Download Quantitative Morphometry on Spinal X-rays

Quantitative Morphometry on Spinal X-Rays: Initial Evaluation of a New Workflow
Tool for Measuring Vertebral Body Height in Fractured Vertebrae
1,2
K ,
Engelke
1Synarc
Stampa
1
B,
Steiger
3
P,
Fuerst
1
T,
Genant
2Inst.
Inc, Hamburg, Germany and San Francisco, CA
of Medical Physics, Univ. of Erlangen, Germany, Optasia Medical Ltd, Cheadle, UK,
4Dept of Radiology, Univ. of California, San Francisco, CA
3
INTRODUCTION
METHODS continued
Vertebral height assessed by 6-point quantitative morphometry (QM)
provides useful information for the diagnosis of both prevalent and
incident vertebral fractures but reliable QM requires specially trained
technicians and is tedious, making it impractical in clinical routine.
The aim of this study was to evaluate a new QM tool SpineAnalyzer,
(Optasia Medical Ltd, Cheadle, UK) in a research setting for the
measurement of vertebral body height in fractured and non-fractured
vertebrae.
• Anterior, middle and posterior heights were calculated from the 6
points using standard algorithms.
• The coefficient of variation was computed between the manual
height measurement and those obtained by SpineAnalyzer default
results. Results were summarized by vertebra by height.
Inferior Posterior
Rim Center
Morphometry Points
• In this study, the default placement provided by SpineAnalyzer
were compared with a standard manual 6-point placement, which
was used as a reference standard. Default placements were not
corrected by the operator.
• In a second step an experienced QM reader corrected the default
SpineAnalyzer placements, if necessary. Results were again
compared with a standard manual 6-point placement.
• Lateral lumbar and thoracic x-rays from three groups of
postmenopausal women (73.7
5 y) with femoral neck T-score
< -2.5 and were used in three patient groups:
• 100 subjects without vertebral fractures (Group 1)
• 50 subjects with milder fractures (deformed 20-30%) (Group 2)
• 50 subjects with more severe fractures (>30%) (Group 3)
Anterior
Middle
Posterior
Anterior
9%
Middle
Posterior
6.0
8%
5.0
7%
6%
rmsSD mm
4.0
5%
Figure 5. Two subjects from Group 3. While default placement for normal and
mildly deformed vertebrae is acceptable, manual correction for more severely
deformed vertebrae is needed.
3.0
4%
3%
2.0
2%
1.0
1%
0%
0.0
T4
T4 T5 T6 T7 T8 T9 T10 T11 T12 L1 L2 L3 L4
Vertebra
T5
T6
T7
T8
T9 T10 T11 T12 L1
L2
L3
L4
Vertebra
Figure 2: Group 1 Percentage (rmsCV%) and absolute (rmsSD) deviation between
manual QM and SpineAnalyzer vertebral body heights. Dashed lines: automatic
analysis, solid lines: including manual corrections. The rmsCV% was largest in the
upper spine while rmsSD showed a slight trend toward larger values in the lumbar
spine. Middle heights agreed better with manual QM than anterior or posterior. Overall
mean rmsSD: auto analysis: 1.13 mm, corrected analysis: 1.02 mm.
Anterior
Middle
Posterior
Anterior
6.0
6.0
5.0
5.0
4.0
4.0
Middle
Posterior
rms SD mm
Right Posterior
Margin
Superior
Osteophyte
Handle
Figure 1. Vertebral body contours
identified by SpineAnalyzer are
shown. Triple contours are used for
the endplates, double contours for
the posterior margin and a single
Left Superior
Endplate Rim
contour for the anterior margin. QM
Anterior
Margin standard 6 points are shown along
Inferior Anterior
Right Inferior
the endplates. Operators are able to
Rim Center
Endplate Rim
adjust the size and rotation of the
contour using control points. IndividInferior Osteophyte
Handle
ual contours themselves can be
Left Inferior
Endplate Rim
adjusted to match the radiographic
Inferior
Cortical Endplate
depiction of the vertebra.
Superior Anterior
Rim Center
Right Superior
Superior Cortical Endplate Rim
Endplate
Superior Posterior
Rim Center
In all 200 subjects’ results were obtained from T4 to L4. In Group 1
eight (8) vertebrae were judged unreadable due to low image quality.
Results in Fig. 2 are based on the remaining 1,292 vertebrae. Fig. 3
shows the results for 61 fractured vertebrae of Group 2 and Fig. 4
the results for 143 fractured vertebrae of Group 3. Results for the
unfractured vertebrae of Group 2 and Group 3 were comparable to
those for Group 1 shown in Fig. 3
rmsCV (%)
Using SpineAnalyzer on lateral spine x-rays in the standard clinical
routine, the reader initiates analysis by placing a point in the
approximate center of each vertebra between T4 and L4. Based on
active shape and appearance models, SpineAnalyzer suggests
default placements of 6 points needed to measure posterior, mid and
anterior heights of the vertebrae. The reader would then make
adjustments, if necessary, allowing the algorithm to re-fit the
contours or manually placing the contours. The computer then
determines the placement of the QM standard 6 points on the
contours allowing calculation of the anterior, middle and posterior
heights.
RESULTS
rms SD mm
METHODS
Left Posterior
Margin
4
HK1,
3.0
3.0
2.0
2.0
1.0
1.0
0.0
0.0
T4
T5
T6
T7
T8
T9 T10 T11 T12 L1
L2
L3
L4
Vertebra
Figure 3. Group 2 Absolute (rmsSD)
deviation between manual QM and
SpineAnalyzer vertebral body heights are
plotted by vertebra. Error rates were
elevated by uncorrected fractured vertebrae and most noticeable below T8.
Overall mean rmsSD: auto analysis: 1.83
mm, corrected analysis: 1.16 mm.
T4
T5
T6
T7
T8
T9 T10 T11 T12 L1
L2
L3
L4
Vertebra
Figure 4. Group 3 Absolute (rmsSD)
deviation between manual QM and
SpineAnalyzer vertebral body heights.
Again, error rates were elevated by
uncorrected fractured vertebrae. Posterior height measurements were less
affected. Overall mean rmsSD: auto
analysis: 2.52 mm, corrected analysis:
1.37 mm.
SUMMARY
Overall in the non-fractured but not in the fractured vertebrae the
performance of the automatic point placement algorithm compared
very well with manual QM analysis, even in this difficult osteoporotic
population. Manual adjustment of the default contours improved the
rmsSD only 10%. However, consistent with the intended use of the
tool, after manual adjustment of contours on fractured vertebrae, the
rmsCV improved substantially (~40% reduction). which has rmsCV
values about 3-4% among readers and 2-3% within readers.
Adjustment of the default placement was necessary in a minority of
vertebrae and, with experience, can be done efficiently. The use of
SpineAnalyzer also helps to standardize point placement among
operators in the clinic. However the algorithm success rate is lower
in vertebrae with more severe fractures, which are of greater clinical
interest but are more readily recognized without morphometric
measurements. We expect that the default point placement can be
improved by developing a more sophisticated placement model
tuned for the shape characteristics and radiographic appearance of
fractured vertebrae. However, even in its current implementation
SpineAnalyzer is useful to highlight mild to moderate deformities for
closer inspection by the clinician.