Limitations of the digital image subtraction technique in assessing alveolar bone crest changes due to misalignment errors during image capture.

The digital image subtraction method depends critically on the very precise registration of sequential radiographs. A method of using a single radiograph to form two identical digital images was created, eliminating any potential effects from irradiation or processing variation, to investigate the effects of misregistration between images. These images were displaced by 0.1-0.42 mm in the X, Y and XY directions before subtraction. A total of 35 alveolar crest margins from five bitewing radiographs were investigated for the extent of false bone loss or gain produced in the subtracted images. The mean grey level gradient across the interdental space/crestal margin interface was 3.1 (SD 1.4) grey levels per 0.05 mm pixel. Small displacements between subtracted images of 0.1-0.14 mm in the Y or XY directions caused 20-25% of crestal pixels to vary by greater than or equal to +/- 2.5% of the grey range. Larger displacements of 0.3-0.42 mm caused 65% of crestal pixels to vary by greater than or equal to +/- 2.5% of the grey range. A higher threshold of greater than or equal to +/- 4.1% of the grey range still showed up to 48% of crestal pixels were in this higher deviation band. A high noise threshold of +/- 8% of the grey range should be used to discriminate against false grey values. A visual method for estimating the direction of displacement error between images is discussed together with a method for setting noise thresholds for individual alveolar crest grey level gradients. In any radiograph, a range of alveolar crest gradients will be found and those with the higher gradients will produce the largest subtraction errors for any given displacement between images.