Deformable registration of lateral cephalogram and cone?beam computed tomography image

Purpose: This study aimed to design and evaluate a novel method for the registration of 2D lateral cephalograms 3D craniofacial cone-beam computed tomography (CBCT) images, providing patient-specific structures from cephalogram without additional radiation exposure. Methods: We developed cross-modal deformable model based on deep convolutional neural network. Our approach took advantage low-dimensional deformation field encoding an iterative feedback scheme infer coarse-to-fine volumetric deformations. In particular, we constructed statistical subspace fields parameterized nonlinear mapping function image pair, consisting target reference CBCT, latent field. Instead one-shot by learned function, was introduced progressively update deformations fields, accounting shape variations anatomical structures. A total 220 clinically obtained CBCTs were used train validate proposed model, among which 120 generate training dataset with 24k paired synthetic CBCTs. The evaluated 2D–3D growing adult orthodontic patients. Results: Strong structural consistencies observed between deformed CBCT in all criteria. achieved state-of-the-art performances mean contour deviation 0.410.12 mm anterior cranial base, 0.480.17 mandible, 0.350.08 maxilla, respectively. surface mesh ranged 0.78 0.97 various structures, LREs 0.83 1.24 datasets regarding 14 landmarks. handled details improved registration. resultant consistent both projective planes space multicategory Conclusions: results suggest that learning-based enables alignment estimates