Treating Osteoarthritis in Knee with Mimicked Interpositional Spacer
Data selection
In order to detect different clusters of similar shapes at least 500 MRI datasets need to be processed. The datasets are taken from The Osteoarthritis Initiative (OAI) database. The OAI database contains about 5000 patients. Therefore, a selection has been made based on the KellgrenLawrence OAScore which is available for almost all patients. The KellgrenLawrence score differentiates between five grades. For each of the five grades 138 female and 145 male patients have been randomly selected resulting in a preselection of 1415 right knee MRI datasets. In a first step 500 of these datasets are processed and analysed. As MRI protocol SAG_3D_DESS_WE (sagittal 3D dualecho steady state with selective water excitation) is used for bone and cartilage segmentation.
Data processing
Bone and cartilage of distal femur and proximal tibia are segmented automatically using linear Statistical Shape Models (SSMs) [1]. Errors in the automatic segmentation are corrected manually. Additionally, for each knee landmarks of the insertion sites of anterior cruciate ligament (ACL) and posterior cruciate ligament (PCL) are placed by hand. A schematic overview is given in figure 1.
The following pictures illustrate the single steps of the pipeline.
Analysis
The aim is to find geometrical clusters in (very) high dimensional data. This fact complicates meaningful clustering. Therefore reduction of dimensionality/feature selection has to be done. Taking the geometries as triangular meshes there are different options how to proceed. One is the well known principal component analysis (PCA) on the vertex coordinates, that gives the SSMs, also known as Point Distribution Models (PDM). A more general approach is Principal Geodesic Analysis (PGA) [Ref Fletcher], which treats data as elements in a Riemannian manifold and carries out statistical operations therein or in the tangential space to a reference element respectively. This procedure has the advantage that it can be sensible to nonlinear shape changes depending on the chosen modeling space. It should be noted that PCA can be interpreted as special case of PGA, namely with the flat Euclidean space as underlaying manifold. We develop a method that employs the deformation gradient of a mesh modeled as element of a Lie group and thus delivering an even richer mathematical structure. It is called the Gradient Domain Model (GRM).
OA related changes of distal femur
The distal femur exibits clear pathological changes in case of a severe OA and is therefore an interesting structure to take into consideration for Clustering and Classification of OA.
Classification and Clustering
To find clusters we apply our GDM on patients that are healthy (KL 0/1) and those that are severely deseased (KL 4). The resultant shape weights are used to train a Support Vector Machine in order to classify between this two groups.
References
[1] Seim, H., Kainmueller, D., Lamecker, H., Bindernagel, M., Malinowski, J., & Zachow, S. (2010). Modelbased autosegmentation of knee bones and cartilage in MRI data. Proc. MICCAI Workshop Medical Image Analysis for the Clinic  A Grand Challange, 215223.
Publications
2017 

David Wilson, Carolyn Anglin, Felix Ambellan, Carl Martin Grewe, Alexander Tack, Hans Lamecker, Michael Dunbar, Stefan Zachow  Validation of threedimensional models of the distal femur created from surgical navigation point cloud data for intraoperative and postoperative analysis of total knee arthroplasty  International Journal of Computer Assisted Radiology and Surgery, 2017 (accepted for publication) 
BibTeX

2016 

Christoph von Tycowicz, Felix Ambellan, Anirban Mukhopadhyay, Stefan Zachow  A Riemannian Statistical Shape Model using Differential Coordinates  ZIBReport 1669 
PDF
BibTeX URN 
Hans Lamecker, Stefan Zachow  Statistical Shape Modeling of Musculoskeletal Structures and Its Applications  Computational Radiology for Orthopaedic Interventions, Springer, pp. 123, 2016, isbn: 9783319234816 
BibTeX
DOI 