Morphing patient-specific musculoskeletal models

Anatomically realistic musculoskeletal models tend to be very complicated. The current full-body model of the AnyScript Model Repository comprises more than 1000 individually activated muscles and hundreds of bones and joints, and the development of these generic body parts represents an investment of dozens of man-years.
Healthy humans differ significantly in size, shape and general morphology, and this variation is even larger for patients with pathological anatomies. Thus, patient-specific models are imperative for reliable biomechanical analyses on which decisions of surgical treatments or development of medical devices can be based. Given the models’ complexity, a workflow that automatically creates patient-specific models bottom-up for each patient is unrealistic. The only solution seems to be development of methods that can transform generic models into patient-specific models, even when the generic model differs significantly from the patient in question.
The scenario therefore entails two sets of data: (i) a generic musculoskeletal model representing a single (average) individual, and (ii) a set of 3-D medical imaging data, typically in the form of a DICOM file obtained from CT, MRI or surface scans. Furthermore, we assume that a set of corresponding anatomical landmarks can be identified in the medical imaging data and on the generic musculoskeletal model. A nonlinear transformation, i.e. a morphing, is created by means of radial basis functions that maps points set (i) to point set (ii). The morphing is subsequently used to transform parts of the generic musculoskeletal model to a patient-specific version, thus changing bone shapes, muscle insertion points, joint locations and other geometrical properties.
Research questions include how to select point sets and whether the resulting models do indeed represent the patients’ biomechanics. As a particularly challenging case, foot deformities based only on point sets recovered from surface scans are considered as shown in the figure. The preliminary results are promising for the cases of severe flat foot and metatarsalgia while other conditions may require CT or MRI data.
The method and its theoretical assumptions, advantages and limitations are presented, and several examples will illustrate morphing to patient-specific models.
[1] Carbes S; Tørholm S; Rasmussen, J. A Detailed Twenty-six Segments Kinematic Foot model for Biomechanical Simulation. Transactions of the 57th annual meeting of the Orthopaedic Research Society, 2011, Long Beach, California.