Craniomaxillofacial Deformity Correction via Sparse Representation in Coherent Space

Abstract

Patients with Craniomaxillofacial (CMF) deformities always get isolated deformed jaws with normal midfaces. Orthognathic surgery is specifically designed to correct jaws of the patients with normal midfaces. To improve performance of the surgery, a patient-specific normal reference jaw shape is necessary to guide surgical planning prior to the treatment. Sparse representation (SR) technique can be used to predict the reference jaw shape with guidance from the jaws of normal subjects with similar midfaces as the patient's midface. Current SR-based prediction method implicitly assumed that patient's midface shape was strongly correlated with his/her jaw shape. Unfortunately, the assumption is difficult to meet in practice due to different data distributions of midface and jaw. To alleviate this issue, we propose a patient-specific reference jaw shape prediction method via sparse representation in a coherent space mapped by canonical correlation analysis (CCA). Moreover, we iteratively refine the predicted reference jaw shape by a multi-layer mapping and refinement (MMR) scheme. Experimental results on clinical data containing 30 sets of computed tomography (CT) models of normal and patient subjects show that the proposed method predicts more accurate patient-specific reference jaw shapes for surgical planning than the state-of-the-art method.

Keywords

• Canonical correlation analysis
• craniomaxillofacial deformity
• multi-layer mapping and refinement
• sparse representation