|
MRI Method of Analysis
Understanding the function of the foot requires knowledge of
how the bones within the foot move while the foot is in different positions.
This study seeks to create a methodology to gain this understanding which is
quantitative, objective, accurate, and relatively fast. The MRI method of
analysis is a means for quantifying the in vivo motion of the bones of the foot
while held in various positions. This method represents an improvement on
previous studies which involved more invasive means and/or less comprehensive
characterization due to technological limitations. In this method scanning
positions were determined using each subject's end range of motion and neutral
position making the used positions slightly different for each subject. Subjects
were MRI scanned at their calculated positions in a specialized foot loading
frame. This method allowed the visualization of bone movements in three
dimensions through the use of Finite Helical Axes (FHA) allowing quantitative
results.
|
A. Foot Plate
B. Polhemus Sensor 1
C. Polhemus Transmitter
D.
Distal Leg Hold
E. Proximal Leg Hold
|
 |
| Figure 1 Setup |
Five subjects (mean of 53.4 ,std of 4.4 years) were enrolled
in this IRB (Human Subject Division, University of Washington) approved study.
Subjects were included if they had a neutrally aligned foot (classified by an
orthopedic surgeon) and were free of any lower extremity pathology. Exclusion
criteria included the inability to self ambulate, current ulceration and partial
foot amputation.
Scanning positions were determined in such a way as to allow
for the varying foot types and foot sizes. The Polhemus Liberty electromagnetic
motion analysis system (Polhemus Liberty: Colchester, VT) was employed towards
this purpose.
While electromagnetic sensors can be tracked in six degrees of
freedom, only sensor orientation (not translation) was necessary. Direction
cosines were used and converted to the desired Cardan angle system (XY'Z'
sequence) for use in the modified Ankle Flexibility Tester (AFT)(Figure 1.)
during scanning. For lower leg rigidity the rear half of the AFT was used during
position acquisition allowing a free-floating foot plate secured to the
subject's foot to be used, giving unrestricted full range of motion (Figure 1).
The rear half was later re-attached for scanning. Validation of the
repeatability of this setup was done as reported in a previous study.
There were three measured positions: maximum plantar flexion,
inversion, and internal rotation (position 1); anatomical neutral (position 5);
and maximum dorsiflexion, eversion, and external rotation (position 8) used to
create eight positions (linearly interpolated). Two positions were generated
between position 5 and 8 while three positions were generated between positions
1 and 5. Once the positions were calculated the two extremes were tested while
the subject was still in the jig to check for AFT interference problems due to
devices' design. Position 1 was backed off for most subjects and the
interpolated positions were recalculated.
Once the positions were determined, the subject's foot was
scanned using an MRI (Phillips Intera gyroscan 1.5 tesola, slice thickness 1.4,
repetition time 5.87, echo time 1.83, flip angle 25 degrees) scanner. Scanning
the eight positions in the MRI took about 1.5 hours per subject. Scanning took a
long time because the smallest possible volume was scanned to give the greatest
quality. This required more MRI technician time.
  |
|
| Figure 2 MRI scanners |
Feet were held in static positions during each scan
using the AFT at the positions described by three Cardan angles; these
represented the three anatomical planes of movement possible within the foot and
ankle. To give the full range of motion necessary for this study and to fit the
entire device through the coil of the MRI scanner, the AFT had to be modified to
give a larger plantar flexion motion as well as angular readouts.
The major medical image software steps/abilities are:
segmentation, registration, visualization, and analysis. The software
(Multi-Rigid) used for this project was created by Yangui Hu to register MRI
scan data to CT data. His software was initially capable of the more difficult
needs of this study; segmention (Figure 2) and registration. Visualization and
analysis features were added later. Segmenation was accomplished by drawing
initial seed lines (See colored points in Figure 3) which the software used as a
starting point for segmentation. The software segmented the bones from these
initial points using a technique involving a blend of graph cuts and level sets.
During processing the segmented volume expands until it comes to a bone's edge
(recognized as a change in image density). This was done simultaneously for all
14 bones (tibia, fibula, calcaneus, talus, navicular, cuboid, three cuneiforms
and five metatarsals). A secondary segmentation was done which refined the
segmentation giving smoother and slightly more accurate results.
 |
| Figure 3 Seed Points |
Registration was the most computatively intensive process as
it involved finding the spatial correspondence of the segmented bone information
within the other non-segmented non-neutral position scans. This step involved
using mutual information to find the transformation matrix for each bone to get
to each non-neutral position. These sets of transformation matrices described
the non-neutral positions relative to the segmented neutral position. To begin
registration three points were selected within three posterior bones (talus,
calcaneus and tibia) in the neutral position and each of the other seven
positions. From these initial sets of points the seven positions were registered
to the neutral position. To validate the registration step the seven
non-neutral positions from one set of MRI scans (NA04R) were segmented. The
segmented bones from these positions were checked against the same bones moved
to the seven positions from the segmented neutral position using the
transformation matrix determined from the registration. To compare how strongly
the bone orientations agreed between these two independent locating means the
bone volume overlap was calculated. The equation used was:  where  is
the volume of each bone which both methods share.
