Shape model-based point set registration with local density weighting applied to the incomplete femur

Abstract

Pre-operative planning aids clinicians by providing insight into patient-specific problems before surgery is performed. For treatment of femoroacetabular impingement (FAI), a range of motion (ROM) simulation is used, based on collision detection on segmented CT or MRI-data. Imaging data of the femur is often incomplete, due to the fact that radiologists want to minimize radiation dose and scanning time. The lack of data negatively influences the range of motion simulation. In this research we present a statistical shape model (SSM) based point set registration method that is able to accurately register to incomplete femur data. It is based on an iterative schema that minimizes the squared distance between the target data and the modes of variation in the SSM. We introduce a weighting vector that is derived from point density in the target data to force each point to equally contribute to the final solution. This greatly improved registration performance in the distal femur region. Furthermore, we introduced an iteratively increasing amount of modes of variation in the solving approach. This lead to faster convergence while achieving more accurate registration. The implemented method outperforms current state-of-the-art registration algorithms.