Search

Background More accurate predictive biomarkers in patients with gastroenteropancreatic neuroendocrine tumours (GEP-NETs) are needed. This study aims to investigate radiomics-based tumour phenotypes as a surrogate biomarker of the tumour vasculature and response prediction to antiangiogenic targeted agents in patients with GEP-NETs. Methods In this retrospective study, a radiomics signature was developed in patients with GEP-NETs and liver metastases receiving lenvatinib. Patients were selected from the multicentre phase II TALENT trial (NCT02678780) (development cohort). Radiomics variables were extracted from liver metastases in the pre-treatment CT-scans and selected using LASSO regression and minimum redundancy maximum relevance (mRMR). Logistic regression and Cox proportional-hazards models for radiomics and combined radiomics with clinical data were explored. The performance of the models was tested in an external cohort (test cohort). Associations between the radiomics score and vascularization factors in plasma were studied using hierarchical clustering and Mann-Whitney U test. Results A total of 89 patients were included in the study, 408 liver metastases where analysed. The CT-based radiomics signature was associated with clinical benefit in the development (training and validation sets) and test cohorts (AUC 0.75 [0.66-0.90], 0.67 [0.49-0.92] and 0.67 [0.43-0.91], respectively). The combined radiomics-clinical signature (including the radiomics score, Ki-67 index and primary tumour site) improved on radiomics-only signature performance (AUC 0.79 [95% CI 0.64-0.93]; PConclusions Radiomics-based phenotypes can provide valuable information about tumour characteristics, including vasculature, that are associated with response to antiangiogenics. Clinical Trial Registration This is a study of the Lenvatinib Efficacy in Metastatic Neuroendocrine Tumours (TALENT) phase II clinical trial (NCT02678780).

Non-invasive differential diagnosis of brain tumours is currently based on the assessment of tumour vascularity through magnetic resonance imaging (MRI) coupled with dynamic susceptibility contrast (DSC). However, given its limited accuracy, reaching a definitive diagnosis often requires complex neurosurgical interventions that compromise the patients’ quality of life. We applied deep learning on DSC images from histology-confirmed patients with glioblastoma, metastasis or lymphoma, the three most common brain malignancies. The convolutional neural network trained on ~ 50,000 voxels from 40 patients provided intra-tumour probability maps that yielded clinical-grade diagnosis. Performance was tested in 400 additional cases and an external validation cohort (n = 128). The tool reached a three-way accuracy of 0.78, superior to standard diagnosis with cerebral blood volume (0.55) and percentage of signal recovery (0.59) perfusion metrics. Our open-access software, Brain Enhancing Region Radiological analysis (BERRY), demonstrates the potential of voxel-wise probability maps for differential diagnosis of brain tumours using standard-of-care MRI.