Comparison of Superpixel and Correlation Based Segmentation for

Improved WBC Segmentation in Microscopic Images

Rayala Ramanjaneyulu and V. Rajmohan

Saveetha University Chennai, India

Keywords: Segmentation, Novel Superpixel Algorithm, Correlation Based Segmentation, White Blood Cells,

Microscopicimage, Accuracy, Medical.

Abstract: The objective of this investigation is to refine the precision of segmentation for white blood cells (WBCs) by

employing an innovative Superpixel algorithm. Additionally, the efficacy of this innovative approach will be

evaluated by comparing it to the correlation-based segmentation technique. Methods and Materials: The

specimens were examined using the new Superpixel Algorithm as group 1, encompassing N=10 samples,

while adopting the Correlation-based method as group 2 segmentation, also comprising N=10 samples. A

preliminary test power of 80%, alongside alpha and beta values of 0.05 and 0.2, respectively, were assumed,

along with a 95% confidence interval. The analysis was conducted using the MATLAB software. Results:

The novel Superpixel Algorithm demonstrated an accuracy of 87% in segmenting white blood cells, whereas

the Correlation-based segmentation achieved 74% accuracy. A statistically significant p-value of 0.001 (p <

0.05, two-tailed), determined through SPSS statistical analysis, indicates a substantial disparity in the

segmentation data. This difference affirms the absence of errors in the acquired data. Conclusion: The

Superpixel method surpasses the correlation-based segmentation approach in terms of enhancing the accuracy

of white blood cell segmentation.

1 INTRODUCTION

Accurate segmentation of white blood cells (WBCs)

within microscopic images holds paramount

importance in the domain of medical image analysis.

This significance arises from the fact that the

accuracy of the segmentation outcomes directly

impacts the overall precision of subsequent analyses

and diagnostic interpretations. Segmentation

algorithms play a pivotal role in this procedure by

delineating the target entities from the background

(Alharbi et al. 2022). Various methodologies for

WBC segmentation in microscopic images have been

explored, including thresholding, edge detection, and

region-based techniques. Among these approaches,

two of the most extensively utilized methods

encompass the innovative superpixel algorithm and

the correlation-based segmentation (Khan and Mir

2022). The novel superpixel algorithm partitions the

image into small, homogenous regions and

aggregates them into larger segments, whereas the

correlation-based segmentation employs the intensity

correlations among pixels to effectuate image

segmentation. Nonetheless, owing to the diverse

dissimilarities in shape, size, edges, and positions of

WBC cells, data collection can prove intricate in

medical devices. Furthermore, the lighting conditions

during image capture influence the extent of contrast

between the background and cell boundaries (AS et

al. 2013). The analysis of WBCs involves both cell

segmentation and feature extraction. During the white

blood cell segmentation process, leukocytes are

extracted from blood smear images. This extraction

endeavors to unveil distinctive attributes that

facilitate their differentiation from other categories of

blood cells (Mask R-CNN 2022).

A comprehensive compilation of research

materials encompasses 80 papers on GitHub, 40

papers on Google Scholar, and 30 papers within the

MathWorks articles repository. These publications

have surfaced within the last five years and pertain to

the specific research domain at hand. The

compositional analysis of white blood cells imparts

valuable medical insights into the patients' condition

(Vickram, A. S et al 2020). The findings derived from

the estimation provide crucial understanding of the

patient's health status, thereby streamlining the

diagnostic process. As a result, the accuracy of the

algorithmic differential white blood cell count

assumes a pivotal role in this context. The

Ramanjaneyulu, R. and Rajmohan, V.

Comparison of Superpixel and Correlation Based Segmentation for Improved WBC Segmentation in Microscopic Images.

DOI: 10.5220/0012543100003739

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Artiﬁcial Intelligence for Internet of Things: Accelerating Innovation in Industry and Consumer Electronics (AI4IoT 2023), pages 31-36

ISBN: 978-989-758-661-3

methodology frequently entails prevalent techniques

like image segmentation, feature extraction, and

classification to accomplish its objectives. Previous

studies indicate that this leads to diminished

accuracy, subsequently raising the accuracy level. To

address this limitation, a comparison is made between

the novel superpixel algorithm for WBC

segmentation and the correlation-based segmentation

(Lagerberg and Korte 2020) (Ramkumar, G. et al.

