Fully Automated Segmentation of Maxillofacial Lesions Based on 3D Feature Descriptors and Symmetry Analysis

This paper presents a method for automatic segmentation of maxillofacial lesions from CBCT (cone beam CT) images based on 3D features and symmetry analysis. The proposed framework consists of three stages: Firstly, three classes of robust keypoint detection algorithms, i.e. SIFT (scale-invariant feature transform), SURF (speeded-up robust features) and BRISK (binary robust invariant scalable keypoints) are implemented. Subsequently, by matching symmetric keypoints, symmetric plane is detected and 3D volume image is divide into two roughly symmetric parts. Finally, by combining non-rigid registration, thresholding and morphological operators, maxillofacial lesion is segmented. The proposed framework has been validated on 125 CBCT datasets containing various cystic lesion and bone deformation. The mean dice similarity coefficient of 0.82 and 0.85 is achieved for cystic lesion and bone deformation, respectively. Moreover, 3D implementation of the proposed framework using GPU (Graphics Processing Unit), reduces the running time of algorithm and increases the feasibility of clinical application in practice.