Data fusion for modelling in paediatric oncology
F. Taddei
1
, S. Martelli
1
, G. Valente
1
, A. Leardini
2
, M. G. Benedetti
2
, M. Manfrini
3
, M. Viceconti
1
1
Istituto Ortopedico Rizzoli, Medical Technology Laboratory, Bologna, Italy
2
Istituto Ortopedico Rizzoli, Movement Analysis Laboratory, Bologna, Italy
3
Istituto Ortopedico Rizzoli, 5th Ward of Oncologic Orthopaedic-Trauma Surgery, Bologna, Italy
Introduction
Limb-salvage surgery is increasingly adopted in the
treatment of primary bone tumours, affecting principally
growing subjects at the long bones of the lower limbs [1].
Biological reconstructions, using intercalary massive allograft
to recover the excised bone, present some acknowledged
advantages especially when the patient’s joint can be spared.
Those reconstructions present a high long term survival rates
(75-89% at 10 years) but still allograft fractures are one of the
major complications. To limit the risk of fracture,
rehabilitation therapy is managed preventing a complete
weight bearing for long periods, which is considered
discouraging for the adoption of these surgical techniques. It
would be important to deeply understand the biomechanics of
reconstructed limbs either to improve the surgical technique,
with respect to the mechanical failure of the implants, and to
verify if shorter rehabilitation protocols may be adopted,
without increasing the fracture risk.
The first step for the evaluation of the fracture risk of the
reconstructed bone segments is the evaluation of the loads
acting on the skeleton during daily motor tasks. Direct and
non-invasive measurements in-vivo of muscle forces is
impossible, hence it is necessary to estimate them using
computational models of the musculoskeletal system. Most
recent studies have demonstrated the need for relying on a
subject-specific musculoskeletal models especially when
abnormalities of the skeleton geometry, and/or on the
muscular system, are present.
The development of such personalised musculoskeletal
models requires the fusion of data coming from different
sources (e.g. medical imaging data, gait analysis data, etc.).
Aim of the present study is to propose a method for
developing subject-specific musculoskeletal models of the
lower limbs and apply them on a real clinical case to predict
the joint and muscles’ forces during normal walking in a
young male subject with a massive skeletal reconstruction.
Materials and Methods
The selected patient was operated at the age of 10 for an
Osteosarcoma at the distal part of his left femur. An
intercalary reconstruction was performed by means of a
massive bone allograft in conjunction with a vascularized
fibula autograft. Fixation was provided by a titanium plate
with screws.
Fig. 1On the left: the CT dataset with the visible skin markers
used for the gait analysis. On the right the musculoskeletal
model of the patient lower limb
At month 42 of follow-up the patient underwent a
Computed Tomography (CT) exam at the lower limbs for
clinical reasons. In that occasion, a gait analysis session was
performed, and a specific protocol was adopted to allow the
registration of the patient’s skeleton with the kinematics
recorded during the gait session. Thirty-four reflective
markers were applied to the patient’s skin prior to CT
scanning, so their positions relative to the patient skeleton
could be estimated (Fig. 1). A subject-specific
musculoskeletal model was build, registered with the recorded
kinematics, and used to understand the biomechanics of the
reconstructed bone as explained below.
The musculoskeletal model.
The skeletal anatomy was segmented from the CT dataset
(Amira® v. 4.1, Mercury Computer System, Inc., USA). The
biomechanical model of the musculoskeletal system of the
lower-limb was defined as a 7-segment (pelvis, femurs, tibiae,
foot), 10 degree-of-freedom (DOF) articulated system
actuated by 82 Hill-type muscle-tendon units (Fig. 1). Each
leg was articulated by three ideal joints: a ball and socket at
the hip (3 DOF) and a hinge (1 DOF) at both the knee and the
ankle [2]. The identification of joint parameters was based on
relevant skeletal landmarks identified on the skeletal surface
with the virtual palpation procedure [3] (LHPBuilder©,SCS,
Italy). All anatomical landmarks defined by ISB standards
were identified and the local coordinate system was computed
for each segment. A generic muscular model of the lower
extremity including 82 muscles [4] was manually registered
on the skeletal anatomy. The muscular system of the operated
leg accounted for modification induced by surgery. Each
muscle was modelled as a Hill-type actuator for which the
muscle tetanic force, the tendon slack length and the optimal
muscle length were personalised fitting individual data. All
the other necessary parameters were taken form literature.
Inertial parameters of each segment were derived from CT
data assuming homogeneous density properties for both the
hard and the soft tissues.
Gait simulations were run using the OpenSim software
[4]. A least-square algorithm was used in a preliminary
inverse kinematics analysis to predict the joints kinematics
from marker trajectories. Musculoskeletal dynamics
simulations were then run by input these kinematics and the
recordings of the ground reaction forces. All five gait
repetitions were analysed.
Model validation.
The models were validated comparing predicted
activations of some of the major muscles of the operated leg
with EMG recordings.
Results
All gaits were successfully analysed.
Validation
Consistent muscle activation patterns were predicted
throughout the five simulated repetitions. Comparing the
predicted muscle excitations with the EMG recordings
(available for a single trial), a general agreement was found