Automatic Segmentation of Mandibular Condylar in Dental OPG Images Using Modified Mask RCNN

Abstract

Proper segmentation of the maxillofacial bones in OPG (OrthoPantomoGram) is vital for identification and prediagnosis planning for maxillofacial surgery. Traditional segmentation is time - consuming and demanding due to inherent properties of bones in the maxillofacial regions. Nevertheless, due to the large consistent dataset requirements of data driven segmentation techniques, such as deep learning, there is an impediment in their clinical applications. In this study, we proposed a modified Mask RCNN based Framework for the automatic and accurate segmentation of the condylar regions in Dental (OPG) images with limited datasets. This proposed technique comprises of three stages namely pre-processing, mask creations and segmentation. Initially the edges of the condylar region in dental (OPG) images are enhance using pre-processing filter subsequently the mask regions of the condylar has been created using polynomial approach then the mask images along with the original images are trained by the proposed deep network architecture finally the segmented condylar region is compared with the ground tooth images created by dental experts and achieves an accuracy of 87.24%. The results suggested that Modified Mask RCNN has segmentation performance that is comparable to other models and has better data compatibility.