411
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
A NEW MERGED SEGMENTATION TECHNIQUE USED
FOR X-RAY CHEST IMAGES
Balasubramani Perumal
Department of Electronics and Communication Engineering,
Kalasalingam Academy of Research and Education, Krishnankoil, Virudhunagar (Dt), (India).
E-mail: palanimet@gmail.com
ORCID: https://orcid.org/0000-0003-4408-9396
Emil Molayil Paul
Department of Electronics and Communication Engineering,
Kalasalingam Academy of Research and Education, Krishnankoil, Virudhunagar (Dt), (India).
E-mail: emilmpaul@gmail.com
ORCID: https://orcid.org/0000-0003-0563-4240
Recepción: 28/11/2019 Aceptación: 15/01/2021 Publicación: 30/11/2021
Citación sugerida:
Perumal, B., y Paul, E. M. (2021). A new merged segmentation technique used for X-ray chest images.
3C Tecnología. Glosas de innovación aplicadas a la pyme, Edición Especial, (noviembre, 2021), 411-429. https://
doi.org/10.17993/3ctecno.2021.specialissue8.411-429
412 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
ABSTRACT
Image processing techniques of a chest X-ray image includes noise removal followed by
segmentation, feature extraction to locate regions and classication. The chest image appears
dierently when viewed from dierent angles or under dierent lightings. Diagnosing TB
stays a challenge. The customary medical specialty, even so count, variety ways, they are
gradual and frequently unreliable. In ancient poster anterior chest radiographs, several
clinical and diagnostic functions build use of computationally designed algorithms that
assist in scientic diagnostic analysis by victimization acquisition of pictures. The Digital
image may be a necessary medium for analyzing, annotating, patient’s demographics
coverage in screening of TB via chest radiography. This paper deals with the essential
segmentation methods of TB. In our methodology, this disease can be fastly and accurately
identiable by the merge segmentation methods of K means with Marker-based Watershed
segmentation which has highest precision and recall values when compared to the several
other segmentation methods which is been is discussed. More than 80 chest Xray images
output for recall and precision is discussed here.
KEYWORDS
Watershed Segmentation, Marker-Based Watershed Segmentation, Otsu, K means,
Gamma correction.
413 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
1. INTRODUCTION
The chest X-ray lms gives certain vital information about the patients. However, the
researchers are been encouraged to develop computer algorithms to assist the radiologist in
diagnosis process. Automatic segmentation is one of the rst steps of such computer-aided
systems. Medical imaging technological know-how could be facts managing an integrated
system designed for the screening of T.B. patients. The screening technique is extended with
the help of creating the reviewing procedure faster via reducing eye fatigue. In this paper
we tend to discuss, regarding the varied automatic approaches for sleuthing the TB. Here
numerous techniques are tailored for preprocessing, segmentation techniques. Our work
combines the algorithm of marker-based watershed method along with the combination of
normalization of the input image, gamma correction and k-means clustering. This enhances
the ROI of the segmented portion. Here a comparison between various segmentation
methods is performed depending on the precision and recall concepts.
2. MATERIALS AND METHODS
Screening is widely used for detection of tuberculosis. It’s hard for a radiologist to interpret
the disease. It’s viable to improve this task through the use of CAD system (Raviglione &
Sulis, 2016). The rst paper mainly deals with the necessity of the computerized method
for the tuberculosis detection along with the various segmentations used. The pros and cons
for each segmentation method is described here. The need for segmentation and how is it
useful for x-ray chest image is referred here.
This paper depicts a voyage of the segmentation methods (Kumar et al., 2018; Karargyris,
Antani, & Thoma, 2011; Guendel et al., 2019). In TB screening, the segmentation part
paves an important role. This paper deals with the involvement of marker-based watershed
segmentation, which fairly well works than distance transform watershed transforms. Due to
the few disadvantages in watershed segmentation, the chest radiographs are well segmented
using marker-based watershed segmentation (Lu, Wang, & Zhang, 2017; Zhang et al., 2013;
Guendel et al., 2019).
