Professional Documents
Culture Documents
MAY 1995
Abstract-A new instrument based on the principle of color- The pattem of healing is poorly understood and clinicians
coded structured light has been developed to measure the area, need an instrument which will give an objective measure of
volume, and color of skin ulcers and pressure sores. A set
of parallel stripes of alternating colors is projected onto the
the dimensions and color of a wound so that changes in the
ulcerated skin and recorded by a CCD camera. The color, short term can be detected. The risk of introducing infection
width, distance, and coding of the stripes have been adjusted to should be minimized by ensuring that the instrument does not
maximize the precision of the instrument. Algorithms have been touch the patient.
developed to extract the stripes and determine their centers to Owing to the lack of precise measurement methods, physi-
better than 0.1 mm, even under circumstances where the skin
reflectivity varies widely over short distances. A depth map for cians have tried to establish an empirical scale for the classi-
the ulcerated skin is calculated by triangulation. The volume of fication of ulcers and sores. Shea [lo] describes a four-point
the ulcer is that sandwiched between the base of the lesion and schedule, for example. Additionally, the measurement of diam-
the original healthy skin, which is simulated by a cubic spline eter has been suggested, but wounds are rarely sufficiently reg-
interpolation between the surrounding areas of healthy skin. The
ular to allow a diameter to be clearly defined. Bulstrode et al.
instrument measures the ulcer’s volume with a precision of about
5% provided the ratio of the ulcer’s volume to area is greater [6] trace ulcer edges through plastic sheets and are able to mea-
than 0.4 Em. The technique has been used in hospital clinics for sure areas to within 25%, but at a risk of infecting the wound,
a wide variety of wounds. Volume measurements are of greater interest since ulcers
tend to heal from the base rather than from the borders.
I. INTRODUCTION Raw assessments have been made by filling the ulcer with
saline from a calibrated syringe. Some of the saline will be
M ANY chronically ill people suffer from skin ulcers
which may develop from minor injuries when the
elasticity of the skin is decreased by a variety of causes such
absorbed by the wound and the estimated precision of 20%
is probably optimistic [ 5 ] . Stereo photogrammetry produces
as old age, a side effect of medication, or a weakened immune accurate results for the volume, but it is slow because the
system. Another common cause is poor blood perfusion in photographs must be processed and traced manually [6], [7].
certain regions of the skin induced by disease, for example Ulcers are also assessed by their optical properties. Ring [9]
varicose veins, and by behavioral influences such as smoking. has measured an infrared line scan through an ulcer. Within the
Anaemia, diabetes, and vitamin deficiencies are examples of ulcer boundary, the thermogram shows reduced temperatures
other contributory factors. caused by the absence of skin, which has a high absorption
Pressure sores are closely related to skin ulcers and origi- (and hence emission) coefficient for infrared, and the presence
nate when immobile patients are exposed constantly to high of fluids which cool the lesion by evaporation. The temperature
pressures on bony prominences. Between 6% and 8% of of the perimeter of the ulcer does indicate the effectiveness
the hospital population suffer from pressure sores [4], which of blood perfusion, though the raised temperatures rarely
respond only slowly to treatment. corresponded with the ulcer boundary, thus ruling out an area
There is general agreement that there are two main phases measurement. Reflectivities in the visible and near infrared
of treating wounds. The first cleansing phase is characterized regions are able to provide data for predicting the healing
by the presence of necroses, pus, or fibrin in the wound [3]. time. Afromowitz et al. [ l ] are able to forecast healing times
This phase lasts for a few weeks during which the wound for bums which have similar optical properties to skin ulcers.
may be cleaned by the application of enzymes and treated They claim that healing times may be predicted with better
surgically and with 0.9% sodium chloride solution [12]. The accuracy than experienced physicians from a scatter plot of
second granulatiodepithelialization phase may last for several the ratio of green (550 nm) to infrared (880 nm) against the
months. Surveys have shown that there are literally hundreds ratio of red (640 nm) to infrared.
of treatments used during this phase [ 131. Amqvist et al. [3] classify healing ulcers by image pro-
Manuscript received July 1 1 . 1993; revised January 25, 1995. The work cessing of color photographs. They identified black areas with
of P. Plassmann was partly supported by the German Academic Exchange necrotic eschar, yellow with necrosis and fibrin, and red with
Service (DAAD).
