JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

JAUNDICE DETECTION SYSTEM USING MACHINE

LEARNING

Rakshitha S 1, Priyanka T 2, Ramya P3, Shijitha R*

U G Student 1,2,3, Assistant professor*
Department of Biomedical Instrumentation Engineering
Avinashilingam Institute for Home Science and Higher Education for Women,
Coimbatore, India

Abstract: When sturdy jaundice to bestowed, babies or adults should be subject to

clinical communication like “serum bilirubin,” which might cause trauma in patients.
Typically, jaundice is presented in diseases like infectious disease or liver disease. We
propose discovering jaundice (icterus) in new-born or adults by employing a not pain
technique to avoid further traumas. By effort digital pictures in color, palm, soles, and
forehead, we tend to analyze RGB attributes and diffuse reflection factor spectra
because the parameter to characterize patients with either jaundice or not and
correlate that parameter with the extent of haematoidin. By applying a support vector
machine, we tend to distinguish between healthy and sick patients.
Keywords: Digital Image Processing, Diffuse Reflectance, Bilirubin, Support
Vector Machine

1. INTRODUCTION

Newborn jaundice is a yellowing of a baby’s skin and eyes. Newborn jaundice is very
common and can occur when babies have a high level of bilirubin, a yellow pigment
produced during normal breakdown of red blood cells. The yellowing may begin within
2 to 4 days after birth and may start in the face before spreading down across the body.
Penhaker created an associate instrument that supported transmitted light-weight through
the skin to see the amount of animal pigment in newborns or adults. each invasive and
non-invasive strategy is developed to see the amount of animal pigment in newborns or
adults.450 nm - 575 nm monochromatic light-weight is employed for skin image
diagnostic[1]. Buttitta associated colleagues produce a non-invasive animal pigment
monitor that employs 2 wavelengths of sunshine aimed toward an infant's blood vessel
system. The reflectivity or disperse {of lightweight sunshine} from light sources into the
infant's blood is determined and analyzed to see the animal pigment level of the infant's
blood vessel system. beginning with the diffuse reflectivity spectrum[2,3], Alla Suresh et
al establish a way to separate intravascular and extravascular animal pigment. The
optical properties of the skin reflectivity spectrum, as well as animal pigment
concentration, were extracted employing a nonlinear improvement formula. A support
vector machine (SVM) may be a variable information classification tool. the likelihood

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:149

JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

of developing some diagnostic algorithms victimization SVM is additionally attracting

interest. though golf player et al. used a linear SVM classifier to classify fluorescence
and diffuse reflectivity spectra of breast tissues in vitro, Lin et al. used each a linear and
nonlinear SVM classifier with RBF kernel to classify fluorescence spectra from tissues
in vivo. The tissue spectra were dimensionally reduced by victimization linear PCA
algorithms before victimization the SVM methodology for classification in each groups'
papers. Lin et al. incontestable that the classification performance of an associate SVM
classifier trained on the complete spectral information was adored that of a classifier
trained solely on the diagnostically important principal elements. The procedure quality
of their combined PCA- SVM approach was explicit to be reduced. during this paper, we
tend to propose a unique idea within which we tend to extract RGB attributes from a
picture and use a prism spectroscope to extract diffuse coefficient spectra from patients
with low and high haematoidin levels, then compare the values with a BILITEST®. we
tend to were ready to produce Associate in Nursing rule to differentiate between low and
high levels of haematoidin for any patient by employing a support vector machine to
correlate. the values with a BILITEST®. Several researches have been reported over past
few years about non-invasive bilirubin detection and in the area of using smartphones to
diagnose various medical conditions.

2. HYPERBILIRUBINEMIA

Bile is a greenish liquid fluid formed by the liver, which is normally white and deposited
in the vesica during inter-digestive intervals before being emptied to the small intestine
when meals are digested. When the liver is unable to break down the pathogenic red
blood cells, the number of hematoid cells in the body climbs to dangerous levels, causing
brain harm and, in rare circumstances, death.. [4,5,6]. There are two types of hyper
bilirubinemia 1.conjucated 2.unconjucated.Figure 1 shows Jaundice affected baby image
Yellowish discoloration of the skin, sclerae, and mucous membranes due to
accumulation of bilirubin pigment.

Figure 1. Jaundice affected baby image.

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:150

JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

3. DIGITALIMAGE PROCCESING
The most ordinarily used color house in technology is that the RGB color house, which
relies on the additive mixture of the three primary colors R, G, B [7, 8]. The terms red,
green, and blue area unit a standardization to supply descriptions for the first colors. the
first colors area unit used as a reference colors within the majority of the imaging
sensors. They kind the base vectors of a three-dimensional orthogonal color-vector
house, wherever the zero-vector represents black. The origin is additionally represented
as black purpose. On the opposite hand, any color is often viewed as a linear
combination of those base vectors within the RGB house. In one such accepted RGB
color house, a color image is mathematically treated as a vector perform with 3 parts.
Specifically, for a picture
𝐼(𝑥,𝑦)=(𝑅(𝑥,𝑦),𝐺(𝑥,𝑦),𝐵(𝑥,𝑦))𝑇
These values are referred to a tristimulus values.

