Application of statistical decision theory to radiographic diagnosis of endodontically treated teeth

The presence or absence of periapical radiolucencies, after a predetermined healing period, have been used as a criteria for therapeutic success or failure in endodontic treatment. However, in radiologic diagnosis it has been shown that variations within and between examiners are substantial. In the present study six endodontists evaluated the periapical tissues in radiographs of 119 endodontically treated roots using a five-graded rating scale. An earlier study on the same radiograph material served as reference and "true" states of the periapical tissues were established. True positive and false positive reports on the presence of periapical lesions could then be calculated. For each observer it was noticed that a higher true positive percentage always was coupled with a higher false positive percentage and vice versa. If pairs of true positive and false positive percentages were plotted on a two-dimensional graph they corresponded well to a Receiver Operating Characteristic (ROC) curve. Variations between the observers could be explained by their adoption of different criteria of periapical disease resulting in different positions on the ROC curve. Influence of observer variation on reported frequencies of periapical lesions was greater, the lower the prevalence of the disease. The best opportunities for revealing relative differences in disease prevalences was created when the examiner defined a strict criterion for disease and reported a positive finding only when absolutely certain.