close

Вход

Забыли?

вход по аккаунту

?

Joint space measurement in hand radiographs using computerized image analysis.

код для вставкиСкачать
ARTHRITIS & RHEUMATISM Volume 38
Number 7, July 1995, pp 891-901
6 1995, American College of Rheumatology
891
JOINT SPACE MEASUREMENT IN HAND RADIOGRAPHS USING
COMPUTERIZED IMAGE ANALYSIS
MICHAEL F. JAMES, GEOFF HEALD, JOHN H. SHORTER, and ROBERT A. TURNER
Objective. To compare computerizedjoint space
(JS) measurements with conventional joint space narrowing (JSN) scores in patients with mild rheumatoid
arthritis.
Methods. Serial paired hand and wrist radiographs from 34 patients with classic rheumatoid arthritis were evaluated. Purpose-written software automatically measured the JS on test images and standard
clinical hand radiographs;JSN was scored “blind” by 6
observers.
Results. The software proved reliable. JS values
differed significantly (men > women; metacarpophalangeal > proximal interphalangealjoints), declining with
disease duration more than with age; JSN scores correlated poorly and varied more.
Conclusion. Computerization permits sensitive
JS measurement and should be of benefit in studies of
early joint disease.
In the rheumatic diseases, assessments of disease progression, outcome, and the effects of treatment rely heavily on radiologic analysis: typically,
patients’ radiographs are scored for the degree or
amount of certain disease features (1-3). Accurate
scoring relies very much on the skill and experience of
the clinician, although it can be improved by comparing standard radiographs of diseased joints. Such
scoring systems are largely qualitative and subjective,
and much effort has been invested toward improving
their reliability and efficiency (4-8). Nevertheless,
Dr. Turner’s clinical work during this project was supported in part by grants from Beecham Laboratories, Bristol, TN
(now SmithKline Beecham Pharmaceuticals).
Michael F. James, PhD, Geoff Heald, John H. Shorter,
MIBiol: SmithKline Beecham Pharmaceuticals, Harlow, Essex,
UK; Robert A. Turner, MD: The Arthritis Center, West Palm
Beach, Florida.
Address reprint requests to Michael F. James, PhD, SmithKline Beecham Pharmaceuticals, New Frontiers Science Park
(North), Third Avenue, Harlow, Essex CM19 5AW, UK.
Submitted for publication May 18, 1993; accepted in revised
form January 24, 1995.
because of these limitations, large clinical trials of long
duration are still required to radiologically demonstrate clinical efficacy in rheumatoid arthritis (RA)
(9-12).
Significant damage to the soft tissues of the
joint (including articular cartilage) begins early in RA;
therefore, truly disease-modifying treatment should be
started soon after the diagnosis. Measuring radiographic progression is also important because clinical
indices may not reflect deterioration of the joint (1316), but scoring early RA is especially a problem
because the radiographic changes may be slight.
To overcome such difficulties, attempts have
been made to develop new, more sensitive radiologic
measurement techniques (17). Computerized methods
for detecting radiologic change (18-25) have several
potential advantages: they are objective, observerindependent, accurate, and reproducible; they use a
continuous measurement scale capable of calibration;
and they can be standardized for multicenter use.
We describe here the use of computerized image analysis for measuring the radiographic joint space
(JS) in the metacarpophalangeal (MCP) and proximal
interphalangeal (PIP) joints of 34 patients with RA.
The computerized measurements were compared with
standard joint space narrowing (JSN) scores for the
same joints.
PATIENTS AND METHODS
Patients. Serial, paired posteroanterior (PA) radiographs of the hands and wrists of 34 patients (16 women, 18
men) who met the American College of Rheumatology
(formerly, the American Rheumatism Association) criteria
for classic RA (26) were evaluated. For women, the mean
disease duration at the first (baseline) visit was 5.1 years
(range 1-17) at a mean age of 46.1 years (range 24-69); for
men, disease duration was 4.7 years (range 1-23) at a mean
age of 46.3 years (range 23-71). Patients received various
nonsteroidal antiinflammatory drugs: 24 took fenoprofen, 17
took aspirin, and 10 took naproxen. Some also received
JAMES ET AL
892
intramuscular gold (15 patients), penicillamine (2 patients),
or low-dose prednisone (1 patient).
For a separate study published elsewhere (8), JSN
was scored according to the method of Sharp et a1 (27).
Eighteen areas of each hand and wrist were scored by 6
independent observers (4 rheumatologists, 2 radiologists),
using a scale of 0-4 (0 = no clinical disease, 4 no visible joint
space). Between observers, scores compared well (mean
Pearson correlation coefficient = 0.82 [ P < 0.0011, range
0.61-0.91), so the mean scores for each patient were used.
The mean baseline JSN score for women was 12.0 (range
3.7-34.8; maximum possible score 144), and for men, it was
12.2 (range 1.7-30.3), indicating low disease activity.
Radiographs. Films (n = 127, divided into 56 sets)
had been taken annually or semiannually, and were independently coded in advance to prevent identification; 2 codes
were allotted to some patients to divide their radiographs
into annual intervals. Films were visually inspected before
measurement and were arbitrarily deemed unsatisfactory if
hands were not PA and positioned flat (many of these did
appear measureable, however). Sixteen sets were rejected,
including 7 sets where an acceptable film remained (serial
measurements were required for the study); unsatisfactory
films (n = 34) were anteroposterior (n = 6) or variably
supinated (n = 21), or had no joint space visible (n = 7). Film
codes were broken only after measurements were completed.