2021).

The microscopic examination of medical blood

smears, encompassing the quantification of diverse

types of white blood cells (WBCs), constitutes a

widely employed blood test in the realm of medical

hematology. Consequently, the enhancement of

precision for white blood cell images within the

microscopic view during the segmentation process

emerges as a crucial objective. This enhancement is

pursued through the integration of the innovative

superpixel algorithm with correlation-based

segmentation methodologies (Faried Effendy. 2022)

(Nanmaran et al 2022). As observed, some of the

drawbacks associated with existing techniques, such

as Correlation-based segmentation, involve

susceptibility to noise and artifacts in the image,

potentially leading to false positives or false

negatives. This approach necessitates a reference

image or template, which may not be available or

suitable for WBCs within microscopic images.

Additionally, Correlation-based segmentation might

struggle to capture the intricate shapes and structures

of WBCs, resulting in inaccurate segmentation

outcomes. The process is computationally demanding

and may necessitate substantial time and resources for

the analysis of extensive images or datasets.

The objective of this study is to surmount the

limitations associated with correlation-based

segmentation by implementing the super pixel

algorithm. The integration of the super pixel

algorithm renders the segmentation process more

resilient to noise and image artifacts, thereby

augmenting segmentation accuracy and conferring

greater adaptability compared to correlation-based

segmentation. This adaptability stems from the ability

to tailor super pixel algorithms to match various

image types and objects. For instance, distinct super

pixel algorithms can be employed based on the size

and shape of the segmented WBCs (Junaid Mir,

2022). From a broader perspective, the application of

a super pixel algorithm exhibits the potential to refine

the precision of white blood cell (WBC) segmentation

in microscopic images, surpassing the capabilities of

the correlation-based segmentation technique.

2 MATERIALS AND METHODS

This study was conducted within the Digital Image

Processing Laboratory, located in the Department of

Electronics and Communication Engineering at

Saveetha School of Engineering, which is a part of

SIMATS (Saveetha Institute of Medical and

Technical Sciences), situated in Tamil Nadu, India.

The central focus of this investigation was to enhance

the precision of white blood cell segmentation within

microscopic images. This improvement was

accomplished through the incorporation of an

innovative super pixel algorithm, and its performance

was assessed in comparison to correlation-based

segmentation. The determination of the sample size

was based on findings from a prior study (Nithyaa et

al. 2021). The current analysis was executed using the

clinical.com platform, maintaining a statistical power

of 80% with a confidence interval set at 0.86% and a

threshold set at 86%. Each group consisted of a

sample size of 10, resulting in an overall sample size

of 20. Furthermore, all images were uniformly resized

to a standard dimension of 512x512.

Super pixel Algorithm

Steps to super pixel algorithm :

• Input image

• SLIC segmentation

• DBSCAN clustering.

• Combining the diminutive and noisy super

pixels.

• Segmentation followed by amalgamation

of sizable and noisy super pixels.

For the purpose of obtaining merged regions, the

initial clusters are combined by considering factors

such as colour similarity and spatial adjacency.

Noisy super pixels with mixed colours are

segmented and subsequently integrated within each

of the merged regions.

Segmented image

The algorithmic steps and calculations are

performed according to the equation provided below

(Equation 1). The calculation of the number of pixels

in the input image (N) is derived from the input

image, and the number of super pixels employed for

segmenting the input image (K). The segmentation

output is obtained by dividing N by K.

S = √N/K (1)

N/K gives Approximate size of each super pixel.

Correlation based segmentation

Steps to correlation-based segmentation

1. Both sequences should be loaded.

2. Codification of images.

3. The correlation procedure.

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics

4. Determine the peak location and intensity

5. peak>threshold? If yes go to step 6, if no

follow step 7.

6. The item is in the scene

7. The item is not the scene

8. What was the last object? If yes go to follow

step 9, if no go back to step 3

9. Evaluation.

Table 1: The process of collecting data for the Superpixel

Algorithm.

