We received the following question regarding the annotation process of the PET/CT data: "Did you use any thresholding, e.g. SUV4 or relative SUVmax, during the annotation process? Or was it done entirely manually?"

Overall, the annotation was done manually based on a visual inspection by one annotator for the FDG and PSMA data, respectively.

FDG data: The annotation process is described in A whole-body FDG-PET/CT Dataset with manually annotated Tumor Lesions: "CT and corresponding PET volumes were displayed side by side or as an overlay and tumor lesions showing elevated FDG-uptake (visually above blood-pool levels) were segmented in a slice-per-slice manner resulting in 3D binary segmentation masks."

PSMA data: The attenuation-corrected PET volumes with voxel values converted to standardized uptake values (SUV) and a corresponding CT were displayed side by side or as an overlay. Tumor lesions with elevated SUV were segmented in 3D space by drawing circular ROIs, in which voxels with SUV values above a user-defined threshold were presegmented automatically and then manually corrected slice by slice, resulting in 3D binary segmentation masks. The threshold was used to accelerate the manual annotation process and was not kept fixed but adapted individually for the images based on visual inspection.