B. F. Jones is with the Department of Computer Studies, University of granulation tissue. The percentages of the colors vary as the
Glamorgan, Pontypridd, Mid Glamorgan CF37 lDL, U.K. wound heals.
P. Plassmann is with the Department of Electronics and Information The aim of this work is to produce an instrument which
Technology, University of Glamorgan, Pontypridd, Mid Glamorgan CF37
IDL, U.K. assesses the status of the healing of skin ulcers and pressure
IEEE Log Number 9410034. sores subject to the constraints that:
0018-9294/95$04.00 0 1995 IEEE
JONES AND PLASSMANN: INSTRUMENT TO MEASURE DIMENSIONS OF SKIN WOUNDS 465
Fig. 1. (a) A schematic representation of the camera and projector housing. (b) A photograph of the instrument in use.
1) no contact is made with the patient, easy disinfection. Electrical safety is ensured by the use of
2) the instrument is easy to operate, fast, and has a precision insulating materials for the housing and a double-insulated
of 5%, transformer; the maximum voltage used is 48 V.
3 ) the instrument should be portable and operate in normal The base of the housing is a rectangular frame which defines
hospital conditions. the optimum distance and field of view when it is held close
Our instrument is based on the technique known as struc- to the skin (see Fig. l(b)).
tured light [14], [15], adapted for conditions where the re- The camera is able to operate at f = 1.4 with a minimum
flectivity of the surface is low and variable. Algorithms to illumination of 25 lux and a color temperature of 3300 K. Its
detect and identify stripes and to reconstruct the damaged skin horizontal resolution is 43 pixels per mm; its vertical resolution
surface are described. is 88 pixels per mm for the green channel and 44 pixel per
mm for the blue and red channels. Two images of the wound
are captured: the first illuminated by unfiltered light from the
11. DESCRIPTION
OF THE INSTRUMENT
HARDWARE halogen bulb, the second illuminated by a slide of alternating
The instrument comprises a PAL color CCD PULNIX red and blue stripes, with a single reference green stripe in the
camera and a projector with a 24-V 250-W tungsten halogen center. The stripes are projected in the direction corresponding
bulb. Both are housed in a portable box which is 500 mm to the camera’s higher resolution. The camera electronics
high, 300 mm wide, and 100 mm deep; the dimensions of generate a standard PAL TV signal with 575 columns and
this box are reduced by using two mirrors (Fig. l(a)). The 330 rows. The dynamic range of the camera is about 50 dB
housing is completely closed and has a smooth surface for (compared to 200 dB for a human eye); small dynamic ranges
466 IEEE TRANSACTIONS O N BIOMEDICAL t N G I N E t K I N G . VOL 42, N O 5. MAY 1995
loo[ Transmittance / X
L I Projector (Py. Pz)
6 Wavelength / nm
"1
Y
Wound Surface
1
6o Tranamlttance / X
Fig. 4. The geometry of triangulation.
40 brlpht blue
cyan white
I blue
Gd71 yellow
i I
bottom top
Fig 5 Vertical luminance scans for the RGB channels perpendicular to the
projected stripes Fig. 6 . The RGB color cube with a volume enclosing red granulation tissue
colors.
picture are checked to ensure that they are not capped at the
maximum gray level. If this is the case, the operator is warned smoothed by a simple 1-D moving average filter. This filter
to repeat the procedure, but with a reduced camera aperture. generates a new gray level, g(y), at position y from the gray
Typical luminance scans for the RGB channels perpendicular level of the original image, f ( y ) , using
to the stripes are shown in Fig. 5. Scans are shown for pictures
Y+2
with and without stripes. Cross-talk between the channels is
evident, especially in the green channel. However, the only g(Y) = 0.2 * f(+ (1)
i=y-2
point in the green channel where the luminance for the picture
with stripes exceeds that without stripes corresponds to the Reducing noise in the spatial domain rather than using a
location of the single projected green stripe. low-pass filter in the frequency domain has two advantages: it
The color of the image captured with white light is mea- is fast to compute and it preserves the positions of the peaks
sured; the ratio of the numbers of black to yellow pixels within and troughs of the stripes.
