performed worse, particularly for specificity. Conclusion: Our preliminary results suggest that greater intra-tumoural heterogeneity in GOJ cancer is associated with metastatic potential. The extraction of quantitative multiparametric features from the primary tumour may facilitate prognostication at staging, identifying patients who will benefit most from more frequent follow-up examinations and consideration of alternative therapies.