The image segmentation using Otsu and K means is well explained in the paper. Depending
upon the threshold value the classication for foreground and background is being done.
414 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
K means clustering method goes in the circle of vector quantization, forms cluster of
k in n mining region (Kasu & Saravanan, 2019). The region growing method of chest
segmentation is described in Kiran et al. (2019). The normalization of the x-ray input chest
image is made due to the poor contrast or due to glare (Guendel et al., 2019; Vidya et al.,
2019; Homan, Kothari, & Wang, 2014). Then gamma correction is done as a luminance
correction (Chen & Ramli, 2003; Deivalakshmi, Saha, & Pandeeswari, 2017).
The ROI is determined by merging the result of K means and watershed segmentation
(Kasu & Saravanan, 2019; Jeyavathana, Balasubramanian, & Pandian, 2017; Betsy & Nizar,
2017). Thus, by calculating the precision and recall of every segmentation mentioned in
this paper and a total analysis of segmentation of all about 80 images are made.
2.1. WATERSHED SEGMENTATION
Automatic investigation of TB microorganism algorithms has conjointly been tested to
be less time, less human error and less man-power. The Watershed transformation is a
powerful tool for image segmentation, it uses the region-based approach. Here the concept
of an image gradient came. It is a directional change in the intensity or color in of an X-ray
chest image. It is been used to extract information from images. The approach consists
of analyzing a picture, with the color thresholding segmentation before greyscale color.
The excellence of the grayscale image has been increased, and binarization is performed.
Here, we perform an experiment for calculating the precision and recall of around 80 chest
X-Ray images using watershed segmentation. The main disadvantage of this method is
over segmentation. However, this can be reduced using internal and external markers. The
original input lung image and the watershed segmented image is shown.
Figure 1. a) Input original image, b) Watershed image.
Source: own elaboration.
415 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2.2. MARKER-BASED WATERSHED SEGMENTATION
In Watershed Segmentation, two methods are involved. The rst methodology quickly
makes use of Watershed Segmentation on the image, to boot known as Watershed Distance
transform Segmentation methodology. The second methodology, Marker-based Watershed
Segmentation, which overcomes the over segmentation problem of the rst method, limits
the variability of regions through exploitation the inside markers to specify the ROI (Kasu
& Saravanan, 2019).
Unfortunately, the real watershed transform corresponds to a minimum of the gradient
that is produced by small variations, due to noise. This leads to the over segmentation
problem, which yields inaccurate results. The disadvantage of this over segmentation can be
overcome by marking the patterns on the lung region of the x-ray image to be segmented.
It should be performed before the watershed transformation of the gradient, which is been
eectively reduced by using marker-based watershed segmentation.
Figure 2. Marker-Based Segmentation.
Source: own elaboration.
2.3. OTSU SEGMENTATION
This Otsu segmentation automatically performs clustering-based segmentation. It converts
gray level image into binary image. It accounts for foreground and background pixels. The
main drawback of this segmentation is that if the variance of both the background and image
416 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
remains high, the additive noises may disrupt the image clarity. The Otsu Segmentation is
very much limited to low size object, with small mean dierence and variance.
Figure 3. a) Input original image, b) Otsu Segmented Image.
Source: own elaboration.
2.4. REGION GROWING SEGMENTATION
The main goal of segmentation is to partition the X-ray chest images into suitable regions.
It rstly, goes through the neighboring pixels of initial seed points, then determines whether
the neighboring X-ray pixels should be added to the region. The iteration process proceeds
next. The main disadvantages of this region growing includes:
1) Proper identication of the seed points.
2) Needs more information of the image.
3) The pixels with similar threshold value regions will be considered as same regions.
4) It is computationally expensive and more sensitive to noise.
Figure 4. a) Input original image, b) Segmented Image.
Source: own elaboration.
417 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2.4. K MEANS CLUSTERING
It is a vector quantization method. It divides the n observations or pixel elements into k
clusters. Here, the observation belongs to the cluster with meaningful value of mean. An
observation for the pixel values of the input image and the K means clustering is depicted
as graph shown below.
