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

Figure 5

(a) PCA succeeded to identify kidneys (Jaccard 0.54), whereas (b) it failed with Jaccard index 0.05. In both subfigures, the two upper panels include manual segmentations of the VOIs (red for heart, blue for brain, and cyan for kidneys), and the bottom panels visualise the PCA clusters representing corresponding VOIs (again, red for heart, blue for brain, and cyan for kidneys). The underlying grayscale image is the PET image. For visualisation purposes, the images were flattened into 2D format by taking the sum over the dimension missing from the 2D representation.

(a)

(b)