O49. A comparison of two model fitting methods for transferring mesh correspondences: Implications to scapular bone using statistical shape modelling

Abstract

Introduction Use of Statistical Shape Models (SSMs) of various shapes in the medical field (diagnosis, morphometric analysis, surgical interventions, etc.) has been promising. To conduct such analysis, an SSM must establish a correspondence between itself and the target shape. This study was focused on comparing two mesh-based fitting methods in order to establish point-to-point correspondence between shapes using the SSM as a prior knowledge. Materials and methods CT scans of 27 dry scapulae were first used to build the SSM of scapula bone using an IMCP-GMM pipeline proposed earlier (Mutsvangwa, 2015). The scapula SSM quality was tested using generality, specificity and compactness criteria. The fitting method was conducted using an open-source software called Scalismo. The fitting process was initiated with a landmark-based alignment step, followed by a rigid alignment using Iterative Closest Point (ICP) algorithm between a target mesh and the SSM and then using two different fitting model methods: (A) ICP non-rigid registration, and (B) Parametric registration with L-BFGS optimizer. The fitting quality from the two methods was tested on two sets of targets (internal: from the SSM learning base and external: not from learning base, four targets each). Correspondence quality was evaluated using the root mean square distance (RMS) between the same indices. Results The internal targets had superior fitting quality (method A: mean distance (MD):(0.42 ± 0.03) mm, RMS:(0.57 ± 0.02) mm; method B: MD:(0.44 ± 0.0.03) mm, RMS:(0.58 ± 0.01) mm) than the external targets (Method A: MD:(1.16 ± 0.21) mm, RMS:(1.29 ± 0.25) mm; method B: MD:(1.17 ± 0.21) mm, RMS:(1.32 ± 0.31) mm). Good correspondence quality using both the methods was achieved, with method A (CorrRMS:(1.46 ± 0.34) mm) performing slightly better than B (CorrRMS:(1.56 ± 0.31) mm). Conclusion Scalismo is an efficient toolbox for SSM building as well as for benchmarking. Although both the methods were effective, more evaluations would be necessary by changing the various parameters in the fitting process or by increasing the compactness of the SSM.