Image capture. Radiographs were viewed on a standard illuminated light box with a Hitachi KP41 video camera
( y setting = 1) through a Nikon AF Micro Nikkor 55-mm
lens. The displayed image (720 x 560 pixels x 256 gray
levels; monitor magnification -6.5, giving -21.4 pixels *
mm-’) was captured on a “Photon” framestore card (Analytical Measuring Systems Ltd., Cambridge, England) within
an IBM P Y 2 - 5 0 ~computer incorporating a maths coprocessor (Intel 80287-10) to accelerate processing. The radiograph
was positioned to display the JS vertically in the center of
the monitor screen (Figure lb). The lens aperture was
adjusted to maximize image brightness without producing
image saturation (“flare”).
Image analysis. Having captured the radiographic
video image, the operator defined the limits of the joint by
placing 3 cursor points close to the proximal JS margin.
The program then proceeded automatically to locate
the actual proximal JS margin by drawing a circular arc
through the cursor points and, from the arc, to find the bone
outline where the radiographic densities changed most rapidly from light to dark. The radiographic JS (naturally
circular to accommodate finger flexion) was then “straightened out” by the software: the image’s component lines,
running radial to the arc and across the JS, were rotated to
the horizontal, vertically aligned using the “found” points
on the proximal JS margin, and then stacked. Finally, the
radiographic densities within the stacked JS were averaged
to give a density profile across the entire width of the JS
(Figure lc).
Measurements. The steepest slopes along the density
profile were automatically located by the program to give 2
measurements (Figure Id): 1, between the proximal joint
margin and the leading edge of the distal joint margin,
corresponding to the JS perceived by the eye; and t, between
the proximal joint margin and the distal limit of the com-
(a) Lateral view of MCP joint
(b) Resulting plain radiograph
(horizontal PA view)
(c) Derived joint space profile
(d) Joint space measurements:
’I ’ =distal metacarpal margin to
proximal phalangeal margin.
‘t ’ =distal metacarpal margin to
the distal limit of the phalangeal
compacted subchondral bone.
Figure 1. Basis for the computerized joint space (JS) measurements. The 3-dimensional finger joint (a) forms a 2-dimensional plain
radiograph (posteroanterior [PA] view) (b), in which the radiographic densities through the joint’s thickness are integrated in the
resulting film. The software automatically found the JS margins,
straightened out the JS, and profiled its gray levels (c); its maximum
slopes were used to give at least 2 JS measurements (d): I , corresponding to the JS as perceived by the eye, embodies the JS
evaluated in clinical scoring procedures; and t includes the subchondral compacted bone. Since t varied between individuals (but was
very reproducible within individuals), 2 values of f could be measured (t2 < r2). Metacarpophalangeal (MCP) and proximal interphalangeal joints were measured in the same way.
pacted subchondral bone of the distal joint margin. (Because
I is derived from the radiographic landmarks used in assessing the joint space clinically, I, rather than t, is the principal
measurement addressed in this report; nevertheless, in recognition that more than one computerized radiographic
measurement is possible, r is also included.)
The location of 1 was unambiguous for all joints. For
the MCPs, the location of t could vary idiosyncratically
depending on the individual anatomy of the phalanx, although within individuals, it was very reproducible. For
some MCPs, 2 measurements o f t were possible: a movable
cursor permitted the observer to identify an additional point
on the JS profile, and the program calculated its distance
from the proximal joint margin. Objectivity was maintained
RADIOGRAPHIC JOINT SPACE MEASUREMENT
by selecting only those places where the slope value of the
profile (outputted to the monitor) was maximum. MCP t
values were designated tl or t2 according to size ( t l < t2).
Complete measurements took about 1.5 minutes.
Measurement checking. Measurements were made in
triplicate. To ensure internal consistency and to prevent drift
between the first and last measurements of the whole series,
the replicate measurements of each joint were checked at the
end of the measurement process. If necessary, individual
joints were remeasured to adjust the final JS value, but in
general, agreement was excellent.
Thus, for each radiograph, at least 3 replicate measurements of l and t were made on the MCPs and PIPs of the
index, middle, and ring fingers of both hands. PIPs were
measured laterally as well as medially, since this joint is
divided.
Measurement calibration. Image magnification was
calibrated in pixels * mm-’ and repeatedly checked during
the measurement process using the radiograph of an aluminum cone that had been machined in 1.0-rnm steps.
The calibration software required the operator to
place cursor points on the captured image of the calibrating
standard to define X and Y distances: once the operator had
input the reference units, the calibration values were displayed and incorporated within the measurement program.
No drift in the calibration was observed.
Measurement accuracy. To check for accuracy and
image distortion, standard graph paper, ruled in 1-mm
squares, was imaged through the camera system at the
magnification used for radiograph measurements (at the
standard magnification, the full screen “measured” -33.5 x
26.25 mm): squares of sides that were 5-25 mm from the
same image were repeatedly measured, and the deviation
from the mean calibration value was assessed by linear
regression analysis.
The accuracy of the radiographic JS measurement
software itself was assessed using 4 phantoms that were
constructed using computer graphics software (Adobe Illustrator 5; Adobe Systems, Mountain View, CA). Concentric
discs of diameters 12.00 mm (clear), 14.00-17.00 mm (black),
and 15.00-18.00 mm (clear) were graphically superimposed
on a 90% “screen” background to provide target shapes in
which a black annulus of line widths 1.00 mm, 1.50 mm, 2.00
mm, or 2.50 mm (representing the test radiographic JS
measurement, equivalent to r) was bordered by white regions
of fixed size (representing the bone ends). The test objects
were printed onto clear film using a lithoprinter (Linotronic
300; Linotype-Hell, Brentford, England) at a resolution of
100 dots * mm-’ and 4.72 lines mm-’. After affixing the
film to a glass microscope slide, the JS of the phantom (the
right half of the respective annulus) was then measured using
the joint analysis software, at the standard magnification
used for radiographs.
The phantom was imaged using a modified gray scale
in which the brightest level was colored red (instead of
white) on the monitor. Since small variations in gray (e.g.,
between the 5 brightest levels in a 256 gray-level image) are
visually indistinguishable, this procedure ensured that the
illumination level of the phantom was accurately controlled
and image distortion was prevented.
