Sparse dictionary learning for 3d craniomaxillofacial skeleton estimation based on 2D face photographs

Precisely estimating patient-specific reference bone shape models is important for the surgical planning of patients with craniomaxillofacial (CMF) defects. In this chapter, we introduce an automated method based on sparse dictionary learning for this purpose. This method combines pre-traumatic conventional portrait photographs and posttraumatic head computed tomography (CT) scans for reference 3D CMF skeleton estimation. Specifically, based on the CT images of training normal subjects, a correlation model between the facial and bony surfaces is constructed via sparse dictionary learning. Then, for a patient with large-scale defects (e.g., caused by trauma), a three-dimensional (3D) face is first reconstructed from the patient's 2D pre-traumatic portrait photographs. By feeding the reconstructed 3D face into the correlation model, an initial reference shape model for the patient is generated. After that, the initial estimation is refined by applying nonrigid surface matching between the initially estimated shape and the patient's posttraumatic bone based on the adaptive-focus deformable shape model (AFDSM). Furthermore, a statistical shape model, built from training normal subjects, is utilized to constrain the deformation process to avoid overfitting during refinement. This method has been evaluated using both synthetic and real patient data. Experimental results show that the patient's abnormal facial bony structure can be recovered, which is considered clinically acceptable by an experienced CMF surgeon.