Figure 5. a) Input original image, b) Segmented Image.
Source: own elaboration.
Green colour for cluster 1, rose colour for cluster 2, and yellow colour for cluster 3.
Here value of K=3 is made. The rst 2 values for k are for the left and right lungs and third
value is for the background. It can be applied to larger data sets too. It uses an iterative
technique.
Figure 6. a) Input original image, b) K Means Image.
Source: own elaboration.
418 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
3. PROPOSED METHOD
Our planned system contains the preprocessing followed by segmentation, feature extraction
strategies, trained information sets and also the classier. The preprocessing technique
includes median, wiener and morphological lters (Top hat and bot hat lters). The median
lters out the salt and pepper noises, but once the impulse noise is larger than 4%, it lters
out the helpful data within the chest pictures. It will bring an improbable loss in sleuthing
the TB. Therefore, the ltered output from the median lter is then fed to the Weiner
lter. This lter, in turn, smoothens and restores the required data of the chest pictures.
This output is then fed to the image improvement lters say top hat and bot hat lters.
Even once the background of the image and also the target color remains identical, the
morphological lter analyses the same and identies them as separate values even our eyes
cannot determine. The output is then fed for segmentation. Here, marker-based watershed
segmentation is used to avoid the over segmentation from distance transform watershed
segmentation. The input image is then fed for image normalization, which reduces the
after eect due to poor contrast or glare. Then the image is fed for gamma correction,
which is mainly used for luminance correction. K means clustering is also employed. The
combination of water marked watershed algorithm and the outputs followed by the k means
yields an image segmentation method which got nest recall and precision value. The Flow
chart illustration depicted below give the details of the segmentation.
419 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Graphic 1. Flow Chart.
Source: own elaboration.
4. RESULTS
4.1. PERFORMANCE PARAMETERS FOR SEGMENTATION ANALYSIS
4.1.1. PRECISION
In the chest x-ray images, precision is determined by the fraction of relevant elements
(pixels) from the retrieved elements of the lung image. It is dened as:
PR= TPO÷ (TPO+FPO)
PR-Precision
TPO-True Positive
FPO-False Positive
420 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4.1.2. RECALL
In the X-ray chest image, the recall is the fraction of relevant X-ray pixel elements of the
chest image that is been retrieved from the over the total amount of relevant pixel elements
in the X-ray chest images. It is dened as:
Re=-TPO÷ (TPO+FNE)
Re-Recall
TPO-True Positive
FNE-False Negative
Figure 7. Details about precision and recall.
Source: own elaboration.
The graphical representation of both precision and recall values of various segmentation
methods are discussed below.
421 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4.1.3. FOR OTSU
010 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
images -->
Recall
Recall
Figure 8. a) Recall values for Otsu segmentation for 80 chest images.
Source: own elaboration.
010 20 30 40 50 60 70 80 90
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
images -->
Precision
Precision
Figure 8. b) Precision for Otsu Segmentation for 80 chest images.
Source: own elaboration.
422 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4.1.4. FOR WATERSHED SEGMENTATION
010 20 30 40 50 60 70 80 90
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
images -->
Recall
Figure 9. a) Recall for Watershed Segmentation for 80 chest images.
Source: own elaboration.
010 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
images -->
Precision
Figure 9. b) Precision for Watershed Segmentation for 80 chest images.
Source: own elaboration.
423 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4.1.5. FOR MARKER-BASED WATERSHED SEGMENTATION
010 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
images -->
Recall
Figure 10. a) Recall for Marker-based Watershed Segmentation for 80 chest images.
Source: own elaboration.
010 20 30 40 50 60 70 80 90
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
images -->
Precision
Figure 10. b) Precision for Marker-based Watershed Segmentation for 80 chest images.
Source: own elaboration.
424 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4.1.6. COMPARISON BETWEEN VARIOUS SEGMENTATION METHODS
010 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
images-->
Precision
watershed
otsu
regiongrowing
marker based
kmeans+marker
Figure 11. a) Precision Value comparison for watershed segmentation, Otsu, region growing, k means merged
marker-based watershed segmentation for 80 chest X-ray images.