the wound's area is an indicator of necroses, and a large The luminance of the stripes is calculated using a dynamic
number of red pixels indicates the presence of granulation threshold. The threshold line, which defines the width of the
tissue. The ulcerated area is outlined using a mouse to trace green stripe, is derived from the image without stripes. In the
the perimeter, and the color of each pixel within the ulcer is G channel in Fig. 5 , the green stripe is the only region for
measured after convolving with a 3 x 3 averaging filter. The which the image with stripes has a larger luminance than the
color is represented as a point in an RGB color cube. Fig. 6 image captured with white light. This is used to locate the
shows a volume in a typical RGB color cube enclosing all green stripe. The high-frequency oscillations in the G channel
the coordinates which represent red granulation tissue. The arise from cross-talk from the R and B channels. The low-
volume is defined as a reference after manually analyzing a frequency oscillation follows the changing reflectivity of the
large number of ulcer photographs. The impression of red is skin; in the white light image, this oscillation is just large
caused by the balance between the three primary colors, rather enough to enclose the stripes.
than the absolute intensity of the red channel at a specific #en the green stripe is identified, the centroid for pixels
point. This may be seen in Fig. 5 where the central area within the stripe is calculated. In Fig. 5 , the plot, which is
appears red, but the R channel has a lower luminance than labeled (blue-red), shows the difference between the blue and
in the surrounding skin where the red is balanced by strong red planes, and the median of the local peaks and troughs. The
luminances in the G and B channels. Following the proposal edges of the stripes are defined to be the points of intersection
of Amqvist et al. [3], the pixels are divided into four classes: of the median and the (blue-red) scan; these pixel positions
red, yellow, black, and unclassified. The counts are displayed are labeled a and b.
on the computer screen as a histogram. The centroid, C, is given by
The area and volume measurements start with a procedure to
h I b
locate the center of the stripes, after which the remainder of the
image can be discarded. In other structured light applications
y=a
[14], [15], standard algorithms for finding the center of stripes
operate in conditions where the surface is opaque and has a The center of the stripe algorithm is based on that of Takagi
uniform reflectivity and color. In this application, the lack of and Hata [ 111, and calculates the center typically to within one-
uniformity of the surface demands a more robust algorithm. tenth of a pixel width. The role of the dynamic threshold is
Noise is introduced to the original image from a variety of crucial in maintaining the true width of the stripe when the
sources, such as the camera electronics. The raw image must skin reflectivity changes. The dynamic threshold is unsuited
therefore be smoothed. The center of stripe calculation uses for the R and B channels where the stripes are enclosed in the
a single column of pixels at a time and the raw image is white light luminance. In this case, the stripes are enhanced
468 IEEE TRANSACTIONS ON BIOMEDICAL tNGINtERING, V O L 42. NO S. MAY I Y Y S
area v o l u m e
cm’ c p 3
I
standard dev /f \ \
04
cl 1 2 3 4 5
time / months
1 9 16 25 36 49 62
area/cm
Fig. 9. Change in area and volume of a pressure sore over a five-month
(a) period.
Fig. 9 shows the area and volume measurements for one [5] W. Berg, C. Traneroth, A. Gunnarson, and C. Losing, “A method for
patient with a pressure sore located on the buttocks over measuring pressure sores,” Lancet, vol. 335, pp. 1445-1446, 1990.
[6] C. J. K . Bulstrode, A. W. Goode, and P. J. Scott, “Stereophotogrammetry
a period of six months. The uncertainty of the first two for measuring rates of cutaneous healing,” Clin. Sci., vol. 71, pp.
measurements is high because they were made with an early 4 3 7 4 4 3 , 1986.
version of the instrument; the later version has an improved [7] R. A. Frantz and D. A. Johnson, “Stereophotography and computer-
ized image analysis: A three-dimensional method of measuring wound
precision. healing,” Wounds: Compendium Clin. Res. Practice. vol. 4, pp. 58-64,
This experiment helps to confirm the view of some physi- 1992.
cians that pressure sores heal from the base first, which is [8] H. F. Kuppenheim and R. R. Heer, “Spectral reflectance of white and
negro skin between 440 and 1000 nm,” J. Applied Physiology, vol. 4,
confirmed by the rapid decrease in volume between one and pp. 800-806, 1952.
three months. Then, the wound heals from the boundary, which [9] E. F. J. Ring, Video Thermal Imciging in Thermological Method.\.
