Traditionally, musculo-skeletal (MS) models are based on an "average" model and simply scaled (using for example the height of the patient) to generate a so-called subject-specific model. However, this is a highly doubtful procedure as individual's M-S characteristics may differ considerably. Our goal is therefore to use state of the art 3-D image analysis techniques and develop a reliable and fast methods to parameterize 3-D images, in order to generate reliable subject-specific MS models in Work Package 3.


Muscle volumes

In order to build a subject-specific MS model, one important parameter is given by the volumes of the different muscles. These volumes can be 'seen' on the MRI images, but to compute them we should indicate on every slice which pixels belong to which muscles. This process is called segmentation.

Our MRI images typically contain about 350 image slices, and segmenting all the muscles is a time consuming task. Therefore, we are looking for algorithms that can help to automate and speed up the segmentation. One possible approach is to use the knowledge from one manually segmented subject to predict the segmentation of other subjects. In that way a reasonable approximation can be obtained, which can be improved afterwards by using more specific pixel information (Figure 1).

Figure 1: Muscle segmentation.

Muscle attachment sites

Muscles attachment sites are also an important parameter to build patient-specific M-S models. These sites are typically difficult, if not impossible, to delineate on MRI images. Our strategy is thus to experimentally delineate these regions on cadaver bones, and align these regions with 3D bone models obtained by segmentation and subsequent 3D reconstruction. By morphing these 3D bone models onto the bone models of the subject, subject-specific muscle attachment sites can then be obtained (Figure 2).

Figure 2: 3D femur model with muscle attachment sites delineated experimentally (left). Bone and muscle attachment sites after morphing to the subject-specific femur bone volume (right).

As a result, generic M-S models can be updated with subject-specific information extracted from MRI images, i.e. bone models, muscle volumes and muscle attachment regions (Figure 3)

Figure 3: Update of muscle attachment sites based on morphed generic bone using AnyBody Modeling System. Comparison of muscle attachment points using simple anthropometric scaling, based on subject's height and weight (left) and morphing available in AnyBody Modeling System (right).

Additional information