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

Table 1

Summary of target applications of different unsupervised segmentation methods for PET images. Column “Studied area” indicates the body part segmented in the reference in the last column. Columns “Dynamic,” “Dimensions,” and “Organism” state if the method was designed for dynamic data including multiple time points, number of spatial dimensions of the images to be segmented, and the species used as an example in the reference, respectively. Due to the lack of access to the full studies [16, 21], some information is missing for them. Notably, the dimensions and dynamic/static nature indicated here refer to the final input for the segmentation as used in the reference. It may differ from the original raw images, e.g., dynamic images may be summed over time points prior to segmentation to mimic a static image, or 3D image can be sliced into 2D images.

Studied area	Dynamic	Dimensions	Organism	Reference
Multiorgan phantom, real brain, and real lung	Yes	2D	Phantom and human	[6]
Brain	No	3D	Human	[7]
Brain (real and phantom)	Yes	3D	Phantom and human	[8]
Brain (real and phantom)	Yes	2D	Phantom and rat	[9]
Brain (real and phantom)	Yes	2D	Phantom and rat	[10]
Simulated, brain	Yes	2D	Artificial and human	[11]
Brain	Yes	2D and 3D	Human	[12]
Brain	Yes	3D	Human	[13]
Brain	Yes	2D	Human	[14]
Brain (real and phantom)	Yes	3D	Phantom	[15]
Brain	Yes	—	Human	[16]
Lung, breast	No	2D	Human	[17]
Several	No	3D	Phantom, human, and rabbit	[18]
Total body (real and phantom)	Yes	3D	Phantom and rat	[19]
Prostate	Yes	3D	Human	[20]
—	Yes	3D	Mouse	[21]