Professional Documents
Culture Documents
preliminary
Blood type in the medical world can be divided into 4 groups, namely A, B, AB and O. So far, human
blood group detection is still done manually by a laboratory assistant, by taking two drops of blood to be
identified. The blood will be placed on a blood group paper and divided into 2 parts. Each part of the blood will
be dripped with antisera A and antisera B fluids to observe the agglutination process.
Observation of the agglutination process is carried out by the laboratory assistant through eye
observation so that the error rate is still very possible due to the fatigue or saturation of the laboratory
assistant's eye, which makes the accuracy of the data depending on the eye of the laboratory assistant.
Therefore, the researcher had the idea to apply the blood group identification process using image processing.
There are several studies that discuss this background. In the first study, there is a system that
determines blood type using image processing using the robert, prewitt and sobel method, with the success
percentage of the robert operator 38.89%, the prewitt operator 83.33%, and the single operator 94% [1]. In
the second research, a blood group detection system using digital images using fuzzy logic, with a success rate
of 83% [2]. In the third study, there is a classification system for blood type using Fuzzy C-Mean Clustering
(FCM) and Learning Verctor Quantization (LVQ), with success rates of 92% to 98% [3].
From some background and research that has been done previously, this final project will develop a
blood type detection system with the ABO system using image histograms and Artificial Neural Networks
(ANN).
Classification of blood groups using image histograms with Artificial Neural Networks (ANN) is an
attempt to determine blood type by using image processing based on the minimum and maximum values of
three color channels, namely RGB and its pixel location from the image histogram which will then be classified
using Artificial Neural Networks (ANN).
2. Literature review
2.1 Image processing
Image processing is an image processing, especially by using a computer, into an image of a better
quality [5]. Mathematically, image processing can be defined as two-dimensional processing of an image by a
computer that is, an image is defined as a function of two real variables, such as t (x, y) with an amplitude such
as the brightness of the image at coordinate points (a, b) [6].
Digital images are generally two-dimensional (2D) expressed in the form of a matrix with a finite
number of elements. Each element of the image matrix has a certain x and y coordinate position and of course
has a value. In general, a digital image is a representation of pixels in 2D space expressed in a matrix of size N
rows and M columns as in Figure 1, each element of the image matrix is called a pixel (picture element, image
element or mop). The value of each pixel f at the x and y coordinate positions represents the color intensity
and can be encoded in 24 bits for color images (with three color components, namely RGB: R =red, G = green, B
= blue) 8 bits for the gray image orgray-level or 1 bit for binary image.
3. Research methods
Broadly speaking, the block diagram for blood group classification uses an image histogram
withartificial neural networks (ANN) can be seen in Figure 2, there are 3 parts in the block diagram there are
parts input, matlab, and the output. InputThis system is in the form of an image of a blood group card, which
previously contained a blood sample mixed with antisera. Then for the matlab used consists of 4 parts as data
processing, namely segmentation,cropping, histogram, and ANN. Forthe outputit is the type of blood group
that has been classified by ANN withinput neurons are the minimum and maximum values of RGB and pixel
location of the blood sample image.
The block diagram above describes the sequence of detecting blood groups, input system in the form
of a blood group card image, then the segmentation process is carried out with the aim of removing the color
background so that the only color available is the color of the blood sample. After the segmentation process,
the process is carried out croppingor cutting with the aim of dividing the image into 2 parts between sample A
and sample B. The cut image is the original image not the segmentation result. After this process, a histogram
process is carried out on the image to determine the RGB value of the image. The RGB value that is looking for
is the highest and lowest value along with the location of the values of the three color kernels. If the RGB value
and pixel location of the two images are known, this data will be used as ANN input to classify the type of
blood group from the values obtained.
The blood and antisera will then be drawn so that it can be used as input for the system in the form of
images.
3.2 Segmentation Design
Segmentation is used to separate the background color from blood samples on blood group paper in
order to facilitate analysis. The color that is removed in this case is green, so that the blood sample color and
the background are black.
Segmentation in this process is segmentation based on color similarity, which means the process of
forming an area where a number of pixels are connected to each other because they have the same color or
visually have similar colors. This segmentation process will form segments according to the characteristics of
each color. The segmentation process in this system uses HSV elements. The HSV color space refers to hue,
saturation, and value.
3.3 Cropping Design
Croppingor image cutting is a process of cutting an image which has the aim of separating images that
have certain characteristics from the original image. Image cropping is used to separate blood samples mixed
with antisera A and antisera B, making it easier for further processing. So that after the cutting process is
carried out, there will be two blood samples, namely a blood sample with antisera A and a blood sample with
antisera B.
3.4 Image Histogram Design
The image histogram is used to get the highest and lowest RGB values with the pixel location of the
blood sample. That later the clotting blood sample will have a range of color values, as well as the non-clotting
blood sample, this intensity range value will later be used as neuron input for the ANN training process. In
addition, this histogram process will get the pixel location of the lowest and highest values of each color
kernel.
From this histogram process, the parameters will be used as input (input) for ANN. These parameters
include: Max R, Location Max R, Max G, Location Max G, Max B, Location Max B, Max R1, Location Max R1,
Max G1, Location Max G1, Max B1, Location Max B1, Min R, Location Min R, Min G, Location Min G, Min B,
Location Min B, Min R1, Location Min R1, Min G1, Location Min G1, Min B1, Location Min B1.
3.5 Artificial Neural Networks Design
In this section, ANN training will be conducted to form the characteristics of artificial neural networks.
This is done by pairing the input and target target (supervised learning) using 7 RGB values from the sample of
each image. The following is a flowchart for ANN training data:
As for the overall flowchart in the blood group classification process using an image histogram with
ANN is as follows:
From the flowchart above, it can be seen that the whole system is divided into 3 parts, namely
preprocessing, training and identification. In the preprocessing stage, the reagent image will be processed
through several processes including segmentation, cropping, and histogram.
At the training stage, ANN training is carried out with input in the form of a minimum and maximum
value of 3 color kernels namely Red, Green and Blue, along with the location of the pixels, after which the
weight value adjustment process is carried out.
The third stage is identification, namely the process of introducing incoming data to training data, by
matching the weight values. At this stage the output of the blood group classification that has been carried out
by ANN will be determined.
The ANN algorithm used in this system is backpropagation. The backpropagation model is a guided
training method. This network consists of one input layer with 24 neuron units, one hidden layer, and one
output layer with 2 output units. The activation function used in the hidden layer is tansig, while the activation
function in the output layer is also tansig.
From the table above, it can be analyzed that detection failure during testing can occur due to two
things, firstly because the image processing process is not optimal, the cutting process is not good and the
intensity level of the image color varies so that it affects the characteristic extraction process. Second, in the
ANN, more data needs to be added at the timetraining to improve its accuracy.
From table 2, it can be seen that the percentage of conformity between manual observation and the
program system is obtained 2 times the mismatch of identification from 12 trials. The percentage of errors is
obtained as follows:
jumlah error
Average error = x 100%
jumlah seluruh data uji
2
= x100%
12
= 16.67%
5. Conclusion
From the results of the research and analysis of the system testing that has been made, several
conclusions can be drawn, namely: the application of an image histogram with ANN for blood type detection
has been successfully carried out and the error percentage is 16.67%. In taking blood sample images, the
lighting factor greatly affects the RGB value in the image.
From the results of system testing, there are several shortcomings. We recommend that when taking
blood sample images using a scanner so that the results obtained are the same. In addition, it is also necessary
to add several other image operators besides the histogram to add several parameters that can be used to
improve system accuracy. Further research should be inclined toprototype and use a reading system real time.