The accuracy of the phantoms was confirmed using a
893
standard calibration graticule ruled in 0.01-mm steps (Carl
Zeiss, Oberkochen, Germany). With the calibration software
described above, radial measurements were made horizontally around the annulus, encompassing points used for the
phantom JS measurement.
Statistical analysis. Data (expressed as the mean
SEM) were compared using RS/1 software (release 4.01)
(28). In general, comparisons were made using Student’s
t-test (grouped means) or Student’s paired t-test (within
individuals). Where the distribution of data was skewed,
robust methods were used to compare medians, and where
variances were unequal, a weighted analysis was performed
(29). Differences were accepted as statistically significant at
the 5% level (P < 0.05).
RESULTS
Checking image distortion. Measured images of
graph paper squares (sides = 5-25 mm) returned
horizontal and vertical measurements very close to
expected values. The mean (+SEM) deviation was
0.003 k 0.080% (X values) and 0.009 k 0.045% (Y
values) for 22 squares. Figure 2 shows that the actual
and observed measurements were highly correlated
(r > 0.99); linear regression analyses showed that the
slopes for the X and Y measurements were very close
to unity and significantly different from zero (P <
0.001); intercepts were essentially zero.
Assessing the accuracy of the joint space measurement. To test the accuracy of the joint space
measurement software, phantom annular joint spaces
were employed. The accuracies of the phantoms were
measured with a microscope slide calibration graticule
and found to deviate <0.75% from their nominal
values (Table 1); the variation in these replicate measurements was small (coefficients of variation [CV =
SD/mean %] were 0.16-0.44). However, the deviations
were accepted as real, and corrected for.
Measurements of the test annuli then used the
JS measurement software. Replicate measurements
were found to vary little or not at all and, overall,
deviated < I % from the corrected phantom JS (Table
1). Thus, within the limits imposed by this methodology, the JS measurement software had the accuracy
and precision expected of any measurement system.
Assessing radiographic measurement variation.
Having assured ourselves of the reliability of the
software, the variation inherent in clinical radiograph
measurements was then investigated. Eight replicate
JS measurements were made on 4 baseline patients (2
women and 2 men chosen at random, while ensuring
that all joints were measurable) to give a range of JS
measurements. All joints were measured once (PIPs
JAMES ET AL
894
A
A
PIPl(47)
PIP t (48)
K P I (24)
W P t (24)
0
Median & 95th percentile
0
0
-
2
E
E
v
4 IPlPt
0
t
0
51
MCP I
0X-values
-A
0
5
10
15
Y-values
20
25
0.0
Figure 2. Assessing image distortion. Standard graph paper was
imaged at the magnification used for radiographic measurement, and
squares of sides of 5-25 mm were repeatedly measured. The graph
shows that the observed measurements correlated highly with the
actual values (r > 0.99). The deviation from the mean calibration
value was 0.003
0.080% (X values) and 0.009 2 0.045% (Y
values); significant linear regressions (P< 0.001) fitted the X and Y
values with slope values of unity and intercepts close to zero (Y =
1.004X - 0.051 and Y = 0.999X + 0.01 1, respectively, mean 2 SEM
of 22 squares).
*
laterally and medially) before repeating the measurement cycle (representing at least 5 days' work, spread
over several weeks).
For each joint, the CVs of the replicates were
Table 1. Assessing joint sDace measurement accuracy*
Deviation of
Measured
Nominal
Calibrated
measured from
calibrated
test joint space test joint space test joint space
(mm)
(mean 2 SEM) (mean -t SEM)
(%)
1.OO
1S O
2.00
2.50
1.005 5 0.001
1.503 % 0.002
1.985 2 0.003
2.483 2 0.001
1.013
1.520
2.026
2.482 +- 0.006
I
t
Actual size (mm)
0.80
1.13
2.07
-0.04
* Test joint spaces were constructed graphically and calibrated (8
measurements) using a microscope graticule. Test joint space measurements using the software were found to vary little or not at all,
overall deviating <1% from the calibrated values.
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Mean JS measurement (mm)
(Individual joints)
Figure 3. Assessing the variability of the methodology. From baseline radiographs of 2 women and 2 men with rheumatoid arthritis, 8
replicate joint space (JS) measurements were made of the metacarpophalangeal (MCP) and the proximal interphalangeal (PIP)joints of
digits 2 4 of both hands, the PIPs being measured both laterally and
medially. The graph shows the replicates' coefficients of variation
(CoV = S D h e a n %) plotted against their mean value; 1 joint was
too indistinct to measure satisfactorily (PIP I , CoV = 30.5%). Some
replicates did not vary; thus, the median and 1-sided 95th percentile
for each joint type are shown. The pooled values for 1 and t gave
combined 95th percentiles of 8.3% and 2.8% for PIPs and MCPs,
respectively. (See Figure 1 for further details.)
plotted against their mean values (Figure 3); 1 joint
was too indistinct to measure satisfactorily (PIP I, CV
30.5%) and was thus omitted. Some replicates did not
vary; thus, the median and I-sided 95th percentile
values of the CV for each joint type are shown. After
pooling the 1 and t CVs, combined 95th percentile
values of 8.3% (PIPs) and 2.8% (MCPs) were obtained.
These low overall values indicate the excellent reproducibility of the system.
Baseline JS measurements within and between
patients. Further work then investigated detectable
differences in JS size in the baseline measurements of
all patients. We found that most of the joints in the
selected films were measureable (94%); indistinct or
895
RADIOGRAPHIC JOINT SPACE MEASUREMENT
nonparallel joint margins affected PIP t values most
(6% of joints) and MCP 1 and t values least (1.4%).
Table 2 shows that the baseline JS measurements were significantly larger in men than women.
