Estimation of Texture Variation in Malaria Diagnosis

  • A. Vijayalakshmi
  • B. Rajesh Kanna
  • Shanthi Banukumar
Conference paper
Part of the Lecture Notes in Electrical Engineering book series (LNEE, volume 490)

Abstract

Malaria parasite has been visually inspected from the Giemsa-stained blood smear image using light microscope. The trained technicians are needed to screen the malaria from the microscope; this manual inspection requires more time. To reduce the problems in manual inspection, nowadays pathologist moves to the digital image visual inspection. The computer-aided microscopic image examination will improve the consistency in detection, and even a semiskilled laboratory technician can be employed for diagnosis. Most of the computer-aided malaria parasite detection consists of four stages namely preprocessing of blood smear images, segmentation of infected erythrocyte, extracting the features, detection of parasite and classification of the parasites. Feature extraction is one of the vital stages to detect and classify the parasite. To carry out feature extraction, geometric, color, and texture-based features are extracted for identifying the infected erythrocyte. Among these clause of features, texture might be considered as a very fine feature, and it provides the characteristics of smoothness over the region of interest using the spatial distribution of intensity. The proposed work demonstrates the merit of the texture feature in digital pathology which is prone to vary with respect to change in image brightness. In microscope, brightness of the image could be altered by iris aperture diameter and illumination intensity control knob. However, the existing literature failed to mention the details about these illumination controlling parameters. So the obtained texture feature may not be considered as distinct feature. In this paper, we conducted an experiment to bring out the deviation of texture feature values by changing the brightness of the acquired image by varying the intensity control knob.

Keywords

Malaria diagnosis Image brightness Gray-level co-occurrence matrix Digital pathology Digital microscopy 

1 Introduction and Background

Malaria is an infectious disease caused by Plasmodium parasite. There are four types of parasites namely Plasmodium vivax, Plasmodium falciparum, Plasmodium ovale, and Plasmodium malariae which cause malaria disease [1]. These parasites invade human red blood cells, it passes through four stages of development life cycle, and the stages are ring, trophozoite, schizont, and gametocytes. Symptoms of the malaria disease can be identified from clinical examination followed by the laboratory test to confirm the malaria affected victim. There are several laboratory methods are available like microscopic examination of stained thin or thick blood smear, quantitative buffy coat (QBC) test, rapid diagnosis test, and molecular diagnosis methods [1]. WHO recommends microscopic examination of thin Giemsa-stained blood smear as the gold standard for diagnosing malaria parasite [1]. However, microscopic examination is a quite time-consuming task, laborious process, and trained technicians are being employed to accurately detect the malaria parasite. To supplement this cognitive task, research fraternity trusted digital image processing approaches for efficient malaria diagnosis. Almost, all the image-based malaria diagnoses follows four functional pipelines namely image acquisition, image preprocessing, erythrocyte segmentation, feature extraction and classification [2] as illustrated in Fig. 1.
Fig. 1

Steps for malaria parasite detection and classification

1.1 Image Acquisition

It is the process of collecting malaria blood smear images from the digital camera mounted on the microscope. The stained smears are observed from microscope with the magnification of 100× objective lens. For computer-aided malaria parasite detection, thick and thin blood smear images are used. Images obtained from thin blood smear are used to identify the type of malaria and its severity stages [3], whereas thick blood smear is used for quantifying the infected erythrocytes [4]. Though several dyes have been used to stain erythrocyte, popularly used dyes are Giemsa and Leishman [5].

1.2 Preprocessing of Blood Smear Images

It is the process of removing the unnecessary details present in the acquired malaria parasite image for better visualization and for further analysis. In the existing literature, image filters like median [5], geometric mean [6], Gaussian [2], Laplacian [7], wiener [8], low pass [9], SUSAN [10] were used to eliminate the noise. And to enhance the contrast adaptive or local histogram equalization [11], dark stretching [12], partial contrast stretching algorithm, and histogram matching [13] are used. Most of the literature suggested Gray World Assumption could be the ideal choice, because it is used to eliminate the variation in the brightness of the image [14].