Iterations

Accuracy (%)

84.67

84.21

85.14

85.95

86.32

86.84

88.54

89.36

89.34

90.24

The correlation coefficient (ρ) acts as an indicator

of both the strength and direction of the linear

correlation existing between two variables,

designated as x and y. The covariance (cov(x,y))

between x and y measures the degree to which these

variables display concurrent fluctuations. Equation 2

is expressed as the mean value of the products of the

deviations of x and y from their individual means.

ρx,y = cov/(x,y)/σx σy (2)

Where, (cov) means covariance between x & y

σx, σy are variance of x & y

Data Collection: A dataset of microscopic images

featuring white blood cells (WBCs) was assembled

from publicly accessible sources such as medical

image repositories. The selection of images aimed to

encompass a diverse spectrum of WBC types and

imaging conditions.

Image Pre-processing: The amassed medical

microscope images underwent pre-processing

procedures designed to eliminate noise, artifacts, and

background details. Techniques like image denoising

and image thresholding were employed for this

purpose.

Segmentation Algorithms: The MATLAB

environment was employed to implement both the

Superpixel algorithm and the correlation-based

segmentation algorithm. The Superpixel algorithm

was utilized to generate Superpixels, which in turn

were applied for segmenting WBCs (R. Aarthi,

2019). On the other hand, the correlation-based

segmentation algorithm adopted a correlation-centric

approach for the WBC segmentation process.

Evaluation Metrics: The appraisal of the

effectiveness of the two algorithms encompassed the

utilization of diverse metrics, encompassing

accuracy, sensitivity, and specificity. These metrics

were computed for each image within the dataset and

were averaged across the entire dataset, furnishing a

comprehensive comparative assessment of the two

algorithms.

3 STATISTICAL ANALYSIS

Statistical analysis of the collected data for the

segmentation of white blood cells was carried out

using SPSS version 21 (Kelly Small, 2022). Within

the SPSS software, the Independent Sample t-test and

group statistics were calculated. In this context,

sensitivity, specificity, and noise were considered as

independent variables, whereas accuracy was treated

as the dependent variable.

Table 2: A comparison between the Superpixel algorithm and correlation-based segmentation.

Comparison of Superpixel and Correlation Based Segmentation for Improved WBC Segmentation in Microscopic Images

Figure 1: (a)&(b) White blood cells extracted from microscopic images (c) Mean accuracy Superpixel algorithm and the

Correlation-based segmentation algorithm.

4 RESULTS

Table 1 presents the compiled data for the Superpixel

Algorithm, utilizing a sample size of 10 (N = 10),

pertaining to the segmentation of white blood cells

through microscopic images. The results highlight

notable improvements in accuracy. Table 2 exhibits

the collected data concerning the Correlation-Based

Segmentation Algorithm, employing a sample size of

10 (N = 10), focusing on the segmentation of white

blood cells within microscopic images. Remarkably,

the findings suggest comparatively diminished

accuracy.

Figure 1a: White blood cells extracted from

microscopic images (datasets) were obtained from

blood samples for analysis. These cells were

employed as input for data collection via image

extraction. The resultant data encompassed counts of

both Red Blood Cells (RBCs) and White Blood Cells

(WBCs), along with corresponding accuracy

measurements.

Figure 1b: White blood cells extracted from

microscopic images (datasets) were sourced from

blood samples for examination. These cells served as

input for data collection through image extraction,

facilitating the determination of counts for both Red

Blood Cells (RBCs) and White Blood Cells (WBCs),

alongside accuracy measurements.

With an impressive overall accuracy of 87.06%,

this approach underscores its effectiveness in

successfully segmenting white blood cells within

microscopic images. It has been observed that a

comparison between the Superpixel and correlation-

based segmentation yields a plotted mean accuracy.

The highest accuracy is attained by the Superpixel

algorithm for image segmentation at 87.06%

5 DISCUSSIONS

To enhance the precision of white blood cell (WBC)

segmentation in microscopic images, two

methodologies were employed. In a comparative

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics

assessment with correlation-based segmentation, the

Superpixel technique exhibited superior performance

in enhancing segmentation accuracy, yielding an

average accuracy score of 88% (Sera Tort. 2020). In

the SPSS statistical analysis, the resultant data

showcased a significantly lower p-value of 0.001%

(p<0.05), indicative of its statistical significance.

