Computer-based image analysis of natural approximal caries on radiographic films.

This paper reports on the development of computer software for automatic detection and measurement of approximal caries lesions on digital images obtained from radiographic films. If the program detects a lesion, it defines the lesion's border and measures and reports these lesion parameters: percent mineralization, area, maximum enamel penetration, and mean enamel penetration. Two experiments are reported; both were conducted with the use of extracted human premolar teeth. In the first experiment, thin sections were obtained from 13 approximal surfaces with caries lesions. Microradiographs were made of the sections, and software measurements of the lesions were obtained from the microradiographs. The software was able to separate the lesions into two groups: those that had penetrated less than 100% of the enamel and those that had penetrated 100% of the enamel. The software measurements agreed with evaluation of lesion penetration by polarized light microscopy of the thin sections. In the second experiment, simulated clinical radiographs were made of 16 approximal surfaces, 12 with lesions and four without. Thin sections were obtained for microradiography. Software was used for automatic detection and measurement of the lesions on both types of film. The ability of the software to detect the lesions on the clinical films was compared with that of 10 experienced clinicians. Correlations of the same parameter from the two types of radiographic films were statistically significant, 0.86 less than or equal to r less than or equal to 0.95 and p less than 0.05. The coefficients of variation were low for both microradiographs, 0.4-2.3%, and for clinical films, 0.7-8.5%.(ABSTRACT TRUNCATED AT 250 WORDS)