Additional file 1 of Tumor morphology on CT radiomics is largely driven by the local anatomical environment, not the primary tumor type

Additional file 1: Supplementary Figure S1. Consort flow diagram, depicting the inclusion process in the discovery cohort. Supplementary Figure S2. Example of original and eroded segmentations. Contrast-enhanced CT scan (portal venous phase with soft tissue windowing) showing a liver metastasis in a colorectal cancer patient from the discovery cohort. In the yellow outline, the original segmentation is shown, which is meant to cover the full lesion. In solid red, an eroded segmentation is shown, with one voxel systematically removed from the periphery. Supplementary Figure S3. Clustering analysis exclusively on central tumor regions. (A) Lesions labeled by tumor environment and outlined per cluster. (B) Lesions labeled by tumor type and outlined per cluster. Supplementary Table S1. Included publicly available datasets within the validation cohort. Supplementary Table S2. Acquisition parameters for the discovery and validation sets. Continuous variables are reported as median values with interquartile ranges, while categorical variables are presented as absolute counts and corresponding percentages. Model names occurring in less than 1% of their respective datasets are grouped under the category “Other.”