As mentioned previously, most FE modeling studies use literature motion to drive the FE models [195, 199, 204, 212, 213, 216, 230–233]. This is due to a lack of motion capture availability in clinical settings. Even though the conclusion derived from these studies would not change, literature-based motion may not be suitable for patients with traumatic knee joint injuries, such as ACL rupture, or surgeries, such as ACLR, as both conditions may lead to abnormal joint kinematics, which may alter the joint and tissue loading.
In FE models where the gait cycle is taken into account from motion capture systems, two different approaches are commonly used. Kinetic models use forces
and moments to simulate patient motion. However, this requires including the patella, patellar tendon and quadriceps tendons, with their contributions estimated using musculoskeletal modeling [74, 205, 207, 209, 240, 252]. Kinematic driven models use translations and rotations to simulate patient motion [209, 216]. Both these methods have two major drawbacks: (1) the joint translations cannot be measured directly from motion capture systems and have to be estimated for example using musculoskeletal modeling [205, 230], and (2) joint angles are susceptible to errors due to skin marker movement [235–237]. For large groups of subjects, the necessity of additional post-processing of gait data makes both approaches too complicated.
Verification
Direct validation of the stresses and strains predicted by the FE models is almost impossible inin-vivo. For the knee joint, this would require surgical implantation of pressure sensors into the cartilage-cartilage contact area, which is not feasible; thus verification methods must be used, of which there are a few options available.
First, the FE model results can be verified against literature data, to ensure they are within physiologically-relevant ranges [190, 209, 233, 240, 253]. Second, radiographic methods [39, 195] or semi-quantitative MRI methods [197, 231] can be used to ensure that the FE model predictions are in agreement with information from experimental follow-up. However, both methods are susceptible to inter- and intraobserver variability and misclassification [26, 29, 126, 153, 254]. Moreover, these methods are unable to offer sufficient information on soft-tissue integrity or composition, that would directly relate to the biomechanical parameters from the FE models.
Patient Number
One limitation of most FE models is the low number of subjects that have been evaluated [64, 67, 189, 190, 204, 205, 207–210, 216, 230, 233, 255–257]. This is due to several factors that drastically increase the model generation time and complexity.
First, the knee geometry is obtained from clinical images and typically manually segmented [188,209,216,230,233]. Second, many studies use complex materials, such as FRPVE, to describe the soft-tissues [39, 40, 258, 259]. This is simultaneously time-consuming in terms of implementations and computationally demanding. Third, when including all major knee joint structures, both generation and simulation times become longer [204, 205, 216].
Some studies have tried to decrease the model generation time by using semi-automatic or semi-automatic segmentation techniques to obtain the joint geometry [197, 260–262]. More recently, atlas based approaches for FE model generation have been proposed [195]. In this approach, knee joint anatomical dimensions are measured from MRI. Then a template model that best matches these dimensions is selected and scaled. In this template approach, all boundary and loading conditions as well interactions and material properties are already considered. With this approach full FE models can be generated in a few minutes. This method would be extremely useful in a clinical setting, since it is significantly faster than fully patient-specific approaches. However, the approach currently only uses literature gait, which may affect the accuracy of the FE models.
5 Aims and hypotheses
There are numerous studies investigating the onset and development of knee OA utilizing different imaging methods and gait analysis combined with computational modeling. However, there is still a lack of computational tools suitable for large cohort studies and clinical use, that can be used to predict joint locations susceptible to OA. Subject-specific computational models would be particularly useful in assessing the effectiveness of surgical interventions, such as ACLR, and evaluateing non-surgical management options for avoiding or delaying OA onset and progression. In a clinical setting, patient-specific computational models have to be easy to generate and the time for the converged solution needs to be short. Moreover, to ensure reliability, patient-specific FE model predictions should be verified in large prospective clinical trials.
The specific aims ofStudies I-IVwere to:
1. Evaluate different methods for implementing subject-specific motion in terms of simulated contact mechanics and mechanical response of articular cartilage, and to identify the method most suitable in large cohort studies and/or in a clinical setting.
2. Utilize the relatively fast FE modeling approach from Study I, to generate subject-specific FE models for a large number of ACLR and healthy subjects at the 1-year follow-up time-point. Another aim was to identify and verify locations and cartilage volumes at risk of developing OA. Excessive maximum principal stresses were assumed to lead to collagen network degeneration and excessive shear strains were assumed to cause PG loss.
3. Quantify the relationship between collagen specific parameters (maximum principal stress and changes in T2 relaxation time between 1- and 3-year follow-up time points) and PG specific parameters (absolute shear strain and changes inT1ρrelaxation time between 1- and 3-year follow-up time points).
4. Apply a previously developed adaptive degeneration algorithm, to investigate three possible mechanisms for degeneration via PG loss around focal cartilage lesions, namely maximum shear strains, absolute shear strains and fluid velocity. Qualitatively and quantitatively compare model predictions against local changes in T2 and T1ρ relaxation times between the 1-year and 3-year follow-up time-points.
We examined the following hypotheses:
1. FE models with simplified geometries and inputs can produce similar responses within the knee joint as the more complicated models, and potentially demonstrate the ability of simplified FE models to be used in studies with high number of subjects. (Study I)
2. Areas susceptible to OA in patients with ACLR can be identified by with FE models using maximum principal stresses and absolute shear strains. These
identified areas match local changes inT2andT1ρrelaxation times of articular cartilage, respectively. (Studies IIandIII)
3. Localized changes in FCD content around cartilage lesions correspond to local changes inT2andT1ρrelaxation times. (Study IV)
6 Materials and Methods
This thesis consists of four individualStudies (I-IV). The methods used in each study are summarized in this chapter. For more a detailed description of the materials and methods, please see the original publications in the appendix.
6.1 WORKFLOW
The workflows ofStudies I-IVare illustrated in Fig. 6.1. The knee joint geometry is based on manually segmented CUBE MRI (Fig. 6.1a). The knee joint motion is based on motion capture (Fig. 6.1b). ForStudies Iand IV, the quadriceps muscle force, obtained from musculoskeletal modeling, was also implemented (Fig. 6.1e).
InStudies I-III, the articular cartilage surfaces were modeled as TIPE material and menisci were modeled as TIE. In Study IV, articular cartilage was modeled as FRPVE material with swelling and menisci as FRPE. In all studies ACL, PCL, MCL, LCL were included and modeled as bi-linear springs (Fig. 6.1d). Additionally, in Studies IandIV, the medial and lateral patello-femoral ligaments (MPFL, LFPL), quadriceps tendon (QT) and patellar tendon (PT) were also included and modeled as linear springs between the insertion points (Fig. 6.1d). In terms of verification, the FE model results were compared against literature data (Study I), quantitative follow-up information (T2 and T1ρ relaxation times, Studies II-IV) or semi-quantitative follow-up information (WORMS scores,Study III).
Figure 6.1: Workflow of Studies I-IV: a) Knee joint geometry is obtained from manually segmented MR images; b) Knee joint motion is obtained using motion capture; c) FE models; d) Overview of geometry and motion inputs for the FE models; e) Musculoskeletal modeling used in Studies I and IV; f) Verification methods used in each study.
6.2 FE MODELING