The sums of the visual JS values for each patient (i.e.,
1 for the 3 MCPs and 3 PIPS measured in both hands,
denoted “hand” here) were also significantly larger in
men than women. The MCP JS of any finger was also
significantly greater than its corresponding PIP JS.
There was no evidence for consistent differences in
joint size between hands (patients’ handedness was
not known). Since there were also no significant
differences between the lateral and medial PIP JS
measurements, these were averaged.
Baseline JS measurements and age and disease
duration. The baseline JS 1 measurements were found
to correlate inversely with both age and disease duration (Figures 4 and 5). Since the variation in JS
measurement between patients was very large, these
correlations were weak (r 5 0.43),even where the linear
regressions were significant (i.e., those shown). Since
most patients had disease of <I0 years’ duration, the
duration data were heavily influenced by 3 patients (2
women and 1 man, RA duration 13, 17, and 23 years).
However, the numerical decline in JS measurementwith
disease duration (-0.02 mm year-’) most closely reflected the decline in the paired measurement (-0.04
mm year-’), suggesting that disease, rather than age
(-0.005 mm * year-‘), principally influences the JS reduction observed here. There was no correlation of
disease duration with age (data not shown).
-
-
Table 2. Computerized joint space measurements in 34 patients
with rheumatoid arthritis*
Joint
Women
(n = 16)
Men
(n = 18)
1.33 t 0.03 (94)
1.78 zk 0.05 (62)
2.68 t 0.04 (40)
1.60 2 0.03 (107)
2.04 2 0.05 (44)
3.08 -C 0.03 (91)
0.81 t 0.02 (94)
1.33 t 0.04 (94)
12.59 t 0.61 (16)
1.00 2 0.02 (105)
1.67 2 0.05 (104)
15.25 2 0.50 (18)
MCP
1
tl
t2
PIP
1
t
Hand I
* Computerized measurements (I, tl, and t2 in millimeters) were of
digits 2-3 of both hands, using purpose-written software; measurements were combined to give a total value (“Hand”). MCP =
metacarpophalangeal; PIP = proximal interphalangeal; I = joint
space as perceived by the eye; t = joint space including the
subchondral compacted bone. The choice o f t (designated t l or r2,
by size) varied between individuals, but was highly reproducible
within individuals. All joint space measurements were significantly
different in women compared with men (P < 0.05, by Student’s
t-test). Values are the mean SEM (number of joints).
*
0
o.2
t
0.04
20
:
25
30
35
40
45
50
55
60
65
70
I
75
Age (yean)
Figure 4. Decrease in the mean JS measurement with age. MCP 1
and PIP I of digits 2-4 of both hands were measured in baseline
radiographs of 16 women and 18 men with rheumatoid arthritis.
Significant linear regressions fitted the female and male PIP and the
male MCP measurements (Y = -0.0048X + 1.033; Y = -0.0044X
+ 1.20; Y = -0.0048X + 1.82, respectively), but the correlations
were weak (r = 0.32, 0.27, and 0.23, respectively). W = women;
M = men. (See Figures 1 and 3 for further definitions and details.)
Changes in JS measurements after baseline. Because followup radiographs were taken at least annually, it was possible to measure the absolute change in
individual JS measurements after baseline. Table 3
shows that while on a paired basis, MCP I , PIP 1, and
hand 1 values were all reduced significantly in women
after 1 year, there was no such change in men; JSN
scores also increased significantly in these joints in
women, but not in men. (To prevent possible bias,
only those joints providing the respective JS rneasurements were used for calculating the change in JSN
score.) The values for t decreased significantly only in
men (in MCP t2), whereas the JSN scores increased
significantly in MCP tl and r2 measurements in men.
Hand 1 values decreased significantly only in women,
but the comparable increase in JSN scores was not
significant.
Since women’s joints are smaller, the reduction
896
JAMES ET AL
--.
MCPI: W
PIPI: w
MCPI: M
PIPI: M
A
0
A
- - - .o-
A
A
A
4
A
o.2
t
0.04:
0
2
:
4
:
6
:
8
:
10
:
12
:
14
:
16
:
18
:
20
:
22
:
24
Disease Duration (years)
Figure 5. Decrease in the mean JS measurement with disease
duration. Significant linear regressions fitted the female and male
PIP and the female MCP measurements (Y = -0.017X + 0.90; Y =
-0.019~ + 1.09; Y = -0.025X + 1.46, respectively), but the
correlations were weak (r = 0.36,0.43, and 0.35, respectively). (See
Figures 1 , 3, and 4 for definitions and details.)
score increased, which is consistent with women's
smaller joint size.
Comparing changes in JS measurements and
JSN scores after baseline. The relatively poor correlation between JSN score and JS measurement was
highlighted when the paired changes after baseline
were compared (Figures 7A and B).
For MCP joints, the change in JSN score was
weakly inversely correlated (r = -0.23) with the JS
measurement change, although the linear regression
was significant. Nevertheless, as Figure 7A shows,
there were a substantial number of joints whose JS
measurements changed considerably (increasing as
well as decreasing) but whose JSN scores changed
very little. Conversely, there were also joints which
changed in score, with little change in measurement.
For PIP joints (Figure 7B), there was no correlation (r = 0.04) between the change in the JSN score
and the change in the JS measurement. While there
was less absolute variation in PIP 1 than in MCP 1,
there appeared to be a greater number of PIPs that
attracted substantial changes in JSN score in the
absence of commensurate changes in JS measurement.
Four of the 34 patients had radiographs on 5
consecutive semiannual visits. Figures 8a-h show the
mean JS measurements and JSN scores for each time
point. In these patients the mean JS measurements
Table 3. Change in computerized measures of joint space and in
joint space narrowing (JSN) scores in 1 year*
in JS measurement after baseline represents a faster
deterioration. For MCP 1, this was -2.85% in women
versus -1.41% in men; for PIP I, -2.66% versus
0.19%; and for hand 1, -3.53% versus -1.44%, respectively.
