Journal of Medical Engineering & Technology, Vol. 33, No.

6, August 2009, 426436

Area assessment of psoriasis lesions for PASI scoring

M. H. AHMAD FADZIL*{, DANI IHTATHO{, AZURA MOHD AFFANDI{ and S. H. HUSSEIN{
{

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

Department of Electrical & Electronic Engineering, Universiti Teknologi PETRONAS, 31750 Tronoh,
Perak, Malaysia
{
Dermatology Department, Hospital Kuala Lumpur, Jalan Pahang, 50586 Wilayah Persekutuan, Malaysia

Psoriasis is a skin disorder which is caused by a genetic fault. Although there is no cure
for psoriasis, there are many treatment modalities to help control the disease. To
evaluate treatment ecacy, the current gold standard method, PASI (Psoriasis Area and
Severity Index), is used to measure psoriasis severity by evaluating the area, erythema,
scaliness and thickness of the plaques. However, the determination of PASI can be
tedious and subjective. In this work, we develop a computer vision method that
determines one of the PASI parameters, the lesion area. The method isolates healthy and
healed skin areas from lesion areas by analysing the hue and chroma information in the
CIE L*a*b* colour space. Centroids of healthy skin and psoriasis in the huechroma
space are determined from selected sample. The Euclidean distance of all pixels from
each centroid is calculated. Pixels are assigned to either healthy skin or psorasis lesion
classes based on the minimum Euclidean distance. The study involves patients from
dierent ethnic origins having three dierent skin tones. Results obtained show that the
proposed method is able to determine lesion areas with accuracy higher than 90% for 28
out of 30 cases.
Keywords: Psoriasis; Area assessment; Segmentation; PASI score

1. Introduction
Psoriasis is a chronic, inammatory, non-contagious skin
disorder which is characterized by red plaques covered by
silvery-white scales. Most researchers agree that it is caused
by a genetic fault in which the immune system is mistakenly
triggered to produce skin cells faster than normal [1].
Psoriasis aects about 3% of the world population. There
are ve types of psoriasis: plaque, guttate, inverse, pustular
and erythrodermic. Plaque is the commonest form of
psoriasis, accounting for 80% of cases. A typical plaque
psoriasis is shown in gure 1.
There is no cure for psoriasis; however, there are many
treatments available to help control the disease [2]. During
treatment, a dermatologist will monitor the extent of the
psoriasis continuously to ascertain the treatment ecacy
[3]. The current gold standard method used to assess
psoriasis severity is the Psoriasis Area and Severity Index

(PASI) [4,5]. PASI assesses four body regions: the head,

trunk, upper extremities and lower extremities. For each
region, the surface area involved, redness, thickness and
scaliness of the plaques are determined. The scores from the
regions are weighted and summed to give a PASI score
ranging from 0 to 72 as follows:
PSAI 0:1 Rh Th Sh Ah 0:2 Ru Tu Su Au
0:3 Rt Tt St At 0:4 Rl Tl Sl Al ;

where A area (06), R redness (04), T thickness (04)

and S scaliness (04), and the subscripts h, u, t and l
indicate head, upper extremities, trunk and lower extremities, respectively.
Although PASI is the gold standard for evaluating the
extent of psoriasis, it is not used in daily practice. The
determination of PASI parameters is time consuming [4,5].

*Corresponding author. Email: fadzmo@petronas.com.my

Journal of Medical Engineering & Technology
ISSN 0309-1902 print/ISSN 1464-522X online 2009 Informa UK Ltd.
http://www.informaworld.com/journals
DOI: 10.1080/07434610902744066

427

Area assessment of psoriasis lesions

Table 1. PASI area score.

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

Figure 1. Plaque psoriasis lesion.

