Quantitative analysis of the early peri-implant bone loss and formation using longitudinal intraoral radiographs

Abstract

Longitudinal intraoral radiographs are routinely used to assess bone-implant interface after dental implant placement. Bone loss in peri-implant bone has been measured to assess the degree of bone loss. However, bone patterns have not been thoroughly analyzed although they are strongly associated with the strength of bone-implant interface. Therefore, we sought to investigate the early change of bone patterns in peri-implant bone. Peri-implant bone images were sampled longitudinally, or after dividing into coronal vs. apical and proximal vs. distal parts by placing region of interests (ROI) in the intraoral radiographs. Image normalization and registration were performed as preprocessing. Then, ROI images were processed by erosion, tophat operation, and recombined in multiple scales. Intensity and texture features were calculated from the grayscale image within the ROI. Complexity features computed from the binary mask image and skeletonized image. Orientation features were computed from gradient map of grayscale ROI image. In the longitudinal analysis, complexity feature (the number of branching point per ROI) and texture feature (Inverse difference moment) were increased but texture features such as difference of entropy and information measure of correlation 1 were decreased between 4 weeks and 8 weeks after implant placement (n=45, Wilcoxon pairwise signed rank test, p<0.025). Peri-implant bone intensity increased between 8 weeks and 20 weeks (p<0.015). Thickness of peri-implant bones gradually increased over 25 weeks (Pearson’s analysis, ρ = 0.139, p<0.015). In the comparison of the coronal and apical parts of the implant, all features were significantly different at 4 weeks after implant (p<0.025). However, complexity feature (length of bone marrow per ROI) (p>0.06) was not different at 8 weeks. Furthermore, complexity feature such as thickness (p>0.09) and star volume of bone marrow (p>0.045) and texture feature (inverse difference moment) (p>0.075) were...