Comparing baseline JS measurements and JSN
scores. To facilitate direct comparison with the JSN
scores, only I is considered here. While a moderate
negative correlation related the JSN score to the JS
measurement of the same joint (Figures 6A and B)
(r 0.5 and 0.6 in MCPs and PIPs, respectively), and
significant linear regressions fit the data, it is noteworthy that 1 values had to decrease considerably (especially in MCPs) before the score increased noticeably.
The slopes of the regression lines also indicated that
MCP 1 needed to decrease much more than PIP 1
before the score increased, especially when the JS was
large. Nevertheless, measurements of female JS appeared to fall further than those of men before the JSN
-
Women
Joint, measure
MCP
1
JSN
(n = 16)
Men
(n = 18)
-0.056 f 0.020 (94)t
0.079 2 0.031 (94)t
-0.031
0.081
2
2
0.016 (106)
0.023 (106)
JSN
-0.045 2 0.024 (61)
0.073 f 0.036 (61)
-0.016
0.139
rt
2
0.032 (36)
0.046 (36)t
t2
JSN
-0.032 i 0.026 (37)
0.060 2 0.036 (37)
-0.051 2 0.021 (82)t
0.066 t 0.024 (82)t
-0.029 2 0.011 (94)t
0.094 C 0.030 (94)t
-0.005 2 0.006 (105)
-0.025 t 0.040 (105)
tl
PIP
1
JSN
t
JSN
0.012 2 0.015 (94)
0.094 f 0.030 (94)t
0.0004
-0.018
rt
rt
0.017 (105)
0.040 (103)
* Computerized measurements (I, tl, t2 in millimeters)were of paired
annual radiographs. Joint space narrowing was scored 0 4 (0 = no
narrowing; 4 = no joint space) independently by 6 observers and the
means for each joint were used. (See Table 2 for further definitions
and details.) Values are the mean SEM (number of joints).
t P < 0.05 versus baseline, by Student's paired t-test.
*
RADIOGRAPHIC JOINT SPACE MEASUREMENT
2.6’
2.67
MCPI:Women
MCPI:Men
A
A
A
2.4
2.4-2.2..
897
-
0
PIPI: Men
0
2.0.’
2.0-*
A
1.8.-
A
A
0
-0
1.0.-
c
A
1.6-’,
1.6.-
A
A
A
0
\
-
0
A
\
A A A
0.0
,
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
,
\.
2.2
2.4
JS Measurement (mm)
(Individual join&)
4
0.0
00
\
0.q
0.41
\oo
o a b o o
0.6
0:O
’
PIPI: Women
o
0
2.2..
A
#D
*
CD
.&COO
:
0.2
:
0.4
:
0.6
:
0.8
:
1.0
: - ;
1.2
1.4
:
1.6
;
1.8
:
2.0
:
2.2
f
2.4
JS Measurement (mm)
(Indlvtdual johts)
Figure 6. Baseline joint space narrowing (JSN) scores compared with paired JS I measurements. For individual MCPs (A) and PIPs (B),
significant linear regressions fitted the data for both the women and the men, although the distributions appear asymmetric (for pooled data,
Y = -0.54X + 1.00; Y = -1.35X + 1.64 for MCP I and PIP I , respectively). Correlations were r = 0.50 and 0.52 (MCPs), and 0.72 and 0.62
(PIPS) for women and men, respectively. (See Figures I and 3 for further definitions and details.)
changed remarkably little over the 2-year period: MCP
I values may have decreased slightly (except perhaps
for patient 36, whose values appeared to increase), but
PIP 1 values appeared to drift either up or down,
although the changes were small.
By contrast, the JSN scores changed markedly
and reversibly in magnitude, often in a statistically
significant manner. The magnitude of the changes in
JSN scores also appeared to be unrelated to the
magnitude of changes in JS measurements in the same
joints (apparently more so in PIPs than in MCPs).
Unsurprisingly, in view of their subjective nature, the
standard errors associated with JSN scoring were
much larger than those of the computerized measurements; MCP measurement errors tended to be larger
than those of PIPs because MCPs are more sizevariable.
DISCUSSION
We have developed software that sensitively
and reproducibly measures the radiographic joint
space of the MCP and PIPJoints. The measurements
were remarkably unambiguous and stable, even between patients, especially the t values. Nevertheless,
we tested the software’s reliability in several ways: (i)
Its accuracy was demonstrated by measuring both
squared graph paper and graphically constructed
phantoms that resembled the radiographic joint space.
Deviations of <1% from the standards were obtained
(Figure 2 and Table 1); (ii) The methodology’s precision was shown by remeasuring the MCP and PIP joint
spaces on baseline films of 4 of the patients. Good
coefficients of variation were found (Figure 3); (iii) Its
stability was ascertained by remeasuring the radio-
898
JAMES ET AL
2.0
-
A
A
1.8.'
1.6"
2.0
MCPIkWomen
MCPI: Men
-
1.6"
A
1.4.'
1.4"
A
1.2"
A
1.0"
A
A
1.0"
A
0.8"
A
0.6$
A
0.6"
m.
MAA
f
8
0.2.-
0.0"
," -0.2 -.
t
s -0.4 ..
A
-0.6
-0.6
-
-0.8
-0.8
-
..
-1.0 --1.2 --
-1.0
-
-1.2
-
:I
-I
.4 -.
-1.6 ..
--
-2.04
:
:
:
:
:
.
. .
.
0
. O
ausUna8
0.
=am).
ccOp30
000
.
0
00
0
-1.8
:
PIPI: Women
PIPI: Men
00
0.4"
I
o
0
1.2"
A
o.8y
-1.8
.