Source: own elaboration.
010 20 30 40 50 60 70 80 90
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
images-->
Recall
watershed
otsu
regiongrowing
marker based
kmeans+marker
Figure 11. b) Recall Value comparison for watershed segmentation, Otsu, region growing, k means merged
marker-based watershed segmentation for 80 chest X-ray images.
Source: own elaboration.
425 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
5. CONCLUSIONS
Our paper gives a novelty approach of merging the segmentation methods for achieving a
perfect ROI for getting better results for the precision and recall factors. Our approach is
a new technique which merges two outputs of segmentation parts and accurately xes the
ROI. The input image is preprocessed, then fed for marker-based watershed segmentation.
Meanwhile, the same input X-ray image is normalized in the other section, due to poor
contrast or glare, then gamma correction is performed for the luminance part why because
depending on the value of the gamma factor, the foreground and background visibility is
made clearer. By increasing the value of gamma clear contrast between the backgrounds
are behold in the x-ray image with mote clarity. Then k clusters are formed from the n
neighboring elements. The output from this section is merged with the marker-based
watershed segmentation and gets the correct ROI. Thus, new merged method is adapted
for segmentation whose value is found to be great for the precision and recall value for
around 80 images which is been shown in the graphs. This combination knocks out all the
other segmentation methods so far discussed.
REFERENCES
Betsy, A., & Nizar, B. P. K. (2017). Lung Tuberculosis Detection Using X-Ray Images.
International Journal of Applied Engineering Research, 12(24), 15196-15201. https://www.
ripublication.com/ijaer17/ijaerv12n24_175.pdf
Chen, S.-D., & Ramli, A. R. (2003). Contrast enhancement using recursive mean-separate
histogram equalization for scalable brightness preservation. IEEE Transactions on
Consumer Electronics, 49, 1301-1309. https://doi.org/10.1109/TCE.2003.1261233
Deivalakshmi, S., Saha, A., & Pandeeswari, R. (2017). Raised Cosine Adaptive Gamma
Correction for Ecient Image and Video Contrast Enhancement. In TENCON 2017-
2017 IEEE Region 10 Conference. https://doi.org/10.1109/tencon.2017.8228256
Furutani, K., Hirano, Y., & Kido, S. (2019). Segmentation of lung region from chest
x-ray images using U-net. In 2019 Joint International Workshop on Advanced Image
426 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Technology (IWAIT) and International Forum on Medical Imaging in Asia (IFMIA). https://
doi.org/10.1117/12.2521594
Guendel, S., Ghesu, F. C., Grbic, S., Gibson, E., Georgescu, B., Maier, A.,
& Comaniciu, D. (2019). Multi-task Learning for Chest X-ray Abnormality
Classication on Noisy Labels. Computer Vision and Pattern Recognition. https://arxiv.
org/abs/1905.06362
Homan, R. A., Kothari, S., & Wang, M. D. (2014). Comparison of normalization
algorithms for cross-batch color segmentation of histopathological images. In 2014
36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society.
https://doi.org/10.1109/embc.2014.6943562
Jeevitha, D., & Rajasekaran, J. (2015). Histogram based tuberculosis analysis with
parallel pixel processing. Journal of Recent Research in Engineering and Technology, 2, 2349
–2260. http://www.jrret.com/archives/2015/paper/jun15/J61511.pdf
Jeyavathana, R. B., Balasubramanian, R., & Pandian, A. (2017). An Ecient Feature
Extraction Method for Tuberculosis detection using Chest Radiographs. International
Journal of Applied Environmental Sciences, 12(2), 227-240. https://www.ripublication.