In addition, the four parameters are visually determined and

may result in inter-individual variations even by experienced
dermatologists. Hence, an objective, practical and reliable
method is required. Table 1 shows the PASI scoring of
lesion area parameter A. In determining the lesion areas
there are typically three objects in the image to deal with,
namely the background, healthy skin areas and lesion areas.
The background colour is selected so that it can be
dierentiated easily from the region of interest (ROI), i.e.
the body area. Green can be used as the background colour
since it is clearly distinguishable from the skin when using
the red band of the RGB colour space [6].
A number of ways to segment psoriasis lesions from
healthy skin have been reported in the literature. A
common approach is to threshold on greyscale images
[7,8]. The drawback to this method is that most of the dark
regions (due to insucient illumination on surface curvature) in the skin image are also classied as lesion areas. It
has been found that subtracting the green band from the
blue band in RGB colour space can eectively discriminate
lesions from healthy skin [6,9]. However, this method may
not work well for dark skin cases. A neuro-fuzzy classier
has been used to segment psoriasis lesions [10]. Homogenous regions of healthy skin and psoriasis lesion were
extracted automatically using colour and texture information, and these regions were used to train the classier.
However, due to variation of skin colour, the classier
needs to be trained for every image in order to segment the
lesion accurately. Moreover, this method is computationally high.
The goal of this research is to develop a computer vision
method that can perform eective segmentation of ROI
and psoriasis regions for lesion area assessment in PASI
scoring.

Lesion area

0%
510%
10 to 530%
30 to 550%
50 to 570%
70 to 590%
90 to 100%

0
1
2
3
4
5
6

Chinese and Indian ethnic origins. Patients were identied

from the Dermatology Clinic, Hospital Kuala Lumpur.
Patients with other types of psoriasis or other dermatological diseases are excluded. Based on the PASI standard,
images of face, anterior and posterior sides of the trunk,
and both left and right upper limbs and lower limbs were
digitally photographed for each patient. Sample images
from patients of three dierent ethnic origins are shown in
gures 2, 3 and 4. The face images are not included for
privacy reasons.
Since psoriasis lesions can appear in a wide variety of
colours, segmentation based on a selected colour will be not
eective [11]. However, a lesion can be recognized by its
colour dissimilarity from healthy skin. CIE L*a*b* is
widely used to measure dissimilarity between two colours.
It is a colour space which is perceptually linear with the
human visual system. As shown in gure 5, L* represents
lightness ranging from 0 to 100, a* represents degree of
rednessgreenness (negative values for greenness and
positive values for redness), and b* represents degree of
yellownessblueness (negative values for blueness and
positive values for yellowness) [1214]. As most digital
cameras store images in standard Red Green Blue (sRGB)
format, conversion into CIE L*a*b* is necessary.
There are two steps to transform from sRGB to CIE
L*a*b*. Firstly, an sRGB image is transformed into the
CIE XYZ colour space using the linear transformation
given in equation (2). The resulting CIE XYZ image is then
transformed into CIE L*a*b* image using equation (3).
2

3 2
X
0:4124 0:3576
4 Y 5 4 0:2126 0:7152
Z
0:0193 0:1192

3
3 2
0:1805
RsRGB
0:0722 5  4 GsRGB 5
BsRGB
0:9505

L* 116Y=Yn 1=3  16
h
i
a* 500 X=Xn 1=3  Y=Yn 1=3
h
i
b* 200 Y=Yn 1=3  Z=Zn 1=3

2. Method
Data were obtained based on the PASI standard by
photographing patients with plaque psoriasis from Malay,

where Xn, Yn, Zn are the tristimulus values of the reference

illuminant (light source).

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

428

M. H. Ahmad Fadzil et al.

Figure 2. Patient 2 (Malay ethnic origin): (a) anterior of right arm, (b) anterior of left arm, (c) posterior of left arm, (d)
posterior of right arm, (e) anterior of trunk, (f) posterior of trunk, (g) anterior of legs, and (h) posterior of legs.

Figure 3. Patient 5 (Chinese ethnic origin): (a) anterior of right arm, (b) anterior of left arm, (c) posterior of left arm, (d)
posterior of right arm, (e) anterior of trunk, (f) posterior of trunk, (g) anterior of legs, and (h) posterior of legs.

There are other parameters that can be derived from CIE

L*a*b* to interpret colour, namely the hue, hab, and
chroma, C*ab. Hue represents the dominant wavelength of
colour and chroma represents the saturation of the colour.

The hue and chroma are obtained from the CIE L*a*b*
image using equations (4) and (5), respectively.
hab tan 1 b* =a*

429

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

Area assessment of psoriasis lesions

Figure 4. Patient 8 (Indian ethnic origin): (a) anterior of right arm, (b) anterior of left arm, (c) posterior of left arm, (d)
posterior of right arm, (e) anterior of trunk, (f) posterior of trunk, (g) anterior of legs, and (h) posterior of legs.