1.8"
:
:
:
:
:
JS Measurement change (mm)
(Individualjoints)
I
0
-2.04
: :
:
: :
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2
0
:
0.0
:
0.2
:
0.4
:
0.6
:
0.8
:
1.0
4
1.2
JS Measurementchange (mm)
(Individualjoints)
Figure 7. Change in joint space narrowing (JSN) scores with paired JS I measurements.For individual MCPs (A) and PIPs (B),a significant
linear regressionfitted the MCP data (Y = -0.23X 0.076;r = 0.23) but not the PIP data (r = 0.04). For both data sets, there was considerable
change in JSN scores in 1 year without comparable change in 1 values; similarly,there were changes in the 1 values (especiallyin MCPs) in the
absence of comparable changes in JSN scores. (See Figures 1 and 3 for further definitions and details.)
+
graph of a machined aluminum step-wedge throughout
the measurement process. No drift was evident. Thus,
the methodology has the necessary attributes of a good
measurement system.
Because the radiograph is a 2-dimensional representation of a 3-dimensional structure (Figure l),
correlating finger anatomy with the radiographic image
may not be straightforward. We therefore decided to
make at least 2 joint space measurements: 1 and t. The
1, which was measured between the proximal joint
margin and the leading edge of the distal joint margin,
corresponds to the JS as perceived by the eye, and
should be comparable to clinical evaluations of the JS
embodied in subjective scoring procedures such as
JSN. We have therefore concentrated on 1 values in
this report. The t was measured between the proximal
joint margin and the trailing edge of the distal joint
margin, including the subchondral compacted bone.
We wished to know whether t might be less influenced
by the small differences in finger flexion that were
evident in the radiographs. Since baseline measurements of MCP t were larger than those of MCP 1 but
had similar SEMs, this might be so, but this relationship was not true for PIPs. Metaphyseal medullary
density could also contribute to t , which may, therefore, vary inappropriately with subchondral osteoporosis as well as JS change. Further work (see below) is
required to assess whether t or radiographic measurements other than I are useful (for instance in serial
films, where the joint becomes increasingly malaligned
or eroded-possibilities limited mainly by software
sophistication).
If I truly measures the radiographic JS, then
certain differences in the patient population ought to
899
RADIOGRAPHIC JOINT SPACE MEASUREMENT
(0) Palient Ul (male)
(c) Patienl#34 (male)
(a) PalientK32 (male)
3.01
"'"T
2.5
OS
k--t
0.0J I
1
Time (years)
Time (years)
:
2
4.0~
-
McPl
+McPll
-Mcpe
- Q - PIPI
PlPt
0.0
1.2
1
-*-
2
0. ~ 1~
0.0 0
T h e (years)
1
2
Time (years)
(1) Patient #34 (female)
(d) Patient 134 (male)
(b) Patient 132 (male)
1
(h) Patient X41 (male)
0.4
0.6
c
'.
0.2..
0.6
4?
1.o
0.1
0.2
0.0
-.
0.5
0.0-
1
Time (years)
2
0.0' 1
1
Time (years)
0
1
rime (years)
2
c----e---e
2
0
Time (years)
Figure 8. Variation in joint space narrowing (JSN) scores and paired JS 1 measurements with time. The hands and wrists of 1 woman and 3
men were radiographed 5 times semiannually. The change in mean JSN scores (b, d, f, and h), but not mean 1 values (8, c, e, and g), was often
significant between time points. Unless otherwise shown, n = 6; where n >2, SEMs that are not apparent are within the size of the symbol.
(See Figures 1 and 3 for further definitions and details.)
be detected, for example, between JS values for males
and females (Table 2) since men are morphologically
larger, and between joints in the same finger (MCP >
PIP). Also, JS is expected to decrease in patients with
rheumatic diseases. Significant reductions in the joint
spaces of women over time were also detected on a
longitudinal, paired basis (respectively -60 pm and 30
pm * year-' for female MCP I and PIP I values versus
-30 p m and 5 p m * year-' for males) (Table 3) and
cross-sectionally, according to age and disease duration (Figures 4 and 5). (There was no correlation
between disease duration and age.) When the smaller
JS measurements of women are considered, JS reduction, in the cohort of patients measured here, occurred
at about twice the rate of that in men (and women are
more affected by rheumatic disease).
Again, characteristic systematic differences
should be evident when comparing subjective, discrete
scores with objective, continuous measurements. Thus,
we found that JSN scores (scored independently of the
JS measurements) correlated poorly with JS measurements. Cross-sectionally, a nonlinear relationship between JSN score and JS measurement seemed more
applicable (Figures 6A and B), tentatively explained
by the observers' unwillingness to ascribe a score
increase unless the JS measurement decreased by a
psychologically significant amount (which appeared to
be unrelated to the real change). Figure 7 shows that
large changes in JSN scores occurred, especially in
PIPS, when there were no commensurate changes in
JS measurements. The expected imprecision of the
JSN score is further illustrated in 4 patients for whom
5 semiannual radiographs were available (Figure 8).
The average 1 (and t ) measurements changed little in
these patients, whereas the appropriate JSN scores
fluctuated considerably. These data highlight the non-
JAMES ET AL
linear, variable nature of the subjective scoring process (30).
Figure 7 shows that while JS measurements
declined (as shown in Table 3), they also apparently
increased, both in the MCPs and the PIPs. Decreased
joint space is expected in arthritis through progressive
loss of articular cartilage. However, synovial hyperplasia and fibrotic replacement of the joint tissues is
also symptomatic: proteinase attack on the joint ligaments, coinciding with edema and synovial hyperplasia during early disease, may increase the joint space
acutely, reflecting the characteristic external joint
swelling of RA. Subjective scoring systems may not
detect such increases because, for each patient, scores
from many joints are combined; also, by scoring JS
narrowing, increases, if any, may not be perceived.