com/ijaes17/ijaesv12n2_02.pdf
Kamra, P., Vishraj, R., & Gupta, S. (2015). Performance comparison of image
segmentation techniques for lung nodule detection in CT images. In 2015 International
Conference on Signal Processing, Computing and Control (ISPCC). https://doi.org/10.1109/
ISPCC.2015.7375045
Karargyris, A., Antani, S., & Thoma, G. (2011). Segmenting anatomy in chest x-rays
for tuberculosis screening. In 2011 Annual International Conference of the IEEE Engineering
in Medicine and Biology Society. https://doi.org/10.1109/IEMBS.2011.6091917
Kasu, N. R., & Saravanan, C. (2019). Segmentation on Chest Radiographs Using Otsu's
and K-Means Clustering Methods. In 2018 International Conference on Inventive Research
in Computing Applications (ICIRCA). https://doi.org/10.1109/ICIRCA.2018.8597371
427 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Kiran, M., Ahmed, I., Khan, N., & Reddy, A. G. (2019). Chest X-ray segmentation
using Sauvola thresholding and Gaussian derivatives responses. Journal of Ambient
Intelligence and Humanized Computing, 10, 4179–4195. https://doi.org/10.1007/
s12652-019-01281-7
Kumar, S. N., Fred A. L., Muthukumar, S., Kumar, H., A., & Varghese, P, S. (2018).
A Voyage On Medical Image Segmentation Algorithms. Biomedical Research, 29, 75-
87. https://doi.org/10.4066/biomedicalresearch.29-16-1785
Lu, S., Wang, S., & Zhang, Y. (2017). A note on the marker-based watershed method for
X-ray image segmentation. Computer Methods and Programs in Biomedicine, 141, 1-128.
https://doi.org/10.1016/j.cmpb.2017.01.014
Manisha, R. K., & Palanisamy, K. S. (2016). Computer-aided diagnosis of tuberculosis
using chest radiographs. Discovery, 52(245), 1012-1019. http://www.discoveryjournals.
org/discovery/current_issue/v52/n245/A9.pdf
Poornimadevi, C. S., & Sulochana, H. (2016). Automatic Detection Of Pulmonary
Tuberculosis Using Image Processing Techniques. In 2016 International Conference
on Wireless Communications, Signal Processing and Networking (WiSPNET). https://doi.
org/10.1109/WiSPNET.2016.7566243
Quadeer, M. A., Naaz, S., & Zerdi, N. (2017). Image Segmentation And Classication
Model For Detection Of Tuberculosis. International Journal of Emerging Technology in
Computer Science & Electronics (IJETCSE), 24(6).
Raviglione, M., & Sulis, G. (2016). Tuberculosis 2015: Burden, Challenges and Strategy
for Control and Elimination. Infectious Disease Reports, 8(2), 6570. https://doi.
org/10.4081/idr.2016.6570
Surya, S. J., Lakshmanan, S., & Stalin, A. (2017). Automatic Tuberculosis Detection
Using Chest Radiographs Using Its Features Abnormality Analysis. Journal of
Recent Research in Engineering and Technology, 4(9). https://www.researchgate.net/
publication/343793401_AUTOMATIC_TUBERCULOSIS_DETECTION_
428 https://doi.org/10.17993/3ctecno.2021.specialissue8.411-429
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
USING_CHEST_RADIOGRAPHS_USING_ITS_FEATURES_
ABNORMALITY_ANALYSIS
Vidya, M. S., Manikanda, K. V., Anirudh, G., Srinivasa, R. K., & Vijayananda,
J. (2019). Local and global transformations to improve learning of medical
images applied to chest radiographs. In SPIE Medical Imaging. https://doi.
org/10.1117/12.2512717
Vital, D. A., Sais, B. T., & Moraes, M. C. (2019). Automatic Pulmonary Segmentation
in Chest Radiography, Using Wavelet, Morphology and Active Contours. In Costa-
Felix, R., Machado, J., & Alvarenga, A. (eds.) XXVI Brazilian Congress on Biomedical
Engineering. IFMBE Proceedings, vol 70/2. Springer, Singapore. https://doi.
org/10.1007/978-981-13-2517-5_12
Zhang, X., Jia, F., Luo, S., Liu, G., & Hu, Q. (2013). A marker-based watershed method
for X-ray image segmentation. Computer Methods and Programs in Biomedicine, 113(3),
894-903. https://doi.org/10.1016/j.cmpb.2013.12.025
