AUTOMATIC SYSTEM FOR BLOOD TYPE CLASSIFICATION

USING IMAGE PROCESSING TECHNIQUES

Ana Ferraz, Vania Moreira, Diana Silva, Vítor Carvalho and Filomena O. Soares

Industrial Electronics Dept., University of Minho, Campus de Azurém, Azurém, Guimarães, Portugal

Keywords: Blood Types, Image Processing, IMAQ Vision.

Abstract: There is still not yet available a low-cost commercial equipment to determine blood types in an emergency

situation. This paper presents the development of a low cost system, based on image processing techniques,

that allows the automatic determination of human blood types in emergency situations. The experimental

method is based on the plate test where the serums specifics of blood types determination are mixed with

the sample blood of the donor. The mixtures blood/serums are captured through a CCD camera and

analyzed using the software IMAQ Vision from National Instruments. The developed image processing

methodology and the obtained results are detailed. The first prototype for automatic human blood

determination is presented.

1 INTRODUCTION

The determination of blood type can be performed

using various experimental tests (Datasheet of

Diamed, 2008) (Datasheet of Diamed-ID, 2008).

The plate test, used in this work, allows the

determination of blood type in a short time. It

consists of mixing the specific reagents of blood

type determination, with the patient blood. The

result depends on the occurrence or absence of

agglutination (Datasheet of Diamed, 2008). The

agglutination of erythrocytes is observed

macroscopically, in a short time, allowing using

image processing techniques to detect the

occurrence or absence of agglutination and therefore

the determination of the corresponding blood type.

In Figure 1, it is presented an image that shows the

difference between the occurrence and absence of

agglutination.

Currently, for the determination of blood types it

is required human intervention, not only in

performing the analytical procedures, as well as in

reading and interpreting the results, being then the

process more susceptible to errors (Alexander,

2007). With the aim to fulfil that gap and to

automate the determination of blood types, some

devices were developed (Alexander, 2007)

(Anthony, 2005) (Lambert, 2005). However, they

have high costs and present some limitations

compared to the method proposed in this work.

Figure 1: (a) Occurrence of agglutination. (b) Absence of

agglutination.

Preliminary studies performed by the research

team allowed the development of a software tool

based on image processing techniques, able to detect

the occurrence of agglutination. However, the

methodology was not fully automatic, requiring the

users to select the image area to quantify (Ferraz,

Carvalho and Brandão, 2008) (Ferraz, Carvalho and

Brandão, 2010). In this sense, this paper presents a

new system to automatically determine the blood

type. The methodology presented in this work is

innovative and at low-cost, being an added value to

commercial solutions.

)

368

Ferraz A., Moreira V., Silva D., Carvalho V. and O. Soares F..

AUTOMATIC SYSTEM FOR BLOOD TYPE CLASSIFICATION USING IMAGE PROCESSING TECHNIQUES.

DOI: 10.5220/0003159003680373

In Proceedings of the International Conference on Biomedical Electronics and Devices (BIODEVICES-2011), pages 368-373

ISBN: 978-989-8425-37-9

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

2 IMAGE ACQUISITION

PROCESS

As the reaction of agglutination is macroscopically

visible, the sample images were captured in real

size, using a CCD camera (Sony Cyber-shot DSC-

S750) with 7.2 megapixel resolution.

To analyze the acquired images, an image

processing application was developed using the

IMAQ Vision software from National Instruments

(IMAQ, 2004). Figure 2 shows the schematic of the

designed system.

Figure 2: Designed system schematic.

3 DEVELOPED SOFTWARE

The software application developed is presented in

this section, where it is detailed each image

processing technique employed. For each step (3.1

to 3.10), it is shown the effect that the applied

technique has in the former image, using IMAQ

Vision software.

3.1 Image Buffer: Add Copy (1)

Stores a copy of the original image with the four

samples (mixed blood/serum) in Buffer # 1 of the

image buffer for later use, Figure 3. Final results will

be overlaid on this image (IMAQ, 2004) (Klinger,

2003) (Relf, 2003).

As the original image will suffer a series of

changes, later there is a need for the original image,

this function allows saving the original image.