Heidelberg: Verlag Chemie, 1985, pp. 101-1 1 I .
shows as a slow decrease in area. [IO] J. D. Shea, “Pressure sores: Classification and management,” Clin.
Orthop.. vol. 112, pp. 89-100, 1975.
[ I I] Y. Takagi and S. Hata, “High speed precise 3-D vision sensor using the
VI. CONCLUSIONS slit light method,” in Pmc. Int. Wkshp. Ind. Applical. Machine Vision.
The area and volume of skin ulcers and pressure sores can be Tokyo, Japan, pp. 235-239, 1987.
[I21 G. Eriksson, A. E. Eklund, and S. Zetterquist, “Comparison of different
measured rapidly using the principle of color-coded structured treatments of venous leg ulcers,” Current Therapeutic Res., vol. 35, pp.
light. 678-684, 1984.
[I31 A. L. Knight, “Medical management of pressure sores,” J . Family
The area can be measured with a precision of about 5% Practice, vol. 27, pp. 95-100, 1988.
provided that the area is greater than 9 cm2 and is at a distance [I41 K. L. Boyer and A. C. Kak, “Color encoded structured light for rapid
of less than 3 cm. active ranging,” IEEE Trcms. Pattern Anal. Muchine Intell., vol. PAMI-9,
pp. 14-28, 1987.
The volume can be measured with a precision of about 5% [I51 M. Goharl’ee and M. Raaberg, “Applications of structured light for
provided the volume-to-area ratio is greater than 0.4 cm, i.e., volume measurement,” in P roc. SPIE: Optics. Illumination cmd Imcige
the wounds are not shallow and large in area. Sensing,for Machine Vision, vol. 728, pp. 279-291, 1988.
ACKNOWLEDGMENT
The authors would particularly like to thank Prof. E. F. J.
Ring of the Royal National Hospital in Bath and of the Univer- Bryan F. Jones received the B Sc (Honors) in
physics from Manchester University in 1965 and
sity of Glamorgan, who first introduced them to the problem the Ph D from Nottingham University i n 1968
of wound measurement and helped with many stimulating After developing computer models of high-
discussions. They would also like to thank Dr. K. Harding pressure gas discharges in the Thorn EM1 research
laboratories Leicester, he joined the University
of the Wound Healing Research Unit, University of Wales of Glamorgan and is now Redder and Research
College of Medicine, Cardiff, and Prof. Ring for assessing the Coordinator in the Department of Computer Studies
instrument in their clinics and for many helpful suggestions on
improving the instrument. Thanks also go to the referees for
their constructive comments which have enabled us to improve
9 His mdin research interests are in medical image
processing and the application of genetic algorithms
to the automatic testing ot software
the paper.
REFERENCES
Peter Plassmann received the Dipl.-lng. degree
[ I ] M. A. Afromowitz, G. S. Van Liew, and D. M. Heimbach, “Clinical in electrical engineering from the Fachhochschule
evaluation of burn injuries using an optical reflectance technique,” IEEE Hannover, Germany in 1989 and the Ph.D. degree
Trc“ Biomed. Eng., vol. BME-34, pp. 114-127, 1987. in electrical engineering from the University of
[2] R. R. Anderson and J. A. Parish, “The optics of human skin, J. Glamorgan. United Kingdom, in 1992.
Investigcitive Dermatology, vol. 77, pp. 13-19, 1981. He is currently a Research Fellow at the Depart-
[3] J. Arnqvist, L. Hellgren, and J. Vincent, “Semiautomatic classification ment of Electronics and Information Technology at
of secondary healing ulcers,” in Proc. 9th Int. Con5 Pattern Recognition, the University of Glamorgan. His current research
New York, IEEE, pp. 459461, 1988. interests are in the areas of image processing and
[4] J. C. Barbenel, M. M. Jordan, and S. M. Nicol, “Incidence of pressure the design of medical instrumentation.
sores in the greater Glasgow Health Board area,” Lancet, p. 548, 1977.