To be considered real changes, the observed
changes in JS measurement must exceed significantly
the variability inherent in the measurement process.
Although an estimate of measurement variability can
be derived by remeasuring the same radiographs, such
an estimate fails to include even the slight variations in
hand position that occur between films. Nevertheless,
that many of the changes in JS measurement observed
here, both incremental and decremental, were much
greater than the overall experimental variation calculated (Figure 3), indicates their validity.
Taken together, the results of this study
strongly suggest the validity of the JS measurement 1.
However, it will be essential to determine: (i) how the
computerized measurements are affected by moderate
finger flexion; and (ii) how followup radiography affects the measurement variability.
In the first case, the experienced eye may be
able to allow for small fluctuations in JS measurement
through changes in hand positioning, but the computer
may not. We are undertaking an ethically approved
study to address this issue. Also, a measurement
closely similar to 1, using the same or very similar
radiographic landmarks but derived with a different or
more robust algorithm, may further improve measurement reliability, and our study would be able to assess
this also. In any case, the introduction of sensitive
measurement techniques will require that greater care
with hand positioning be taken during serial radiography. In RA, finger flexion and malalignment is characteristic of later disease, when significant joint
destruction has already occurred. However, diseasemodifying drugs must be taken early on, before much
radiographic change has occurred (when scoring procedures are especially difficult). In patients with early
RA, involuntary finger flexion should be largely absent
and therefore less of a problem.
In the second case, a measure of the variation
inherent in reradiographing the same hand position
would allow the derivation of an objective statistical
limit to this imprecision; from serial measurements it
may then be possible to determine disease effects in
individual joints. While the measurement reproducibility seen here strongly suggests that many of the JS
measurement changes reflect organic change, this remains to be proven. This issue we are also addressing
in an ethically approved study. By identifying sources
of error and their relative contribution to measurement
variability (hand position, the measurement process
itself, software characteristics, 1 versus t, etc.), it will
be possible to improve the reliability of the method.
Such an aim is hindered in subjective scoring procedures because of their imprecision.
The introduction of greater precision and sensitivity into the investigation of radiologic progression
has important consequences for epidemiologic studies
and clinical trials. Although we have focused on MCPs
and PIPs in this study, we have also written software
to measure the knee, the hip, the metatarsophalangeal,
and the distal interphalangeal joints (24,31-33). The
time required to complete a measurement in this study
(-1.5 minutes) is not clinically appropriate, but since
this work was completed, the software has been
adapted to run routinely on 486-microprocessor, 66MHz computers within “Optimas” (BioScan Inc, Edmonds, WA), with considerable savings of time. A
significantly more difficult problem, but an important
step clinically, is to computerize the measurement of
erosions, but progress is occurring in this area too
(18,19,34,35). We believe the increasing use of computer technology will have a major impact on the
radiographic study of joint disease, greatly improving
evaluation of the effects of disease-modifying arthritis
treatments.
ACKNOWLEDGMENTS
We gratefully acknowledge Greg Harper, Jane Dacre, Jacinta Byme, and Melanie Bell for helpful discussions,
and Peter Blower and Bruce Wallin for their support during
this work. We are also grateful to Chris David and Sally
Bradbury for reprographics help.
REFERENCES
1. Kellgren JH, Lawrence JS: Radiological assessment of rheumatoid arthritis. Ann Rheum Dis 16:485493, 1957
RADIOGRAPHIC JOINT SPACE MEASUREMENT
2. Sharp JT, Lidsky MD, Collins LC, Moreland J: Methods of
scoring the progression of radiologic changes in rheumatoid
arthritis: correlation of radiologic, clinical and laboratory abnormalities. Arthritis Rheum 14:70&720, 1971
3. Larsen A, Dale K, Eek M: Radiographic evaluation of rheumatoid arthritis and related conditions by standard reference films.
Acta Radiol 18:481491, 1977
4. Sharp JT, Bluhm GB, Brook A, Brower AC, Corbett M, Decker
JL, Genant HK, Gofton JP, Goodman N, Larsen A, Lidsky
MD, Pussila P, Weinstein AS, Weissman BN, Young DY:
Reproducibility of multiple-observer scoring of radiologic abnormalities in the hands and wrists of patients with rheumatoid
arthritis. Arthritis Rheum 28:16-24, 1985
5. Nance EP, Kaye JJ, Callahan LF, Carroll FE, Winfield AC,
Earthman WJ, Phillips KA, Fuchs HA, Pincus T: Observer
variation in quantitative assessment of rheumatoid arthritis. I.
Scoring erosions and joint space narrowing. Invest Radiol
21:922-927, 1986
6. Kaye JJ, Nance EP, Callahan LF, Carroll FE, Winfield AC,
Earthman WJ, Phillips KA, Fuchs HA, Pincus T: Observer
variation in quantitative assessment of rheumatoid arthrits. 11.
