Clinical biomechanics
Gait analysis and musculoskeletal modelling
I am interested in the development of patient-specific musculoskeletal models of the human body to better understand and treat complex deformities. These models are mainly designed to address clinically relevant questions, but also to support the development of orthoses or prostheses.
The creation of personalised models requires extensive use and processing of medical imaging. I devised innovative processing pipelines to derive models and capture movement from magnetic resonance imaging (MRI). For example, I used MRI images to determine the kinematics of the knee joint in vivo, build finite element models to study the contact between the cartilage and the menisci, provides thickness maps of the cartilage, and study the material properties of the anterior cruciate ligament in vivo.
I created personalised models of the human lower limbs during gait. Typical medical imaging modalities (e.g. CT-scans, MRI) are collected supine instead of standing. There are large soft tissue artefacts between the supine and standing position which led me to design new medical imaging modalities I developed freehand 3D ultrasound, which merges traditional planar ultrasound with a motion capture system. This allows to reconstruct 3D images using the 3D positioning data of the ultrasound probe. We developed and validated the use of freehand 3D ultrasound to locate the hip joint centre, to locate the condylar axis of the femur, and to measure the torsion of the femur bone. We have also used freehand 3D ultrasound to measure a range of muscle properties. I have also further developed bi-planar x-ray (EOS) to merge imaging with gait data. We used the EOS system to evaluate the accuracy of models to determine the hip joint centre and the condylar axis in volunteering adults and in children with torsional deformities. EOS and freehand 3D ultrasound measurements were validated against a method I devised to measure clinical parameters of the femur bone shape using low-dose CT-scans.
As an example of clinical application, the personalised models were utilised to determine the effect of torsional deformities on hip and knee joint contact forces during gait. The outputs informed surgical decision making to derotate the femurs.
I am one of the leader of the new conventional gait model, which aims to update the biomechanical model used in clinical practice around the world, and allow full traceability of the effect the different versions have on the model outputs.
Statistical learning in clinical research
My clinical role led to research interests in orthopaedics, in statistics, and in machine learning.
Gait analysis informs surgical planning in children with musculoskeletal deformities. Many require multiple surgical procedures involving several muscles and bones. We investigated the short and long-term effect of various surgical procedures from monocentric and multicentric cohort studies. I developed new algorithms to perform computationally tasks previously based on the visual assessment of experts. Algorithms are more easily shared than experts, therefore these algorithms allow improved harmonisation of the analysis across centres.
It is one thing to estimate the effect of deformities, or surgical procedures, but another, maybe more interesting, to provide an explanation for the observed effect. We investigated if radiographic measurements of the proximal femur and acetabulum could predict problems during gait in children post slipped capital femoral epiphysis. We also used correspondence analysis to investigate the statistical relationship between gait patterns and physical examination data.
I have developed a line of research in clinical statistical learning. It is customary to assess the gait of the patients before and after a surgical operation. This provide the opportunity to create statistical models to link the patients' gait to the surgical decisions, and to link the change in the patient’s gait to the surgical procedure. We developed statistical models 1- to capture clinical reasoning and 2- to link gait patterns with the outcomes post-surgery. The algorithms were implemented in a web-based computer-aided decision support system. Currently, I am leading a SPHN project to perform multicentric trial emulation from retrospective gait data.