1.3 Segmentation of Infected Erythrocyte

It is a process of isolating the red blood cell (erythrocyte) and eliminating the other details such as white blood cell, platelet, and other artifacts from the preprocessed image. The isolated segment may include infected and non-infected malaria parasites. Further, to discriminate the infected and non-infected erythrocytes, finer segmentation techniques have been used. From the previous studies, we listed the various segmentation techniques utilized for isolating erythrocytes. They are circle Hough transform [15], Otsu threshold [8], pulse-coupled neural network [16], rule-based algorithm [17], Chan-Vese algorithm [11], granulometry [10], edge detection algorithm [18], Marker-controlled watershed algorithm, and normalized cut algorithm [4, 5]. In most of the literature, watershed transformation algorithm was used to segment the erythrocytes. Similarly to discriminate the infected erythrocytes, few techniques are used such as histogram threshold, moving k-means clustering, fuzzy rule-based system, annular ring ratio method, Zack threshold, and N-cut algorithm.

1.4 Extracting the Features

It is the important step to gather the insights from the erythrocytes. And it is a set of process for deriving size, color, stippling, surface, and pigment features of the infected erythrocyte for subsequent human interpretation to detect the infected malaria parasite. From the segmented erythrocytes, we can extract the coarse-level features like size and color. The size of infected region is being inferred from the efficient image-based models and its area estimation technique [19, 20]. Area and color are first-level eliminating features to differentiate the species in the malaria parasite. It is a vital feature to differentiate the species in the malaria parasite. The existing literature refers some morphological features, which are retrieved using Hu moment [4], Chain code, Bending energy, Roundness ratio [21], Relative shape [22], Shape features [5, 6, 10, 14] and Area granulometry [23]. From the previous literature, the features used to extract the intensity of the malaria images are histogram, color histogram, entropy, color channel intensity, and color autocorrelogram. Among all, texture is a finer feature used by most of research fraternity to discriminate infected and non-infected malaria. It provides the spatial distribution of intensity over the region of interest and gives the quantitative evaluation of the images. Some texture features like gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level size-zone matrix (GLSZM), local binary pattern, flat texture, gradient texture, Laplacian texture, and wavelet feature [6] has been used in previous studies.

1.5 Detection and Classification of the Parasites

It is the process of extracting the information about infected parasite and categorizing its types. The various classification algorithms used in the literature are Bayesian classifier [24], feedforward backpropagation neural network [4, 10, 21], classification and regression tree [5], K-nearest neighbor classifier [9, 14, 16], logistic regression [5], AdaBoost algorithm [25], Naïve Bayes tree [26], support vector machine [3, 6, 7, 9, 17], and multilayer perception network [5, 9].

2 Motivation and Challenges

In microscope, image brightness variation occurs due to the change in the magnification, iris aperture diaphragm, and illumination intensity control knob. As per the WHO standard, the magnification of the malaria blood smear is fixed to 100× objective lens. Hence, the variation of the brightness can occur only because of the manual adjustments of the iris aperture diaphragm and illumination intensity control knob. Moreover, theory of the light microscope defines that brightness depends on the amount of light passing through the condenser controlled via the iris aperture. For low magnification factor (eg.10x), iris aperture is tuned to higher diameter, inversely for the higher magnification, iris aperture is small. It gathers optimum light so we will get bright images. But tuning the illumination intensity controller only makes the variation of brightness in the image. We tried to observe the impact of this brightness variation on the global texture feature.

Image texture provides information about the spatial arrangement of intensities of an image. These texture features are used in the digital pathology to detect or classify the malaria parasites. The characteristics of the texture features behave differently for infected and non-infected erythrocyte regions. So the texture descriptor plays an important role in identifying the malaria parasite. However, when extracting the infected region texture properties, the nearest region property also influences the texture feature [27]. These artifacts influence the definite variation in the texture feature metric. We observed that especially in image-based malaria diagnosis, the previous computer-aided microscopic researches ignored to incorporate this variation in their formulations. From this proposed work, we tried to bring out the existence of non-uniformity in texture descriptor.