Therefore, selecting a* 0 as threshold value will not give

accurate segmentation.
However, the histogram of the CIE a* band shows two
modes with a clear separation. This allows eective
thresholding on this band using the method of Otsu [17].
The results of ROI segmentation are shown in gure 6.
2.2. Segmentation of psoriasis lesions

Figure 5. CIE L*a*b* colour space (reproduced from [15]).

C*ab

p
a2 b2

2.1. Segmentation of region of interest (ROI)

Red is found to be the dominant colour of human skin [16].
In CIE L*a*b* colour space, the a* band represents the
degree of rednessgreenness. Thus, by using green colour
for the background, it is expected that pixels belonging to
the ROI (human skin) will have positive values and pixels
belonging to the background will have negative values for
the CIE a* band. However from preliminary investigations,
it is found that skin areas bordering the background can
have negative values of a*. These areas are not well
illuminated due to the curvature of human body limbs.

In the CIE L*a*b* colour space, the Euclidean distance

between two colours is perceptually linear with its colour
dierence. We use hue and chroma values of CIE L*a*b* to
segment psoriasis lesions from healthy skin. Pixel samples of
healthy skin and psoriasis lesions were segmented manually
from each image, giving an equal amount of pixels. An
example of distribution of healthy skin and psoriasis pixels
in the huechroma plane is shown in gure 7(a).
From the pixel samples, centroids of healthy skin and
lesion class for each image are calculated using the following
equations:
N
1X
huei ;
N i1

N
1X
chromai ;
N i1

ho

co

where N number of pixels.

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

430

M. H. Ahmad Fadzil et al.

Figure 6. Patient 1: (a) anterior trunk image, (b) segmented ROI and (c) segmented lesion.

Figure 7. Pixel distribution in the huechroma plane for the image in gure 6(a). (a) Distribution of healthy skin and
psoriasis, (b) centroids of healthy skin (O) and psoriasis (X), (c) healthy skin and psoriasis lesion region.

Centroids of healthy skin and psoriasis class for

gure 6(a) are shown gure 7(b). The Euclidean distance
of all possible pixels in huechroma space from each
centroid is calculated. Each pixel is assigned to the class
with minimum distance. As a result, we can visualize a
threshold line as shown in gure 7(c).
Figure 8 shows the distribution of all pixels from
gure 6(a) before and after being classied. The result of
lesion segmentation is shown in gure 6(c).
It is found that in face images, regions such as hair, eye,
eyebrow, nostril, moustache, lips and beard are also
segmented as lesion. These regions can however be
dierentiated from psoriasis lesion using the L* value.
Minimum, mean and maximum values of L* from
segmented lesions were selected as references of dark
region, psoriasis lesion and bright region respectively.
Distances of all segmented lesion pixels from three

references are calculated. Each pixel is assigned to the

class with minimum distance. Pixels that belong to dark or
bright regions are excluded from psoriasis region. This
approach is also applied to leg images to eliminate dark
regions that occur due to insucient illumination.
While under treatment, lesions can heal gradually and
usually appear darker in contrast to the healthy skin, as
shown in gure 9(a). Dermatologists categorize healed
lesions as healthy skin, and our proposed method is able to
segment psoriasis without misclassifying healed lesions as
clearly shown in gure 9(b).
2.3. Misclassied objects
In some cases, non-lesion objects such as lips, eyes, eyebrow
and nipples are misclassied as psoriasis lesions. This is due
to the colour of these non-lesion objects being dissimilar to

431

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

Area assessment of psoriasis lesions

Figure 8. (a) Histogram of gure 6(a) excluding the green background, and (b) after being segmented into healthy skin and
psoriasis.

Another example of misclassication is shown in

gure 12. In this case, nipples are misclassied as psoriasis
lesions (indicated by red circles).
2.4. PASI area scoring
Once areas of ROI and psoriasis lesion are obtained, area
percentage of the lesion can be calculated by the following
formula:
Figure 9. (a) Psoriasis lesion and healed lesion, and (b)
segmented lesion.