3.2 Color Plane Extraction: RGB

Green Plane

Allows extracting the green plane from an RGB

image, Figure 4 (

IMAQ, 2004) (Klinger, 2003) (Relf,

2003).

The original image is a RGB image that must be

processed to allow the determination of the

occurrence of agglutination.

Figure 3: Original image.

Figure 4: Image resultant from applying the Color Plane

Extraction: RGB Green Plane function.

3.3 Filters: Convolution Highlight

Details

The convolution filter highlights the regions in the

image where there are sharp changes in pixel values.

These regions correspond to the boundaries of the

samples and other noisy pixels that may be present

in the image. The convolution kernel highlights the

edges of an image and in this case, the function uses

a 3 x 3 kernel (IMAQ, 2004) (Klinger, 2003) (Relf,

2003).

The next function is a threshold function that is

used to separate certain structures of the image. In

this case, it is used to separate the samples

blood/serum of the background, once this function

segments the image in two regions, designated

region “particle” and “region background” (IMAQ,

2004) (Klinger, 2003) (Relf, 2003).

3.4 Threshold

This function applies a threshold to the image

resulting of previous function based on the

Minimum and Maximum threshold values

introduced. All pixels that are not contained between

the Minimum and Maximum values are set to 0 and

all pixels that fall inside the range are replaced by 1

AUTOMATIC SYSTEM FOR BLOOD TYPE CLASSIFICATION USING IMAGE PROCESSING TECHNIQUES

369

(IMAQ, 2004) (Klinger, 2003) (Relf, 2003). In this

function the Minimum value is 128 and the

Maximum value is 255.

The Minimum and Maximum threshold values

were determined by trial and error, when developing

the algorithm, and were kept constant afterwards.

The result of applying this function is presented in

Figure 5.

Figure 5: Image resultant from applying the Manual

Threshold.

This function is then combined with the Local

Threshold: Niblack function,, allowing isolating the

particles corresponding to the mixed blood and

serum.

3.5 Local Threshold: Niblack

Calculates a threshold value for each pixel based on

the statistics of the surrounding pixels. This

algorithm compensates the high lighting variations.

This function uses a kernel; in this case the kernel

size is 115 width and 132 height (

IMAQ, 2004)

(

Klinger, 2003) (Relf, 2003).

Using the previous feature to define the borders

in the image to isolate particles, results in the image

shown in Figure 6.

Figure 6: Image obtained by applying the Local

Threshold: Niblack function to Figure 5.

3.6 Adv. Morphology: Fill Holes

Fills all the holes that are present in the particles.

Holes are filled with a pixel value of 1. The resulting

binary image contains entire particles, without holes,

corresponding to the samples blood/serum (IMAQ,

2004) (Klinger, 2003) (Relf, 2003).

3.7 Adv. Morphology: Remove Small

Objects

This function removes the small particles and the

possible noise in the binary image resulting from the

previous function. It eliminates particles that are not

relevant to the analysis. The particles that are

removed by an iteration of erosion are assumed to be

noisy particles (IMAQ, 2004) (Klinger, 2003) (Relf,

2003).

This function is used to eliminate the small

particles that can interfere in the analysis of the

image. Small drops of blood or serum in the

background of the image are not relevant to the

analysis, and should be therefore removed.

3.8 Adv. Morphology: Remove Border

Objects

It eliminates particles that are at the border of the

image. It removes particles that are not needed for

the analysis of the image, preventing interference

from unwanted particles (IMAQ, 2004) (Klinger,

2003) (Relf, 2003).

This step is necessary to eliminate particles that

are joined together due to high kernel placed in the

Local Threshold: Niblack function, but in fact

should be separated.

3.9 Particle Analysis

This step is particles corresponding to each mixture

blood/serum. These necessary to analyze the

properties of the particles in the image, considered

as four particles can be analyzed using various

properties, being the determination of the center of

mass the most important in this work (IMAQ, 2004)

(Klinger, 2003) (Relf, 2003). This function is

essential because it defines a coordinate system and

the region to analyze.

The result of the previous function is a table that

contains the properties selected and their values. The

values of Center of Mass X and the Center of Mass

Y will be used in the following function.

BIODEVICES 2011 - International Conference on Biomedical Electronics and Devices

370

3.10 Threshold

As described in 3.4.The result of this function is

presented in Figure 7.

