RESUMEN
<p><b>OBJECTIVE</b>To establish and evaluate a computer-based software system (caries diagnosing system, CDS): CDS for better detection of initial approximal caries based on the performance of radiography.</p><p><b>METHODS</b>A total of 190 approximal surfaces from 95 extracted posterior teeth were examined by bite-wing radiography. After analysing and extracting the approximal lesion's characteristics of X-ray performance, using Matlab and VC language, the CDS was established. Then the proximal surfaces of these teeth were detected automatically by CDS, and scored by naked eyes of 8 experienced dentists for the extent of lesions. The histological appearance of sectioned teeth was used as the gold standard to assess sensitivity, specificity, accuracy, and reproducibility.</p><p><b>RESULTS</b>The sensitivity of CDS was 0.728, Pearson's correlation was 0.722, better than that of optic estimation. The reproducibility of CDS was good too.</p><p><b>CONCLUSIONS</b>The CDS can report the lesion site and extension, its diagnostic level is better than that of optic estimation, it is an reasonably sensitive and accurate method for the detection of initial approximal caries in present clinic practice, and a tool of providing objective data.</p>