Area percentage

Area of lesion
 100%
Area of ROI

Using table 1, a score for the PASI area is determined.

that of normal skin. An example of misclassied objects is
given in gure 10. In this case, lips, eyes and eyebrow are
misclassied as psoriasis lesions, as shown in gure 10(b).
The distributions of pixels belonging to normal skin,
psoriasis lesions and lips in the huechroma plane are
shown in gure 11. The lip pixels overlap the lesion pixels;
therefore, lips are misclassied as lesions.
In order to exclude the misclassied objects from the
segmented image, all pixels belonging to that object should
be selected rst. These objects are composed from subsets
of connected pixels. By selecting several pixels manually
from the misclassied objects, we will be able to select all
pixels belonging to these objects by using a region-growing
technique, as shown in gure 10(c). Once all the misclassied objects are selected, they are subtracted from the
segmented image to obtain the correct segmented image as
shown in gure 10(d).

3. Result and discussion

The segmented images are compared with the reference
images to measure the accuracy of the proposed lesion
segmentation method. Reference images are obtained by
segmenting the lesion area manually from the digital images
as benchmark for accuracy. Accuracy is calculated using
the following formula [18]:

Accuracy

True positive True negative

 100%
Total positives Total negatives
9

The true positive is the area which is categorized as

lesion by reference segmented images and also by the

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

432

M. H. Ahmad Fadzil et al.

Figure 10. Misclassied objects: (a) original image (edited for privacy reasons), (b) segmented lesion, (c) non-lesion objects
misclassied as lesions, and (d) segmented lesion after correction.

Figure 11. Distribution of pixels belonging to normal skin, psoriasis lesions and lips in the huechroma plane.

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

Area assessment of psoriasis lesions

433

Figure 12. Misclassication: (a) original image, (b) segmented image with nipples misclassied, and (c) segmented image after
correction.

Table 2. Accuracy of lesion segmentation method. The shaded

areas indicate where accuracies lower than 90% were achieved.
Skin colour

Patient

Fair skin

Brown skin

7
Dark skin
8

Body area

Accuracy (%)

Head
Trunk
Upper extremities
Lower extremities

93.19
97.54
98.17
95.17

Head
Trunk
Upper extremities
Lower extremities

98.72
98.13
94.48
92.21

Head
Trunk
Lower extremities

99.12
99.78
99.41

Head
Trunk
Upper extremities
Lower extremities

96.26
96.96
99.24
95.5

Head
Trunk
Upper extremities
Lower extremities

95.29
96.28
94.1
83.75

Head
Trunk
Upper extremities
Lower extremities

93.62
95.71
95.61
73.18

Head
Trunk
Upper extremities
Lower extremities

99.65
96.75
95.84
92.1

Trunk
Upper extremities
Lower extremities

99.21
98.53
97.34

proposed method. The true negative is the area which is

categorized as normal skin by reference segmented
images and also by the proposed method. Total positive
and total negative are the areas categorized by reference
segmented images as psoriasis lesion and normal skin,
respectively.
The proposed lesion segmentation method was applied
to images of 30 body regions from eight patients: three
patients with fair skin colour, three patients with medium
skin colour and two patients with dark skin colour. The
selection of patients is based on the clarity of lesion border
to ensure minimal error in manual segmentation. The
accuracies achieved by the segmentation method are given
in table 2.
Accuracies higher than 90% were achieved except for
two cases: the lower extremities of patients 5 and 6
(indicated by the shaded area in table 2). Images of these
regions of patients 5 and 6 along with the segmented images
and reference segmented images are shown in gures 13 and
14, respectively. It is seen that the lesions on the calves are
not well segmented by our method, as indicated by the red
circles in gures 13(b), 13(e), 14(b) and 14(e). These lesions
are covered by thin scales which occupy the same region as
normal skin in the huechroma plane. Therefore, these
regions are misclassied as normal skin.
4. Conclusion
Lesion area is one of the parameters of PASI scoring. In
this work, a procedure to assess psoriasis lesion area has
been developed. The primary objective is to calculate the
area percentage of lesions. The method eectively extracts

M. H. Ahmad Fadzil et al.

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

434

Figure 13. Lower extremities of patient 5: (a) anterior, (b) segmented lesion of image (a), (c) reference lesion of image (a), (d)
posterior, (e) segmented lesion of image (d), (f) reference lesion of image (d).