Figure 7: Image obtained by applying the Manual

Threshold function.

3.11 Image Buffer: Retrieve Buffer # 1

It retrieves the copy of the original color image, so

that it can be used by the next function. The original

image has the four samples blood/serum, Figure 8

(IMAQ, 2004) (Klinger, 2003) (Relf, 2003).

Figure 8: Image obtained by applying the Image Buffer:

Retrieve Buffer function in the image of Figure 7.

3.12 Color Operators: Not Or

This step performs a logical OR operation between

the original image input image and the original

image stored in the buffer. This is a bit-wise

operation (IMAQ, 2004) (Klinger, 2003) (Relf,

2003). The result of this function is presented in the

image of Figure 9.

3.13 Color Plane Extraction: HSL

Luminance Plane

This function is used to extract the luminance plane

from the color image obtained with the previous

function (IMAQ, 2004) (Klinger, 2003) (Relf,

2003), Figure 10.

Figure 9: Image obtained by applying the Color Operators:

Not Or function in the image of Figure 8.

Figure 10: Image obtained by applying the: HSL

Luminance Plane function in the image of Figure 9.

3.14 Set Coordinate System

This function defines a coordinate system based on

the stage of particle analysis. The particle analysis

function gives the coordinates necessary to calculate

the center of mass, used in this function. Chosen the

mode horizontal and vertical motion because it

allows adjusting the region of interest positions

along the horizontal and vertical axes (IMAQ, 2004)

(Klinger, 2003) (Relf, 2003).

The definition of the region of interest is an

important task in this method. Based on the

coordinate system, it is selected the region to be

analyzed, depending of the center of mass of the

particle calculated through the particle analysis

function.

This function will be repeated for each of the

particles in analysis. In this case, it will be repeated

four times, one for each blood/serum sample.

3.15 Quantify

It measures the intensity of the pixels in the region

of interest selected, Figure 11. This step uses the

Reposition Region of Interest that when enabled, it

dynamically repositions the region of interest based

on the coordinate system previously defined. Also, it

AUTOMATIC SYSTEM FOR BLOOD TYPE CLASSIFICATION USING IMAGE PROCESSING TECHNIQUES

371

uses the Reference Coordinate System that indicates

the coordinate system to link the region of interest

(IMAQ, 2004) (Klinger, 2003) (Relf, 2003).

The result of the application of this function

consists of a table that contains the area (percentage

of the analyzed surface in relation to the complete

image), the mean value (mean value of the pixels),

the standard deviation (standard deviation of the

pixels) and the minimum and maximum values of

the pixels (IMAQ, 2004) (Klinger, 2003) (Relf,

2003).

This function allows identifying the occurrence

of agglutination in a sample blood/serum based on

the standard deviation value of the pixels.

As in the previous function, this step is repeated

for each of the particles in analysis.

Figure 11: Result of applying the Quantify function.

4 SYSTEM PROTOTYPE

A prototype system that automatically determines

the human blood type, based on the plate test

procedure, was designed (Figure 12).

Figure 12: Prototype developed.

The blood and the four serums drops are

manually placed in the plates inserted in the mobile

drawer, actuated by a DC motor. In the first blade, it

is placed reagent anti-A, in the second reagent anti-

B, in the third anti-AB reagent and finally in the

fourth, reagent anti-D, in accordance to the testing

procedure previously described.

The system is switched on and the drawer moves

to the mixing area, where the mixture blood/serum is

promoted in each blade. It must be referred that

there is no contamination between the four samples.

Next, the drawer moves to the image capture zone.

A step motor moves a Glossy 5 Mega pixels webcam

along the samples for capturing the four images. The

images are saved for future analysis. The system is

controlled by Arduíno microcontroller

(http://www.arduino.cc/).

5 RESULTS

The proposed methodology was tested with several

standard blood types samples. In this section, are

presented the results obtained when applying the

image processing methodology to four blood/serum

samples of a donor blood type, Figure 13.

Figure 13: Blood/serum samples. (a) Serum Anti-A. (b)

Serum Anti-B. (c) Serum Anti-AB. (d) Serum Anti-D.

