A Diagnostic Accuracy Study of an Innovative Automatic Edge Detection Technique to Identify Simulated Implant Fractures on X-Ray Images


The null hypothesis (H0) was rejected because significant differences were observed between the original and Canny images in terms of fracture diagnosis.

Dental implant body fracture occurs rarely; however, it is one of the most serious complications of implant therapy. Several factors have been proposed as risk factors for implant fracture, including implant design and diameter, implant length, implant-abutment mismatch, treated area, long duration, the parafunctional activities, the occlusal overload and the type of prosthesis are the most important.2,25,26,27,28. In many cases, it is impossible to determine whether marginal bone resorption is a cause of implant fracture or a consequence thereof.12. This means that fracture of the implant may occur before bone loss in some cases and therefore it is necessary to diagnose the fracture in a timely manner to minimize the amount of bone defect that occurs around the fractured implant and hence the need for future augmentation procedures. . Several studies have been performed on the prevalence and associated risk factors for implant fracture1,11,29,30; Nevertheless, no attempt has been made so far to identify the best radiographic modality for accurate diagnosis of fractures in dental implants. Since the radiodensity and the geometric configuration of dental implants are very different from natural teeth, a study of the different radiographic methods is necessary to define the best technique.

To our knowledge, this is the first study comparing the diagnostic accuracy of various radiographic methods for the detection of sham fractures in dental implants. We evaluated CBCT and periapical radiographs with the first in two active and inactive modes of MAR and the second in two forms of parallel and oblique angle of X-ray projection. Simulated fractures were created in the cervical part of the devices since it is the most load-bearing area and therefore more susceptible to fractures31. However, it should be noted that the real fractures differ from the artificial fractures in our study in terms of size and location. True fractures are usually narrower and occur at the shoulder of the implant extending vertically downward2.

Another distinctive feature of this study was the application of an automatic edge detection algorithm (Canny algorithm) on radiographic images to determine if it aids in the diagnosis of fractured specimens.

Periapical radiographs are commonly used for postoperative evaluation of dental implants. Moreover, they are frequently used for the diagnosis of root fractures due to their excellent spatial resolution. However, the two-dimensional nature of these x-rays makes it almost impossible to diagnose fractures that are not oriented in the direction of the central x-ray beam. Thus, it is strongly recommended to take a second x-ray with a different angulation when a fracture is suspected16. In the present study, we evaluated PPA and OPA radiographs for the diagnosis of sham fractures. The PPA views were obtained with the central beam directed perpendicular to the implant and the image receptor while the OPA images were obtained by shifting the central beam by 20° in the horizontal plane. The diagnostic accuracy of both techniques was exactly the same when obtained in their native formats (AUC = 0.792). In the present study, the simulated fractures concerned three planes of the fixtures (one proximal and two facio-lingual); therefore, the similar accuracy of PPA and OPA radiographs might be related to the involvement of proximal surfaces that can be correctly displayed in both modalities. Saberi et al.32 reported that the use of OPA images significantly improves the visibility of fenestration and dehiscence defects when confined to the buccal aspect of dental implants. This finding further emphasizes the importance of the surfaces that are involved in fractures or bone defects around the implants.

CBCT images were taken in two modes MAR (on) and MAR (off), both modes having higher accuracy compared to PPA and OPA images. Previous studies have reported that CBCT images are superior to periapical radiographs for detecting root fractures16.33. In the present study, we found that the same fact also exists for sham implant fractures. Although interfering artifacts are produced near the titanium implants, it seems that the three-dimensionality of the CBCT images outweighs the adverse effects of these artifacts. However, enabling MAR mode obviously resulted in more accurate diagnosis of simulated implant fractures (AUC=0.917 for MAR in mode versus. AUC = 0.875 for MAR mode disabled). Kajan et al. and Candemil et al., also observed that the application of the MAR tool in CBCT images improves the detectability of vertical root fractures, especially when the teeth contain intracanal metal posts.34,35,36.

Computer-aided systems are gaining popularity in medical and dental imaging diagnostics. These techniques could be applied to almost all types of radiographic images to improve the accuracy and feasibility of different diagnostic tasks.37. A unique feature of the present study was the application of an automatic edge detection algorithm, i.e. the Canny algorithm for the identification of simulated implant fractures. The Canny edge detection algorithm is an accurate tool for defining the edge characteristics of an object as well as detecting abrupt intensity changes in an image. The algorithm runs on the MATLAB computing platform and has been used in medical imaging to express bone changes in osteoporosis and also quantification of artifacts in CT images.22.23. Canny edge detection algorithm benefits from being less sensitive to sources of image noise, including grayscale inhomogeneity caused by exposure conditions, patient position, and ambient temperature38. Since it has the ability to define intensity changes and sharp edges of an image with the least amount of noise, we decided to apply it on X-ray images to determine if it improves the accuracy of X-ray images for the diagnosis of implant fracture. Three parameters of the Canny algorithm, including Gaussian filter standard deviation, high and low sensitivity thresholds, can be manually adjusted by the operator. In this study, two expert radiologists who were not part of the observers defined the aforementioned parameters in the CBCT and periapical images by testing different values ​​until they both reached a point with the least noise and details. finest images.

By applying Canny’s edge detection algorithm on the images, we achieved higher diagnostic accuracy in PPA, OPA and CBCT MAR(on) images. This means that regardless of the two-dimensional or three-dimensional nature of the radiographic images, applying the Canny algorithm improves the visibility of simulated implant fractures due to the pronounced intensity changes it displays. Looking at the specificity values ​​in Table 1, we come to another important feature of Canny’s algorithm which is the application of this algorithm on X-ray images does not result in the number of false positive results.

AUC values ​​were similar for original and Canny formats of CBCT MAR (off) images. This finding could suggest a lower efficiency of Canny’s algorithm in the presence of metallic artifacts. Therefore, it can be concluded that the Canny algorithm creates more accurate images when metallic artifacts are reduced or eliminated.

By comparing the AUC values, it is perceived that the highest accuracy belongs to Canny CBCT with MAR, followed by Original CBCT with MAR, Original CBCT without MAR = Canny CBCT without MAR = Canny OPA, Canny PPA, PPA d ‘origin = original OPA, respectively . Pairwise comparisons also revealed statistically significant differences between Canny CBCT with MAR and original PPA (p= 0.005), Canny CBCT with MAR and Canny PPA (p= 0.028), and Canny CBCT with original MAR and OPA (p= 0.005).

There are also limitations to this study; the most important being the configuration of the fractures. Since this study was carried out in vitro, fractures were created with standardized contours. However, in real situations, fractures occur with more bizarre shapes that can influence diagnostic accuracy. In the present study, the simulated fractures were 0.3 mm wide and they were created horizontally at the cervical parts of the devices, whereas the real fractures are generally narrower, which may limit the voxel resolution of CBCT images. to display them correctly. Additionally, true fractures mostly occur at the shoulder of the implant, extending vertically downward.2. Thus, it is recommended that further clinical studies be conducted to define the effectiveness of the different radiographic modalities as well as the Canny edge detection algorithm for the diagnosis of true implant fractures.


Comments are closed.