Total joint replacement has excellent long-term outcomes but new designs are difficult to assess due to patient variability. The speaker discusses moving towards population-based computational modeling to generate hundreds of subject-specific models to better evaluate implant performance. Principal component analysis is used to build statistical shape models of femurs to automatically generate new instances for implant positioning and finite element analysis. This allows evaluation of variability across a population rather than a single "average" patient.

1. Moving Towards Population
Based Computational
Modelling of Total Joint
Replacement
Professor Mark Taylor

2. Total Joint Replacement
 Excellent survivorship at 10
years
 New designs regularly enter
the market
 Increasingly difficult to
assess whether design
changes will improve
performance

3. Sources of Variability
The Patient Surgery
•Experience
•Personal preference
•Age/activity level
•Alignment
•Bone quality/geometry
•Surgical approach
•Soft tissue quality
•Body weight

4. Femoral Head Resurfacing
Initial early-mid term clinical
results impressive
However:
 High incidence of femoral
neck fracture in first 6
months
 5 fold increase in revision
rate in small diameter heads
as compared to large
diameter heads1 http://www.orthoassociates.com
1Shimmin et al, JBJS(Br), 2010

5. FE analysis of the
resurfaced femoral head:
Modelling of an individual
patient

6. Subject specific models
3x BW
1x BW

7. Subject specific models
- Significant strain
shielding within the
head
- Increase in strain
on the superior
aspect of the neck
- Peak strain occurs
around the inferior
aspect of the neck

8. Comparison of a small vs. large femur
Small femur Large femur

9. Typical FE analysis of the resurfaced
femoral head
 Typically model the
“average” patient
 Ideal implantation, single
size
 Parametric studies on limited
number of variables
 Attempt to extrapolate results
to larger patient population
 Patient variability swamps
differences?

10. Typical FE analysis of the resurfaced
femoral head
 Typically model the
“average” patient
 Ideal implantation, single
size will not predict small percentage of failures
This
 Parametric studies on limited
Radical re-think of pre-clinical testing needed!
number of variables
 Attempt to extrapolate results
to larger patient population
 Patient variability swamps
differences?

11. FE analysis of the
resurfaced femoral head:
Modelling of 10’s of
patients

12. The brute force approach
- Model multiple femurs
from a range of patients
- Examine mean, standard
xN deviation, range….
- Perform statistical tests
when comparing designs
Radcliffe et al, Clin. Biomech., 2007

13. The brute force approach
Patient Data
200 45
180 40
160
Height (cm) / Weight
35
140
30
120
25
(kg)
BMI
100
20
80
15
60
40 10
20 5
0 0
Hip 609 Hip 613 Hip 628 Hip 631 Hip 636 Hip 608 Hip 626 Hip 607 Hip 625 Hip 612 Hip 610 Hip 630 Hip 614 Hip 635 Hip 627 Hip 634
Hip Number
Height (cm) Weight (kg) BMI
Weight: 95.312 kg (54 – 136)
Height: 1.76 m (1.57 – 1.88)
Age: 40.75 years (18 – 57)
Gender: male dominated

14. Influence of cementing the stem
N=16
Radcliffe et al, PhD Thesis, 2007

15. Influence of cementing the stem
N=16
Radcliffe et al, PhD Thesis, 2007

16. Influence of implant position
N=16
Radcliffe et al, PhD Thesis, 2007

17. The brute force approach
N=16
- Very labour intensive
-Impractical to examine 100’s of
femurs
- Still difficult to compare differences
across sizes
Radcliffe et al, PhD Thesis, 2007

18. FE analysis of the
resurfaced femoral head:
Modelling of 100’s of
patients

19. Principal Component Analysis
Construction of a Statistical Model

20. Statistical Shape and Intensity Model (n=46)
Mode 1 – Scaling of
morphology and properties
Mode 2 – Scaling and neck
anteversion
Model 3 – Neck anteversion
and head/neck ratio
Bryan et al, Med. Eng.
Phys., 2010

21. Generation of New Instances
• Using governing PCA equation it is possible to generate new,
realistic femur models from the variations captured by the
model

22. Automated Implantation
Automated Implantation – Run through Matlab
Hypermesh (Booleans) -> Ansys ICEM (meshing)-> Marc MSC
(FE)
Fully scripted from statistical model to FE results

23. Representative examples from N=400
Modulus
Modulus
Strain

24. Results (N=400)
Bryan et al, J. Biomech., 2012

25. Results (N=400)
Bryan et al, J. Biomech., 2012

26. Results - Comparison between head sizes
N=20
N=25
Small diameter heads show: Bryan et al, J. Biomech., 2012
- Increased strain shielding
- Elevated strains at the superior femoral neck

27. Statistical Shape and Intensity Model
• Developed methodology has
significant potential for improving
preclinical assessment
• There are issues:
• Statistical shape and intensity
models only as good as the
training set
• Robust automation
• Forces may need to link with
musculoskeletal models
• Verification/validation

28. Future directions…….
Drive for ‘real time’ tools
Femoral neck fracture
Diaphyseal fracture reduction
(KAIST, Korea) Implant Positioning (Imperial College, UK) (Brainlab, Germany)

29. Rapid patient specific modelling………
Surrogate model
FR = axb + cyd +……
100’s to 1000s
of simulations

30. FE simulation Surrogate model
Approx. 300 secs Approx. 0.2 secs

31. Acknowledgements
Dr Rebecca Bryan
Dr Ian Radcliffe
Dr Mike Strickland
Dr Francis Galloway
Dr Martin Browne
Dr Prasanth Nair