Statistical contrast enhancement of subtraction images for radiographic caries diagnosis.

The effects of the nonlinear contrast-enhancement technique are examined in terms of diagnostic performance obtainable from subtracted dental radiographs for a variety of simulated exposures. Conventional bitewing radiographs of patients known to be free of caries were digitized by means of a computer interfaced with a conventional television camera. The resulting images were duplicated and stored in the computer. Radiolucencies similar in appearance to interproximal caries were simulated analytically in one set of the images. Reference images were superimposed spatially and subtracted from their counterparts containing the induced interproximal lesions after simulating the effects of quantum limited exposure on both sets. This was done for each separate image element independently by replacing original gray levels in each image with levels determined by a Poisson random deviate. The resulting difference images were contrast enhanced by a method which first smooths out local variations in gray level and then reassigns gray-level values in a way determined by the observed second-order spatial statistics. To aid in localization, these images were then again subtracted from the original noise-degraded pictures without lesions, rendering images similar to conventional radiographs but contrast enhanced. Observer performance by means of these enhanced images was compared with that produced from unenhanced-lesion-containing controls. The results suggest that enhancement increases the certainty with which diagnosis can be made and, further, that diagnostic accuracy can be improved in severely degraded images which simulate the effects of reduced levels of exposure.