3 Experiment

In this session, we describe the experiment which has been conducted by the authors. The purpose of this experiment is to identify the variation in texture values of malaria blood smear images related to its brightness. Texture is one of the important features to improve the sensitivity and specificity of image based malaria detection techniques. We considered the global features of the image rather than segmented erythrocyte. Here, we used Olympus CX21i bright field microscope to view the malaria blood smear images. And Canon EOS1200D digital camera was used to digitize the captured view of microscope. The blood smear is examined with the total magnification of 10× in ocular lens and 100× in objective lens. The procedure is divided into four steps namely smear collection, microscope setup, focusing, and image acquisition with varied brightness, which have been explained below.

3.1 Smear Collection

Giemsa staining primarily uses two solutions: eosin and methylene. Eosin helps to change the parasite nucleus into red, and methylene will change the cytoplasm into blue. WHO recommends Giemsa staining is the reliable method to detect the malaria parasite for early case detection of malaria disease. Therefore, we have collected few Giemsa-stained blood smears infected with P. falciparum- and P. vivax from Tagore Medical College & Hospital, Chennai, India. The pathologist works for Tagore Medical College & Hospital had labeled the infected regions in the acquired image and these labeled samples were used to evaluate the efficiency of the proposed experiment.

3.2 Microscope Setup

Before proceeding to the required microscopic adjustment, we need to properly clean the microscope. For cleaning the ocular lens, objective lens, and condenser lens, lens paper or clean cotton must be used. Raise the condenser knob to check the amount of light the condenser gathers into microscopic stage. Then, adjust the iris diaphragm to control the amount of light passing through condenser. In order to examine the slides with the magnification of 10× in ocular lens and 100× in objective lens, we need to rotate the nose piece and fix it.

3.3 Focusing

To focus the specimen, we have to place the smear on the microscopic stage. Then, add a drop of liquid paraffin on the stained region of the smear to increase the resolving power of microscope. Now, move the microscopic stage in the upward direction so as to make a contact on the liquid paraffin with the objective lens. Ensure that the appropriate light has to pass through the stained region of the smear. Adjust the coarse adjustment knob, and look through the ocular eyepiece until the isolated erythrocyte gets visibility. Thereafter, tune the fine adjustment knob to focus the clear details of the erythrocyte image.

3.4 Image Acquisition with Varied Brightness

After completing the above three steps, we need to alter the brightness for same specimen with varied brightness in image acquisition. There are two ways to change the brightness variations in microscope, either increase/decrease the diameter of the iris aperture or adjusting the illumination intensity control (condenser) knob. In existing practice of microscopy, the iris aperture diameter is always inversely proportional to the magnification factor, since, we kept the objective lens in 100× magnification oil immersion and fixed the diameter of the iris aperture as constant throughout the experiment. We made the brightness variation only by adjusting the illumination control knob to vary the light emitted from the light source. In our experiment, the amount of light emitted from the light source could be varied by allowing high illumination to capture high brightness image and low illumination to high dark image and other images are captured in between these illuminations with fixed intervals. Here, the high and low illuminated images are labeled as High Bright (HB), High Dark (HD) images; the intermediate illuminated images are labeled as Medium Bright (MB), Low Bright (LB), Low Dark (LD), Medium Dark (MD) and are shown in Fig. 2.
Fig. 2

Microscopic image captured with various illuminations

4 Performance Analysis

In this analysis, we try to figure out the variation of texture descriptor value for the same image with its six classes of brightness level as described in experiment section. We derived GLCM for every image and estimated the global texture feature like energy, contrast, and homogeneity from the constructed GLCM. The obtained texture feature value indicates the measure of closeness of the distribution of GLCM, the local intensity variation of GLCM, and uniformity of the image illumination distribution.

