Fig. 5

Illustration of method used to generate lesion feature maps in template space. Each subject's PET–CT image is first registered to template space (1). Voxel-, lesion- and subject-level features are extracted from the subject's lesion mask and used to construct lesion feature maps in subject space (2). Lesion feature maps are subsequently mapped to template space using image registration results (3). In template space, feature values are summarized across lesions sampled at each voxel location in the body (4)