Efficacy of two radiographic algorithms for detection of peri-implant bone defects on cone-beam computed tomography scans

Abstract

Background Early detection of peri-implant bone defects can improve long-term durability of dental implants. By the advances in cone-beam computed tomography (CBCT) scanners and introduction of new algorithms, it is important to find the most efficient protocol for detection of bone defects. This study aimed to assess the efficacy of metal artifact reduction (MAR) and advanced noise reduction (ANR) algorithms for detection of peri-implant bone defects. Materials and methods In this in vitro study, 40 titanium implants were placed in 7 sheep mandibles. Crestal, apical, and Full defects (n = 10 from each type) were created around the implants, and 10 implants were also placed as controls. CBCT scans were obtained in four modes: with MAR, with ANR, with both MAR and ANR, and without any filter. Totally, 28 scans were obtained and evaluated by a radiologist and a maxillofacial surgeon. The observers recorded their observations in a checklist, and data were analyzed by SPSS version 21 using the kappa coefficient of agreement, sensitivity and specificity values, area under the receiver operating characteristic (ROC) curve (AUC), intraclass correlation coefficient, t-test and paired t-test (P < 0.05). Results The inter-observer agreement was high for detection of all defects in all modes except with ANR. No significant difference was found in AUC and diagnostic accuracy of different scan modes (P > 0.05). The most common diagnostic error was related to misdiagnosis of control group with full defect with ANR filter, such that the existing bone was not detected. Defect depth was averagely over-estimated while defect length was under-estimated. Correct diagnosis of defects had the highest frequency when both filters were on. Conclusion The diagnostic accuracy and sensitivity for detection of different defect types were not significantly different in different scan modes but activation of ANR filter significantly decreased the specificity and positive predictive value compared with no use of filter.