Disturbance Region Removal Based Craniofacial Registration of Stereo Photography and CBCT Images

With rapid advancement in scanning technology, three-dimensional craniofacial imaging has secured its clinical acceptance and increasingly played an important role in improved planning and evaluation of pre-/post- orthognathic surgeries. 3D craniofacial imaging enables highly realistic facial and skeletal rendering, and can make more reliable measurements by taking into account of the depth information of the subject. Besides, creating a precise 3D replica of the head including hard and soft tissue structures has long been the goal of much research. Many 3D acquisition techniques exist, such as facial laser scanning, high-resolution 3D photogrammetry, and 3D cone-beam computed tomography. Facial laser scanning suffers its slow capturing and loss of the back of head information. 3D photogrammetry is fast, high-resolution, accurate for soft tissues, non-radioactive, but has only surface information. 3D CBCT reveals the whole volumetric structures, but has poor texture information for the soft tissues. Much previous research investigated ways to superimpose images from different modalities, in which 3D photogrammetry and 3D CBCT were among those most studied. The superimpositions or registrations were either manual or looking for features to match. In case of model match or morphing between two distinct subjects, the previous approaches would suffice. For post-surgical evaluation in which prominent features are usually altered, however, superimposition of such craniofacial images from disparate modalities exhibits great challenges. In this two-year project, we propose a novel disturbance region removal (DRR) approach to cope with the aforementioned difficulty. We propose a two-stage feature identification scheme. In the first stage, the most prominent features are extracted as disturbance features. Neighboring disturbance features form disturbance regions, which are then removed. In the second stage, we extract the secondary features, which then form secondary feature regions (SFR). The disturbance-regions-removed image with SFR information are then matched to obtain the final superimposition results. Our preliminary studies on five pairs of 3dMD photogrammetric and 3D CBCT images show significantly improved superimposition with DRR over those without (p＜0.01).

Project ID:PB10308-2698
External Project ID:MOST103-2221-E182-037