435

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

Area assessment of psoriasis lesions

Figure 14. Lower extremities of patient 6: (a) anterior, (b) segmented lesion of image (a), (c) reference lesion of image (a), (d)
posterior, (e) segmented lesion of image (d), (f) reference lesion of image (d).

the ROI by thresholding the a* band of CIE L*a*b* colour

space. Healthy and healed skin are successfully isolated by
analysing pixels in the huechroma plane and classifying
the pixels based on minimum distances from either
centroids of healthy skin and psoriasis lesion. Healed skin
and other areas prone to misclassication are also corrected

segmented. For 28 out of 30 body regions under study, the

method is able to obtain segmentation accuracy higher than
90%. The high accuracy enables objective area assessment
of psoriasis lesions.
The proposed method forms one of the components for
PASI scoring that will allow dermatologists to monitor

436

M. H. Ahmad Fadzil et al.

psoriasis lesions using the PASI standard in practice. The

authors are currently also investigating methods for
determining other PASI parameters such as redness,
scaliness and thickness.
Acknowledgements

J Med Eng Technol Downloaded from informahealthcare.com by University of Texas at Austin on 05/26/14
For personal use only.

The research is a collaborative work between Universiti

Teknologi PETRONAS and Dermatology Department,
Hospital Kuala Lumpur. The authors gratefully acknowledge the assistance from Ahmad Tarmidzi Zakaria during
data collection.
Declaration of interest: The authors report no conicts of
interest. The authors alone are responsible for the content
and writing of the paper.

References
[1] Henseler, T., 1998, Genetics of psoriasis. Archives of Dermatological
Research, 290, 463476.
[2] Morrow, T., 2004, Evaluating new therapies for psoriasis. Managed
Care, 13, 3440.
[3] Pariser, D.M., 2003, Management of moderate to severe plaque
psoriasis with biologic therapy. Managed Care, 12, 3644.
[4] Feldman, S.R. and Krueger, G.G., 2005, Psoriasis assessment tools in
clinical trials. Annals of the Rheumatic Diseases, 64, 6568.
[5] Fredriksson, T. and Petterson, U., 1978, Severe psoriasisoral
therapy with a new retinoid. Dermatologica, 157, 238244.

[6] Gomez, D.D., Carstensen, J.M., Ersboll, B., Skov, L. and Bang, B.,
2003, Building an image-based system to automatically score psoriasis.
In Proceedings of the 13th Scandinavian Conference in Image Analysis,
557564.
[7] Roning, J., Jacques, R. and Kontinen, J., 1999, Area assessment of
psoriatic lesions based on variable thresholding and subimage
classication. In Proceedings of the Vision Interface99, 303311.
[8] Jailani, R., Hashim, H., Taib, M.N. and Sulaiman, S., 2004, Border
segmentation on digitized psoriasis skin lesion images. In Proceedings
of the 2004 IEEE Region 10 Conference, 3, 596599.
[9] Gomez, D.D., Ersboll, B. and Carstensen, J.M., 2004, Automatic
scoring of the severity of psoriasis scaling. In Proceedings of the Irish
Machine Vision and Image Processing Conference, 204209.
[10] Taur, J.S., 2003, Neuro-fuzzy approach to the segmentation of
psoriasis images. Journal of VLSI Signal Processing, 35, 1927.
[11] Xu, L., Jackowski, M., Goshtasby, A., Roseman, D., Bines, S.,
Yu, C., Dhawan, A. and Huntley, A., 1999, Segmentation of skin
cancer images. Image and Vision Computing, 17, 6574.
[12] Sharma, A., 2004, Understanding Color Management (New York:
Thomson Delmar Learning).
[13] Lee, H.-C., 2005, Introduction to Color Imaging Science (New York:
Cambridge University Press).
[14] Plataniotis, K.N. and Venetsanopoulos, A.N., 2000, Color Image
Processing and Applications (Berlin: Springer-Verlag).
[15] Ohno, Y., 2000, CIE Fundamentals for Color Measurements. In IS&T
NIP16 Conference, pp. 540545.
[16] Maletti, G., Ersboll, B. and Conradsen, K., 2005, A combined
alignment and registration scheme of lesions with psoriasis. Information Sciences, 175, 141159.
[17] Otsu, N., 1979, A threshold selection method from gray-level
histograms. IEEE Transactions on Systems, Man, and Cybernetics, 9,
6266.
[18] Fawcett, T., 2006, An introduction to ROC analysis. Pattern
Recognition Letters, 27, 861874.