Segmentation of Dynamic Total-Body [¹⁸F]-FDG PET Images Using Unsupervised Clustering

Figure 4

An example image with VOIs identified either manually (top-left panel), or using PCA (top-right panel), ICA (bottom-left panel), or GMM (bottom-right panel). The image is flattened into 2D from (a) top or (b) side view by taking the sum of voxel values over the dimension not visible in the 2D representation. The underlying grayscale image is the PET image, and the VOIs are plotted on the top with VOIs corresponding to heart, brain, and kidneys encoded with red, blue, and cyan, respectively. The Harderian glands are highlighted with a red arrow and brown fat with a blue arrow from the side view image (b) with the manual segmentation.

(a)

(b)