Application of computer-aided image interpretation to the diagnosis of periapical bone lesions

An image analysis system was developed for the computer-aided diagnosis of periapical bone lesions in dental radiographs. The system was designed to (1) identify the periapical region, (2) determine the presence of a periapical lesion and (3) estimate the size of the lesion in cases when a lesion had been found. To initiate the procedure, an observer indicates an arbitrary point on the root in a digitized radiograph. From this initial point, the location of the radiographic projection of the apex of the root is automatically computed. Next, the trabecular bone pattern is detected through texture analysis. A local absence of the trabecular bone pattern in the periapical region is marked as a periapical bone lesion. When a lesion has been identified, its size is estimated based on local edge properties. Observer interaction is only allowed to adjust the result of the apex localization procedure if the apex has not correctly been localized. In an experiment with randomly selected radiographs of 111 mandibular roots, the performance of the system was tested against the consensual diagnosis of four expert observers. The sensitivity of the system to identify a lesion was 83.3%, the specificity 75.6% and the diagnostic accuracy 80.2%. The correlation between the size of the lesions as estimated by the system and by the observers was 0.67 (P < 0.01). When the procedure was repeated, the percentage of correctly reproduced lesion sizes by the system was 92.8%. The determination of the presence of a lesion was reproducible in 98.2% of all the cases.