Computer-assisted subtraction radiography in periodontal diagnosis.

In the diagnosis of periodontal disease radiography serves as an adjunct to clinical methods to determine the extent of bone involvement. At subsequent examinations radiography is used to e.g. evaluate the course of disease and the effect of treatment. Differences between examinations due to biological changes often occur against a complex background of stationary anatomical structures and the signals of diagnostic interest are therefore often concealed in structured noise. This imposes severe constraints on the radiographic diagnosis of changes in the periodontal bone over time. A method was developed to reduce the amount of structured noise by subtracting identical image features in serial radiographs which had been digitized, using a high quality TV-camera interfaced with an analog to digital converter and a computer. When the technique was applied to standardized radiographs taken before and after small lesions had been made in the marginal crest of dry human mandibles, observers were able to detect those lesions more accurately than in the conventional radiographs from which the computer-assisted subtraction images were produced. The sensitivity increased from 51% to 91% while the specificity remained unchanged at about 95-96% High geometric reproducibility between radiographs to be subtracted is essential, but serial radiographs taken with angulation differences of 3 degrees yielded a higher degree of diagnostic accuracy, evaluated by the ROC-curve technique, than optimally taken conventional radiographs. Defects, slightly less than a mm wide, made in the marginal crest of dry human mandibles and with a depth in the direction of the radiation beam corresponding to 0.49 mm of compact bone could be detected with a close to perfect accuracy (ROC-evaluation) in subtraction images. Conventional radiographs required lesions more than three times as deep to render a similar degree of diagnostic accuracy. The actual lesion depths were assessed by the Iodine-125 absorptiometry technique. To standardized serial radiographs, taken before and after small marginal defects were induced, various amount of random noise was superimposed by adding to the existing grey-level of each picture element in the digitized radiographs a Poisson random deviate. Resultant noise-degraded images correspond to images produced by from faster imaging systems and it was shown that the application of computer-assisted subtraction yielded a high degree of diagnostic accuracy even when severely noise-degraded images were used. This indicates the potential usefulness of much faster imaging technology than presently used.(ABSTRACT TRUNCATED AT 400 WORDS)