Application of image segmentation for implant bone level measurement in periapical standardized radiographs

Background: The identification of variations in bone levels (BL) from one radiographic examination to the next is time-consuming and highly dependent on examiner training. Traditional measurements rely on the identification of the first bone-to-implant contact, which is prone to inter-examiner variability. This study validates a digital method for detection of BL around dental implants. Material & Methods: 1. Image segmentation was based in a series of automatized steps to isolate the crestal bone and the implants aiming to compute the bone-implant intersection. They consist in the application of filters, edge detectors and histogram thresholding, proceeded by a series of morphological and bridge operators. It finishes with the application of previously trained Active Shape Models representative of the implant profile that adapt by iteraction to the enhanced structures in the image. BL are computed by calculation of the distance between the intersection of the implant and crestal bone and the implant shoulder. 2. Method validation: Two examiners analysed 60 radiographs of Camlog® Screw-line implants with a graphics user interface developed for the segmentation method. The results of the computed BL were analysed for disagreement between observers for internal validation. Accuracy was determined by comparison of the results with the matching manual readings. Results: No differences were found between the measurements obtained manually and by either of the two examiners using the automatized method F(2, 186)= 0.16, p=0.852. Reliability analysis of the measurements obtained by the three raters revealed an intraclass correlation coefficient of 0.839 [0.783-0.884, 95% CI], p0.01. More than 60% of the measurements were considered perfect hits. Conclusions: The new proposed method has proven to be a reliable and accurate tool for BL measurement, contributing for the reduction of inter-examiner variability. Keywords: implant, bone level, software, image segmentation