The results were visualized using custom software written
using the Visualization ToolKit (VTK). Surfaces were calculated from the
segmented data and visualized along with motion obtained from the transformation
matrices obtained during registration. The viewing application had the ability
to rotate the bones, show bone motion with interpolated intermediate positions
(to make the motion smoother), to turn bones on and off, to change to which
motion was shown to be relatice, and to turn on and off calculated FHA for any
combination of two bones. Visualization was primarily a tool to check the
results of segmentation, registration and the calculated FHA. Important results
were drawn from visualization which would be difficult, or impossible, to gather
by only looking at numerical descriptors.
 |
| Figure 4 Segmented Foot |
FHA (screw axes) was the implemented descriptors of motion used
for this study. An FHA describes the movement of a body from one position to
another by a translation along and rotation about an axis. FHA's were considered
the best means to quantify the movement of the bones by reducing the large
amount of information into the smallest possible number of descriptors. Other
possible measurements would have been to use cardan angles or Grood and Suntay
parameters. FHA's were chosen because: 1) they do not have the singularity know
as 'gimbal lock' which can occur with cardan angles, 2) in terms of joint axes,
FHA's were the most clinically relevant to joint motion and 3) they can be
implemented without the need to manually select the anatomical landmarks
required for Grood and Suntay parameters.
|
| Figure 5 Finite Helical Axes |
The FHA's were obtained using the principal axes tips of each bone
at neutral and the transformation matrix to the other positions. The numerical
outputs from FHA's were: 1) direction, 2) rotation about, 3) translation along,
and 4) a location point on each axis. The first three were presented.
Data processing was carried out on a DELL Precision 470
computer with a dual 3.2 GHz processors and 2GB of ram. Preparation and
segmentation took about 45 minutes including computer time. Registration took
5.25 hours being primarily computer time. A vast improvement over previous means
which took about 10 hours for a single position. Representative FHA's are shown
in Figure 3 for a left foot from a single subject. These models appeared
different for each subject, however, the general distribution and orientation of
the axes was similar. Differences between subjects can be gathered from the
standard deviations of the FHA descriptors (Table 1).
|
|
|
Axis Orientation
|
Rotation about
|
Translation along
|
|
|
|
Dorsiflexion
|
Inversion
|
Internal Rotation
|
Calc to Talus
|
1
|
2
|
41.42
|
±
|
6.74
|
79.81
|
±
|
9.35
|
9.69
|
±
|
8.99
|
3.52
|
±
|
0.09
|
0.32
|
±
|
0.20
|
Calc to Talus
|
2
|
3
|
41.04
|
±
|
2.33
|
78.12
|
±
|
6.82
|
11.10
|
±
|
6.33
|
6.18
|
±
|
2.11
|
0.19
|
±
|
0.11
|
Calc to Talus
|
3
|
4
|
45.75
|
±
|
4.25
|
72.06
|
±
|
2.68
|
18.67
|
±
|
4.62
|
5.64
|
±
|
1.98
|
0.18
|
±
|
0.08
|
Calc to Talus
|
4
|
5
|
45.47
|
±
|
6.30
|
61.79
|
±
|
2.15
|
27.19
|
±
|
9.25
|
4.20
|
±
|
0.99
|
0.15
|
±
|
0.09
|
Calc to Talus
|
5
|
6
|
34.22
|
±
|
10.35
|
64.48
|
±
|
8.71
|
18.87
|
±
|
8.37
|
4.08
|
±
|
1.94
|
0.57
|
±
|
0.32
|
Calc to Talus
|
6
|
7
|
30.66
|
±
|
12.81
|
51.61
|
±
|
7.87
|
24.63
|
±
|
6.94
|
2.03
|
±
|
0.55
|
0.41
|
±
|
0.21
|
Calc to Talus
|
7
|
8
|
31.06
|
±
|
9.81
|
55.61
|
±
|
18.18
|
19.39
|
±
|
13.93
|
1.53
|
±
|
1.17
|
0.07
|
±
|
0.04
|
Calc to Talus
|
1
|
8
|
41.39
|
±
|
6.61
|
69.88
|
±
|
5.23
|
18.20
|
±
|
6.04
|
27.43
|
±
|
6.50
|
1.17
|
±
|
0.19
|
Table 1 Summarized data for specified joins.
Bold values are after removing one outlier which was only done if the standard
deviation was reduced by ½ or more.
Research Team
Michael J. Fassbind, M.S.
Bruce Sangeorzan, M.D.
William Ledoux, Ph.D.
Eric Rohr, M.S.
Yangui "Patrick" Hu,
Ph.D
David Haynor, M.D
This Site is for patients and researchers interested in prosthetic and amputation research. Veterans can find VA health care information at the VA home page.
|