A simplified scoring system. Invest Radiol 22:41-46, 1987
7. Dawes PT: Radiologic assessment of outcome in rheumatoid
arthritis. Br J Rheumatol27 (suppl I):21-36, 1988
8. Turner RA, Flint KP, Semble EL, Agudelo CA: Clinical evaluation of radiographic progression in rheumatoid arthritis. Clin
Exp Rheumatol 8583-586, 1990
9. Weisman MH: Use of radiographs to measure outcome in
rheumatoid arthritis. Am J Med 83 (suppl4B):96100, 1987
10. O’Sullivan MM, Lewis PA, Newcombe RG, Broderick NJ,
Robinson DA, Coles EC, Jessop JD: Precision of Larsen
grading in assessing progression of rheumatoid arthritis in
individual patients. Ann Rheum Dis 49:28&289, 1990
11. Sharp JT, Wolfe F, Corbett M, Isomaki H, Mitchell DM, Furst
DE, Sibley J, Shipley M: Radiological progression in rheumatoid arthritis: how many patients are required in a treatment trial
to test disease modification? Ann Rheum Dis 52:332-337, 1993
12. Genant HK: Methods of assessing radiographic change in
rheumatoid arthritis. Am J Med 7k35-47, 1983
13. Ianuzzi L, Dawson N, Zein N, Kushner I: Does drug therapy
slow radiographic deterioration in rheumatoid arthritis? N Engl
J Med 309:1023-1028, 1983
14. Scott DL, Grindulis KA, Struthers GR, Coulton BL, Popert AJ,
Bacon PA: Progression of radiologic changes in rheumatoid
arthritis. Ann Rheum Dis 4323-17, 1984
15. Fuchs HA, Callahan LF, Kaye JJ, Brooks RH, Nance EP,
Pincus T: Radiographic and joint count findings of the hand in
rheumatoid arthritis: related and unrelated findings. Arthritis
Rheum 31:44-51, 1988
16. Brower AC: Use of the radiograph to measure the course of
rheumatoid arthritis: the gold standard versus fool’s gold.
Arthritis Rheum 33:316-324, 1990
17. Dacre JE, Buckland-Wright JC: The course and outcome of
rheumatoid arthritis: radiological measures of outcome. BaiC
lieres Clin Rheumatol6:39-67, 1992
18. Browne MA, Gaydecki PA, Gough RF, Grennan DM, Khalil SI,
Mamtora H: Radiographic image analysis in the study of bone
morphology. Clin Phys Physiol Meas 8:105-121, 1987
19. Gaydecki PA, Browne M, Mamtora H, Grennan DM: Measurement of radiographic changes occurring in rheumatoid arthritis
by image analysis techniques. Ann Rheum Dis 46:296-301, 1987
90 1
20. Allander E, Forsgren PO, Pettersson H, Seideman P: Computerized assessment of radiological changes of the hand in rheumatic diseases. Scand J Rheumatol 18:291-296, 1989
21. Dacre JE, Coppock JS, Herbert KE, Perrett D, Huskisson EC:
Development of a new radiographic scoring system using digital
image analysis. Ann Rheum Dis 48:194-200, 1989
22. Dacre JE, Huskisson EC: The automatic assessment of knee
radiographs in osteoarthritis using digital image analysis. Br J
Rheumatol28:50&510, 1989
23. James MF, Heald G, Shorter J, Turner RA, Harper GP: Joint
space measurement in hand radiographs using image analysis
(abstract). Br J Rheumatol 30 (suppl 2):62, 1991
24. James MF, Byrne JM, Dacre JE, Harper GP, Heald G, Shorter
JH, Turner RA: Computerised image analysis of radiographic
joint pathology (abstract). In, Medical Imaging: Image Processing and Analysis. London, Institution of Electrical Engineers
Colloquium Digest 051, 1992
25. Byrne J, Heald G, James MF, Kay M, Shorter JH, Dacre JE:
Digital image analysis, a rapid and reproducible method for joint
space measurement (abstract). Osteoarthritis Cartilage 1:60-61,
1993
26. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF,
Cooper NS, Healey LA, Kaplan SR, Liang MH, Luthra HS,
Medsger TA Jr, Mitchell DM, Neustadt DH, Pinals RS, Schaller
JG, Sharp JT, Wilder RL, Hunder GG: The American Rheumatism Association 1987 revised criteria for the classification of
rheumatoid arthritis. Arthritis Rheum 31:315-324, 1988
27. Sharp JT, Young DY, Bluhm GB, Brook A, Brower AC,
Corbett M, Decker JL, Genant HK, Gofton JP, Goodman N,
Larsen A, Lidsky MD, Pussila P, Weinstein AS, Weissman BN:
How many joints in the hands and wrists should be included in
a score of radiologic abnormalities used to assess rheumatoid
arthritis? Arthritis Rheum 28:132&1335, 1985
28. BBN Software Products: RS/l Release 4 Users Guide: Statistical Tools. Cambridge, MA, BBN Software Products, 1988
29. Hoaglin DC, Mosteller F, Tukey JW: Understanding Robust
and Exploratory Data Analysis. New York, John Wiley & Sons,
1983
30. Maxwell C: Sensitivity and accuracy of the visual analogue
scale: a psycho-physical classroom experiment. Br J Clin Pharm
6:15-24, 1978
31. Byrne J, Dacre JE: Digital image analysis of hands (DIP, PIP,
MCP), knees and hips: a rapid and reproducible method for joint
space measurement (abstract). Arthritis Rheum 35 (suppl 9):
5134, 1992
32. Byrne J, Heald G, James MF, Shorter JH, Dacre JE: Measurement of patellofemoral joint space using digital image analysis
(abstract). Br J Rheumatol 32 (suppl 1):82, 1993
33. Byrne J, Jawed S, James MF, Dacre JE: Radiographic tibiofemoral joint space changes over 5 years in patients with knee pain
and correlation with clinical findings (abstract). Br J Rheumatol
32 (suppl2):44, 1993
34. Buchanan HM, Buchanan WW, Cockshott WP, Kamath MV,
Ghista DN: Radiological changes in rheumatoid arthritis: measurement of area of juxta-articular bone as outcome event in
clinical therapeutic trials of antirheumatic drugs. Clin Rheumato1 7:481485, 1988
35. Moreland LW, Reddy MS, Koopman WJ, Webber RL, Alarcon
GS, Jeffcoat MK: Digital subtraction radiography for the assessment of bone changes in rheumatoid arthritis. J Rheumatol
19:1697-1 703, 1992
Документ
Категория
Без категории
Просмотров
1
Размер файла
1 033 Кб
Теги
measurements, using, space, image, hands, computerized, radiographic, joint, analysis
1/--страниц
Пожаловаться на содержимое документа