4. METHODOLOGY
We propose a non-invasive bilirubin detection technique for jaundice prediction based on
yellow discoloration of skin in which jaundice can be detected directly from the image of
skin. Since the current gold standard to detect bilirubin levels is total serum bilirubin
(TSB) determination by invasive blood sampling which is stressful and painful for the
neonates thus a non-invasive technique is needed which is cheaper than any of the
available Transcutaneous Bilirubinometers. By acquiring digital images in color, in palm
and forehead, we analyze RGB attributes as the parameter to characterize patients with
either jaundice or not, and we correlate that parameter with the level of bilirubin. By
applying support vector machine, we distinguish between healthy and sick patients.
Figure 2 shows the workflow of the proposed method.

Figure 2: workflow

We have collected some sample images of babies having Jaundice and applied some basic
image processing techniques like color balancing, Image segmentation, Feature Extraction
to estimate bilirubin and result was very promising to predict jaundice. Every color is
composed of three basic colors: Red, Green and Blue also known as RGB. If each of these

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:151

JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

components are presented in eight bits then each value will have a range from 0 to 255.
For white light the value of R, G and B component is 255 while for black light all these
values are 0. Now, we are concerned with the yellow color for which the R and G
component is 255, but B component is 0.
4.1 Color Balancing
Color Balancing involves the computation of mean values of red, green and blue pixels of
an image to overcome the effects of different conditions of light. A number of color
transformations are employed to approximate the characteristics such as conversion to
Hue, Saturation and Value
4.2 Image Segmentation
Image Segmentation is used to segment a small portion of image in order to decrease the
search area so the algorithm ignores the pixels outside the segmented area. This segmented
part is then compared with the standard bilirubin level color chart. The algorithm applies
threshold to find the best possible match of segmented image from the color chart.
4.3 Feature Extraction
Feature Extraction includes colormap transformations and feature calculation. In order to
detect the discoloration in a better way, algorithm transforms the original RGB values into
Lab color spaces. We calculate the mean value for each color channel thus giving 6
features. These features will be used to determine the bilirubin levels.

5. RESULTS AND DISCUSSION

We analyzed the baby's skin colors, and the result was encouraging. The jaundice-
affected baby has high red and green component values comparing with the not-affected
baby. But specific issues have to be brought into consideration. The ambient lighting is a
critical factor for this. As we are heavily relying on image processing, ambient lighting
can be a significant factor. The skin color of the baby can also play a vital role. The table
1 shows some sampled images and their RGB values and the result.

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:152

JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

Table 1: Sample images and RGB Values

Affected value
Baby images R Value G value B Value
and result
12000
Affected by
251 249 0 jaundice

8000
Not affected by
100 139 1
jaundice

15000
Affected by
245 231 0
jaundice

7000
Not affected by
120 140 1 jaundice

14000
Affected by
250 237 0 jaundice

The result is displayed in figure 3. We can see that R and G values are pretty close. This is
just a sample case. We will need a lot more than just a sample to develop the system

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:153

JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

Figure 3:RGB Values

6. CONCLUSION
In this paper, a non-invasive bilirubin detection technique for jaundice prediction based
on yellow discoloration of skin is developed. The input image is given and SVM
classification is performed as a change in color of skin. The main symptom of jaundice
we had taken the color texture and our algorithm collects the feedback and uses the baby
face to predict whether it is affected by jaundice or not. The image of the patient is
captured and the SVM performs image processing algorithm and the result is displayed
through web application using cloud data transfer.

REFERENCES

[1] M, Penhaker, V. Kasik, andB.Hrvolova,"Advanced Bilirubin measurement photometric

method”, Electronica electro technical, vol.19,3,2013.
[2] Anthony Buttitta and Steven Falk, "Noninvasive bilirubin monitor", European Patent
Application,EP0747002A1,11.12.1996.
[3] Clinical system model for monitoring the physiological status of jaudience by extracting
bilirubin components from skin diffuse reflectance spectra, Advanced Biomedical and
Clinical Diagnostic Systems VII, edited by Anita Mahadevan-Jansen, Tuan Vo-Dinh,
Warren S. Grund fest, Proc. of SPIEVol.7169,2009.
[4] G. M. Palmer, C. Zhu, T. M. Breslin, F. Xu, K. W. Gilchrist, and N. Ramanujam,
“Comparison of multi excitation fluorescence and diffuse reflectance spectroscopy for
the diagnosis of breast cancer” IEEE Trans. Biomed. Eng.50, 1233-1242(2003).
[5] W.Lin,XYuan,PYuen,WIWei,JSham.PCShi.“Classificationofinvivoautofluorescencespec
trausingsupportvectormachines”. JBiomed.Opt.9,180-186(2004).
[6] LeeGoldman,AndrewSchafer,Goldman`sCecilMedicine,SectionXIII,Diseasesoftheliver,
GallbladderandBile Ducts,SaundersofElsevierInc,2000.
[7] González,R.C.,Wintz,P.(1996).Procesamientodigitaldeimágenes.Addison-Wesley.
[8] J.J.BaezRojas,M.Guerrero,J.CondeAcevedo,A.PadillavivancoandG,UrcidSerrano,Segm
entacionde image nesde color,Rev.Mex.Fis,50,6,pp. 579,2004.
[9] V.Vapnik, of Statistical Learning Theory,Chap.4,Springer-Verlag,NewYork(1995,2000).

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:154

JOURNAL OF ENGINEERING, COMPUTING & ARCHITECTURE ISSN NO:1934-7197

[10] A Practical Approach to Neonatal Jaundice, Sarah k. Moerschel, Lauren b.

Cianciaruso, and Lloyd r. Tracy,Am Fam Physician.2008May1,77(9),1262.

Volume 11, Issue 6, JUNE - 2021 http://www.journaleca.com/ Page No:155