Digital filtering of bone scans: an ROC study.

The aims of this study were to investigate if digital filtering can increase the sensitivity and specificity of bone scans, and to find the type of filter most suitable for bone images at various count levels. We also wished to examine if filtering allows the administered activity or the examination time to be reduced, and if it is easier to detect low-contrast uptake using a digital filter. Images containing a total of 100, 350 and 1000 kcounts were acquired in a 256 x 256 matrix using a transmission phantom simulating the thoracolumbar region. Receiver operating characteristic (ROC) analysis demonstrated that digital filtering increases the sensitivity and specificity of bone scintigraphy. A low-pass filter for images with low statistics (100 kcounts), which contain 2-5 counts per pixel in the ribs and vertebrae respectively, and a Metz filter for images with normal (5-19 counts per pixel) to good (20-54 counts per pixel) statistics, increase the area under the ROC curve significantly (99% confidence level) compared to unfiltered images. Filtering also increases the detectability of low-contrast objects compared with unfiltered images. Digital filtering might be an alternative to raising the number of counts in the image. An alternative is to reduce the administered activity and hence the effective dose to the patient, or to reduce the examination time-which is an advantage when treating elderly patients or patients with pain.