Although osteoporosis
is diagnosed using bone mineral density (BMD), a measure of the quantity
of bone, It is not the only risk factor for hip fracture. Many factors
play a role, including the shape of the proximal femur (the upper part
of the thigh bone).
If we compare the hip to a bridge or building, BMD relates to the concrete
or brick it is built from, whilst the shape of the femur relates to
its design. It is clear that both of these factors are important in
determining its strength.
Fig 1
Template for active shape model of the femur
The shape
of the femur was modelled using a statistical model (an Active
Shape Model [1]) to examine the shape of
femurs in standard pelvic radiographs (X-rays). Fifty radiographs
were available from postmenopausal women. Of these, 26 had suffered
a fracture whilst the other 24 were used as a control group.
The aim of this study was to investigate whether we could identify
the fracture group from the shape of the femur alone. (When
a fracture had occurred, the unfractured femur was analyzed).
The active shape model allows observation and measurement
of differences in the whole shape of the femur across a large group
of images (a different approach to either engineering models or
geometrical measurements). An advantage over geometrical measurements
such as the neck length and neck width is the ability to measure
changes in the shape independently from changes in the size of the
hip.
The shape is described using a number of landmark
points that are placed around the outline of the femur (Fig. 1).
The co-ordinates of these points can then be analyzed to build a
model of shape variation in the group. The analysis finds a number
of 'modes of variation' - patterns of shape variation. Each image
is given a score for each mode of variation, indicating how its
shape compares to the others in the group. If the fracture and control
groups have different shaped femurs, these scores can be used to
identify them.
Figure 2 shows the second mode of variation. This mode was one
of those that showed a significant difference between the fracture
and control groups (P = 0.019). The animation shows the
shape represented by +2 to -2 standard deviations in the mode2
score. The red outline represents the shape found at -2 standard
deviations. This is more likely to fracture than the blue outline
(+2 standard deviations). The green outline shows the average
shape.
The red outline has a relatively longer thinner femoral neck,
with a larger neck shaft angle than the green and blue outlines.
These features have been found to relate to hip fracture individually,
however by using the active shape model, it is possible to automatically
identify when they occur together, and also to identify other
shapes associated with fracture risk.
Figure 2
Mode of variation (#2)
Stepwise discriminant analysis selected modes
2, 4, 5 and 7 to build a classifier to distinguish between the control
and fracture subjects. The result of this was saved as a new variable
Pshape.
Pshape was not significantly correlated
with BMD, age or body mass index (P > 0.05).
Table 1 shows the results of the study. The results
are given in percentage accuracy (0 - 100%) and by the area under
the receiver operating characteristic (ROC) curve which varies from
0.5 for a classifier that is no better than random to 1.0 for a
perfect result.
Table 1
Results
Variable
Percentage accuracy
Area under the ROC curve
Pshape
74
0.81
Femoral neck BMD
74
0.79
Pshape and femoral neck BMD
82
0.89
For more information on osteoporosis and hip fracture, please visit the
website of the National Osteoporosis Society www.nos.org.uk
References
[1] Gregory,J.S.; Testi,D.; Stewart,A.;
Undrill,P.E.; Reid,D.M.; Aspden,R.M., A method for assessment of the shape
of the proximal femur and its relationship to osteoporotic hip fracture.
Osteoporosis International 2004 15(1) 5-11 DOI:
10.1007/s00198-003-1451-y
Department of Orthopaedics, University of
Aberdeen
