The consideration of the individual knee ligament attachments is crucial for the application of patient specific musculoskeletal models in the clinical routine, e.g. in knee arthroplasty. Commonly, the pre-operative planning is based on CT images, where no soft tissue information is available. The goal of this study was to evaluate the accuracy of a full automatic and robust mesh morphing method that estimates locations of cruciate ligament attachments on the basis of training data.

The cruciate ligament attachments from 6 (n=6) different healthy male subjects (BH 184±6cm, BW 90±10kg) were identified in MRI-datasets by a clinical expert. The insertion areas were exported as point clouds and the centres of gravitation served as approximations of the attachments. These insertion points were used to annotate mean shapes of femur and tibia.

The mean shapes were built up from 332 training data sets each. The surface data were obtained from CT scans by performing an automatic segmentation followed by manual cleaning steps. The mean shapes were computed by selecting a data set randomly and aligning this reference rigidly to each of the remaining data sets. The data were fitted using the non-rigid ICP variant (N-ICP-A). Due to this morphing step, point correspondences were established.

By morphing a mean shape to the target geometries, including the cruciate ligament attachments, the distribution of the insertions on the original mean shape was obtained. Subsequently, a statistical mean was computed (annotated mean). The annotated mean shape was again morphed to the target data sets and the deviations of the respective predicted insertion points from the measured insertion points were computed.

The training data was successfully morphed to all 6 subjects in an automatic manner with virtually no distance error (10-5 mm). The mean distance between the measured and morphed ligament attachments was highest for the ACL in the femur (4.26±1.48 mm) and lowest for PCL in the tibia (1.63±0.36 mm). The highest deviation was observed for femoral ACL (6.93 mm).

In this study, a morphing based approach was presented to predict origins and insertions of the knee ligaments on the basis of CT-data, exemplarily shown for the cruciate ligaments. It has been demonstrated, that the N-ICP-A is applicable to predict the attachments automatic and robust with a high accuracy. This might help to improve patient-specific biomechanical models and their integration in the clinical routine.

For a proper functional restoration of the knee following knee arthroplasty, a comprehensive understanding of bony and soft tissue structures and their effects on biomechanics of the individual patient is essential. A systematic description of morphological knee joint parameters and a study of their effects could beneficial for an optimal patient-specific implant design.

The goal of this study was the development of a full parametric model for a comprehensive analysis of the distal femoral morphology also enabling a systematic parameter variation in the context of a patient specific multi-parameter optimisation of the knee implant shape.

The computational framework was implemented in MATLAB and tested on 20 CT-models which originated from pathological right knees. The femora were segmented semi-automatically and exported in STL-format.

First, a 3D surface model was imported, visualised and reference landmarks were defined. Cutting planes were rotated around the transepicondylar axis and ellipses were fitted in the cutting contour using pattern recognition. The portions between the ellipses were approximated by using a piecewise cubic hermite interpolation polynom such that a closed contour was obtained following the characteristics of the real bone model. At this point the user could change the parameters of the ellipses in order to manipulate the approximated contour for e.g. higher-level biomechanical analyses. A 3D surface was generated by using the lofting technique. Finally, the parameter model was exported in STL-format and compared against the original 3D surface model to evaluate the accuracy of the framework

The presented framework could be successfully applied for automatic parameterisation of all 20 distal femur surface data-sets. The mean global accuracy was 0.09±0.62 mm with optimal program settings which is more accurate than the optimal resolution of the CT based data acquisition. A systematic variation of the femoral morphology could be proofed based on several examples such as the manipulation of the medial/lateral curvature in the frontal plane, contact width of the condyles, J-Curve and trochlear groove orientation.

In our opinion, this novel approach might offer the opportunity to study the effect of femoral morphology on knee biomechanics in combination with validated biomechanical simulation models or experimental setups. New insights could directly be used for patient-specific implant design and optimisation.