The reduction of inter- and intra-observer variability for defining regions of interest in nuclear medicine

The use of region-of-interest (ROI) techniques to quantify data obtained in radionuclide images is commonplace. However, the reproducibility of quantitation due to inter- and intra-observer variations using particular methods of deriving ROIs is often not appreciated. We examined such variations in the results obtained by four independent observers of varying experience using four methods of depicting a ROI about an organ. The set of image data consisted of renal scans with varying target-to-background ratios, and the ROI facilities included two edge-detection methods. The results indicated that, once observers were experienced with edge-detection methods, a lower inter- and intra-observer variation could be achieved, although the technique of 'shrinking' a ROI about a subjectively chosen display level was reasonably satisfactory. In terms or reproducibility, the least satisfactory method of depicting a ROI was the commonly used manually guided 'bug' around arbitrarily chosen display levels representing the boundary of an organ.