A diagnostic accuracy study on an innovative auto-edge detection technique for identifying simulated implant fractures on radiographic images

Abstract Implant fracture is a rare but devastating complication of treatment in partially or fully edentulous patients which requires prompt diagnosis. Nevertheless, studies on defining the most accurate technique for the detection of implant fractures are lacking. In the present study, the Canny edge detection algorithm was applied on multiple radiographic modalities including parallel periapical (PPA), oblique periapical (OPA), and cone beam CT (CBCT) with and without metal artifact reduction (MAR) to examine its accuracy for diagnosis of simulated implant fractures. Radiographs were taken from 24 intact implants and 24 implants with artificially created fractures. Images were evaluated in their original and Canny formats. The accuracy of each radiograph was assessed by comparison with a reference standard of direct observation of the implant. The greatest area under the receiver operating characteristic curve belonged to Canny CBCT with MAR (0.958), followed by original CBCT with MAR (0.917), original CBCT without MAR = Canny CBCT without MAR = Canny OPA (0.875), Canny PPA (0.833), original PPA = original OPA (0.792), respectively. The Canny edge detection algorithm is suggested as an innovative method for accurate diagnosis of clinically suspected implant fractures on CBCT and periapical radiographies.