It is observed from the previous literature, Gray World Assumption (GWA) provides the efficient image preprocessing technique to nullify the brightness and contrast variation in image-based malaria diagnosis. Hence, in this proposed analysis, we once again extracted the aforementioned texture feature for the earlier images (six classes) after preprocessing with GWA. However, we found that there is no significant nullification of invariant texture feature after GWA preprocessing. Figures 3, 4, and 5 show the evidence in persistence of variation in GLCM global texture descriptor, even after employing GWA preprocessing.
Fig. 3

Variation in GLCM texture property—contrast

Fig. 4

Variation in GLCM texture property—homogeneity

Fig. 5

Variation in GLCM texture property—energy

5 Conclusion

Though we have detected the variation of texture feature by varying the image brightness, we are in the process of categorizing the texture features which are sensitive and insensitive for malaria diagnosis. To normalize the variation of sensitive texture features, we are also in the process of formulating a model to nullify the texture variance.

References

  1. 1.
    Tangpukdee N, Duangdee C, Wilairatana P, Krudsood S (2009) Malaria diagnosis: a brief review. Korean J Parasitol 47(2):93–102.  https://doi.org/10.3347/kjp.2009.47.2.93CrossRefGoogle Scholar
  2. 2.
    Chayadevi M, Raju G (2014) Usage of art for automatic malaria parasite identification based on fractal features. Int J Video Image Process. Netw Secur 14:7–15Google Scholar
  3. 3.
    Linder N, Turkki R, Walliander M et al (2014) A malaria diagnostic tool based on computer vision screening and visualization of Plasmodium falciparum candidate areas in digitized blood smears. PLoS One 9:e104855Google Scholar
  4. 4.
    Khan NA, Pervaz H, Latif AK, Musharraf A (2014) Unsupervised identification of malaria parasites using computer vision. In: Proceedings of 11th international joint conference on computer science and software engineering (JCSSE), Pattaya, Thailand, pp 263–267Google Scholar
  5. 5.
    Das DK, Mukherjee R, Chakraborty C (2015) Computational microscopic imaging for malaria parasite detection: a systematic review. J Microsc 260(1):1–19Google Scholar
  6. 6.
    Das DK, Maiti AK, Chakraborty C (2015) Automated system for characterization and classification of malaria-infected stages using light microscopic images of thin blood smears. J Microsc 257:238–252Google Scholar
  7. 7.
    Ghosh M, Das D, Chakraborty C, Ray AK (2011) Plasmodium vivax segmentation using modified fuzzy divergence. In: Proceedings of international conference on image information processing, Shimla, India, pp 1–5Google Scholar
  8. 8.
    May Z, Aziz SSAM, Salamat R (2013) Automated quantification and classification of malaria parasites in thin blood smears. In: Proceedings of international conference on signal and image processing applications (ICSIPA), Melaka, Malaysia, pp 369–373Google Scholar
  9. 9.
    Diaz G, Gonzalez FA, Romero E (2009) Asemi-automatic method for quantification and classification of erythrocytes infected with malaria parasites in microscopic images. J Biomed Inform 42:296–307Google Scholar
  10. 10.
    Soni J (2011) Advanced image analysis based system for automatic detection of malarial parasite in blood images using susan approach. Int J Eng Sci Technol 3:5260–5274Google Scholar
  11. 11.
    Purwar Y, Shah SL, Clarke G, Almugairi A, Muehlenbachs A (2011) Automated and unsupervised detection of malarial parasites in microscopic images. Malar J 10:364Google Scholar
  12. 12.
    Hanif NSMM, Mashor MY, Mohamed Z (2011) Image enhancement and segmentation using dark stretching technique for plasmodium falciparum for thick blood smear. In: Proceedings of 7th international colloquium on signal processing and its applications (CSPA), Penang, Malaysia, pp 257–260Google Scholar
