A technique is presented for segmented attenuation correction in posit
ron emission tomography (PET) based on the local thresholding techniqu
e (LTS) described previously. To accommodate the substantially differe
nt body sections encountered in whole-body PET, an adaptive thresholdi
ng has been added to yield more uniform results throughout the body. B
y evaluating the intensity distribution of a set of transverse transmi
ssion images, the algorithm determines an optimal threshold for separa
ting two or three different groups or classes of similar pixels. Inter
class variance is maximized and intra-class errors are minimized. The
algorithm also switches automatically between a three-class mode (back
ground, lungs/air pockets, soft tissue) and a two-class mode (backgrou
nd, soft tissue), thereby achieving more uniform segmentation where lu
ng spaces and bowel air pockets are alternately present then absent fr
om the volume of interest. The addition of adaptive thresholding virtu
ally eliminates the need for operator intervention. The clinical imple
mentation requires short-duration, count-limited, transmission images
that would otherwise be too noisy for direct attenuation correction, E
mission images corrected with adaptive LTS were shown to be equivalent
, both quantitatively and qualitatively, to those corrected using conv
entional measured attenuation correction.