Figure 14 shows the final images obtained with

the application of image processing techniques to the

original sample images of Figure 13. The

corresponding quantification is presented in Table 1.

Figure 14: Image resulting from application of the image

processing techniques developed, to images of Figure 13.

(

a) b)

BIODEVICES 2011 - International Conference on Biomedical Electronics and Devices

372

Table 1: Results of quantification of images of Figure 14.

Fig. Area Mean

Value

Standard

Variation

Min

Value

Max

Value

(a) 0,9 162,1 7,7 137,0 181,0

(b) 0,8 56,9 45,1 5,0 201,0

(d) 0,8 146,0 8,8 109,0 173,0

Analyzing Figure 14, it is observed that the

agglutination occurred in images (b) and (c), but not

in images (a) and (d). By correlating this information

with the information from Table 1, it is observed

that the standard deviation, in the images (b) and (c)

is well above 16, while in the images (a) and (d),

the standard deviation is less than 16. The value 16

for the standard deviation is a limit established for

determining the occurrence of agglutination in a

sample. This value was established from trial and

error. Thus, it is observed that when agglutination

occurs, the standard deviation is much higher than

the one obtained when agglutination does not occur,

allowing thus identifying the occurrence of

agglutination and consequently identifying the blood

type of a patient. In this example, given that the

agglutination has occurred in the presence of serum

anti-B (Figure 14-b) and in the presence of serum

anti-AB (Figure 14-d), the blood type presented is B

negative. Note that the agglutination occurs in the

presence of serum anti-AB, because the patient had

B antigens in their red blood cells that agglutinated

in the presence of anti-B antibodies existing in

serum. However, the serum anti-AB, also had anti-A

antibodies, that have not reacted because the patient

did not have A antigens, justifying the slightly less

value of agglutination (42.7), compared to that

obtained with serum anti-B (45.1).

4 CONCLUSIONS AND FUTURE

WORK

With the proposed system, based on image

processing techniques, it is possible to automatically

determine the blood type of a patient, by detecting

the occurrence of blood agglutination. This approach

allows the determination of blood type of a patient,

safely, and it can be used in emergency situations as

the results are obtained within a short time (2

minutes). The PC hardware requisites of the

prototype are minimal and IMAQ software package

allows the correct and fast determination of blood

types.

In future, we intend to optimize the prototype,

reducing human intervention in the procedures.

Another objective is to ensure that the developed

device is portable, allowing its use near the patient,

avoiding travel to the lab that only cause more time

consuming.

REFERENCES

Datasheet of Diamed AG,1785 Cressiers/Morat, Cressier,

2008.

Datasheet of Diamed-ID Micro Typing System, Card-ID.

Diaclon ABO\Rh for patients. Cressier, 2008.

Alexander, S. P., 2007. An Integrated Microoptical

Microfluidic Device for Agglutination Detection and

Blood Typing [Master Thesis]. California: University

of North of Carolina.

Anthony, S. R., 2005. A Simplified Visible/Near-Infrared

Spectrophotometric Approach to Blood Typing for

Automated Transfusion Safety [Master´s thesis].

California: University of North Carolina.

Lambert, J. B., 2005. A Miniaturized Device for Blood

Typing Using a Simplified Spectrophotometric

Approach [Master´s thesis]. California: University of

North Carolina.

Ferraz, A., Carvalho, V., Brandão, P., 2008. Determinação

Automática do Tipo Sanguíneo de Humanos

Utilizando Técnicas de Processamento de Imagem,

CBIS 2008. Bazil: São Paulo.

Ferraz, A., Carvalho, V., Brandão, P., 2010. Automatic

Determination of Human Blood Types using Image

Processing Techniques, BIODEVICES 2010. Spain:

Valencia.

IMAQ, 2004. IMAQ Vision Concepts Manual, National

Instruments, Austin.

Klinger, T., 2003. Image Processing with LabVIEW and

IMAQ Vision, Prentice Hall. New Jersey.

Relf, C. G., 2003. Image Acquisition and Processing with

LabVIEW, CRC Press.

AUTOMATIC SYSTEM FOR BLOOD TYPE CLASSIFICATION USING IMAGE PROCESSING TECHNIQUES

373