  13. 13.
    Abbas N, Mohamad D (2013) Microscopic RGB color images enhancement for blood cells segmentation in YCBCR color space for k-means clustering. J Theor Appl Inf Technol 55:117–125Google Scholar
  14. 14.
    Tek FB, Dempster AG, Kale I (2010) Parasite detection and identification for automated thin blood film malaria diagnosis. Comput Vis Image Underst 114:21–32Google Scholar
  15. 15.
    Ma C, Harrison P, Wang L, Coppel RL (2010) Automated estimation of parasitaemia of plasmodium yoelii-infected mice by digital image analysis of Giemsa-stained thin blood smears. Malar J 9:348Google Scholar
  16. 16.
    Khot ST, Prasad RK (2012) Image analysis system for detection of red blood cell disorders using artificial neural network. Int J Eng Res Technol 1:1–14CrossRefGoogle Scholar
  17. 17.
    Kumarasamy SK, Ong SH, Tan KSW (2011) Robust contour reconstruction of red blood cells and parasites in the automated identification of the stages of malarial infection. Mach Vis Appl 22:461–469Google Scholar
  18. 18.
    Suradkar PT (2013) Detection of malarial parasite in blood using image processing. Int J Eng Innov Technol 2:124–126Google Scholar
  19. 19.
    Rajesh Kanna B, Aravindan C, Kannan K (2012) Image based area estimation of any connected region using y-convex region decomposition. AEU Int J Electron Commun 66(2):172–183.  https://doi.org/10.1016/j.aeue.2011.06.010CrossRefGoogle Scholar
  20. 20.
    Rajesh Kanna B, Aravindan C, Kannan K (2012) Development of yConvex hypergraph model for contour based image analysis. In: Proceedings of the 2nd IEEE international conference computer communication and informatics (ICCCI), vol 2, pp 1–5.  https://doi.org/10.1109/iccci.2012.6158806
  21. 21.
    Ross NE, Pritchard CJ, Rubin DM, Duse AG (2006) Automated image processing method for the diagnosis and classification of malaria on thin blood smears. Med Biol Eng Comput 44:427–436Google Scholar
  22. 22.
    Springl V (2009) Automatic malaria diagnosis through microscopy imaging. Faculty of Electrical Engineering. Master thesis, Czech Technical University, In Prague, Czech RepublicGoogle Scholar
  23. 23.
    Malihi L, Ansari-Asl K, Behbahani A (2013) Malaria parasite detection in giemsa-stained blood cell images. In: Proceedings of 8th Iranian conference on machine vision and image processing (MVIP), Zanjan, Iran, pp 360–365Google Scholar
  24. 24.
    Anggraini D, Nugroho AS, Pratama C, Rozi IE, Iskandar AA, Hartono RN (2011) Automated status identification of microscopic images obtained from malaria thin blood smears. In: Proceedings of international conference on electrical engineering and informatics (ICEEI), Bandung, Indonesia, pp 1–6Google Scholar
  25. 25.
    Vink JP, Laubscher M, Vlutters R, Silamut K, Maude RJ, Hasan MU, Haan G (2013) An automatic vision-based malaria diagnosis system. J Microsc 250:166–178Google Scholar
  26. 26.
    Maity M, Maity AK, Dutta PK, Chakraborty C (2012) A web accessible framework for automated storage with compression and textural classification of malaria parasite images. Int J Comput Appl 52:31–39Google Scholar
  27. 27.
    Materka A, Strzelecki M (2015) On the effect of image brightness and contrast nonuniformity on statistical texture parameters. Found Comput Decis Sci 40. ISSN 0867-6356Google Scholar

Copyright information

© Springer Nature Singapore Pte Ltd. 2018

Authors and Affiliations

  • A. Vijayalakshmi
    • 1
  • B. Rajesh Kanna
    • 1
  • Shanthi Banukumar
    • 2
  1. 1.School of Computing Science and EngineeringVIT UniversityChennaiIndia
  2. 2.Department of MicrobiologyTagore Medical College & HospitalChennaiIndia

Personalised recommendations