Savolainen S, Pohjonen H, Sipilä O, Liewendahl K
Department of Physics, University of Helsinki, Finland.
Nucl Med Commun. 1995 May;16(5):370-7. doi: 10.1097/00006231-199505000-00009.
Objective determination of regions of interest (ROIs) is a prerequisite for the accurate quantification of radionuclide volume distributions in single photon emission tomographic (SPET) images. In this study, we compared four segmentation methods: fixed thresholding (FT), grey level histogram (GL), region growing (RG) and combined region growing and edge detection (RGE). For this purpose, an elliptical phantom containing two cylinders with varying volumes (8-360 ml) and activities of 111In and 99Tcm (2.9-37 kBq ml-1) was employed. Using these methods, the following correlation was observed between true and measured phantom volumes: 111In, r = 0.95 (FT), 0.73 (GL), 0.93 (RG) and 0.92 (RGE); 99Tcm, r = 0.85 (FT), 0.72 (GL), 0.85 (RG) and 0.89 (RGE). Volume determination with FT and RG was not sensitive to the cut-off frequency used in image filtering. A significant correlation was observed between spleen volumes measured with the different segmentation methods, except GL, when applied to the SPET images of 25 patients administered 111In-labelled platelets. On the basis of these results, FT and RG are recommended for the clinical determination of ROIs, although they can be difficult to apply if the signal-to-noise ratio is very low or highly variable, when a combination of different imaging modalities may be the only accurate solution to the segmentation problem. The RGE method can also produce accurate results, but estimation of parameters is laborious with this method. Before being applied clinically, all segmentation methods tested in this study require phantom measurements for the determination of optimal parameters.