Employing the Superpixel method yielded a

segmentation accuracy of 87.06% for white blood

cells in images, whereas the utilization of the

correlation-based segmentation approach resulted in

an accuracy of 74%. These outcomes distinctly

underscore the superior performance of the

Superpixel algorithm (Shahzad et al. 2020). With the

application of the superpixel technique, the accuracy

was determined to be 87.06%, signifying a 13%

enhancement. Comparatively, the accuracy obtained

from a different algorithm (clustering) was measured

at 85% (Deshpande et al. 2022). The segmentation

process, implemented through the Superpixel

algorithm, exhibited a heightened accuracy of 87%,

consequently bolstering precision. Regarding this

research, no contradictory findings were encountered

(Al-Dulaimi et al. 2021).

A limitation of this study lies in the fact that

correlation-based segmentation might necessitate a

substantial volume of training data to achieve precise

WBC segmentation. This requirement could

potentially lead to a time-consuming and costly

process. Furthermore, correlation-based

segmentation could encounter challenges when

dealing with intricate objects, including overlapping

cells or cells exhibiting irregular shapes. In contrast,

super pixel algorithms might demand less training

data and prove to be more time-efficient. Thus, the

super pixel algorithm exhibits superiority over

correlation-based segmentation (Fatichah et al.

2022).

For future endeavours, the focus is on developing

algorithms that can attain high-accuracy

segmentation within a reduced timeframe. The goal is

to produce algorithms capable of achieving optimal

accuracy levels while streamlining the time required

for the segmentation process.

6 CONCLUSION

Through an evaluation of performance metrics and a

comparison of two algorithms, two different

techniques for calculating the accuracy of WBC

segmentation were scrutinized. The results revealed

that the super pixel algorithm achieved a notably

higher mean accuracy of 87% in comparison to the

correlation-based segmentation, which yielded a

mean accuracy of 74%. Consequently, the super pixel

algorithm demonstrated enhanced efficacy in

enhancing image accuracy. Additionally, the standard

deviation for the super pixel algorithm was

determined to be 2.16385, while the correlation-based

segmentation exhibited a higher standard deviation of

2.99625.

REFERENCES

Al-Dulaimi, Khamael, Jasmine Banks, Kien Nguyen,

Aiman Al-Sabaawi, Inmaculada Tomeo-Reyes, and

Vinod Chandran. (2021). “Segmentation of White

Blood Cell, Nucleus and Cytoplasm in Digital

Haematology Microscope Images: A Review-

Challenges, Current and Future Potential Techniques.”

IEEE Reviews in Biomedical Engineering 14

(January): 290–306.

Alharbi, Amal H., C. V. Aravinda, Meng Lin, P. S.

Venugopala, Phalgunendra Reddicherla, and Mohd

Asif Shah. (2022). “Segmentation and Classification of

White Blood Cells Using the UNet.” Contrast Media &

Molecular Imaging 2022 (July): 5913905.

Almassri, Hasan, Tim Dackermann, and Norbert Haala.

(2019). “mDBSCAN: Real Time Superpixel

Segmentation by DBSCAN Clustering Based on

Boundary Term.” Proceedings of the 8th International

Conference on Pattern Recognition Applications and

Methods. https://doi.org/10.5220/0007249302830291.

AS, Vickram, Raja Das, Srinivas MS, Kamini A. Rao, and

Sridharan TB. (2013)"Prediction of Zn concentration in

human seminal plasma of Normospermia samples by

Artificial Neural Networks (ANN)." Journal of assisted

reproduction and genetics 30 (2013): 453-459.

Axel, Leon, Puneet Bhatla, Dan Halpern, Silvia Magnani,

Jadranka Stojanovska, and Chirag Barbhaiya. (2023).

“Correlation of MRI Premature Ventricular Contraction

Activation Pattern in Bigeminy with Electrophysiology

Study-Confirmed Site of Origin.” The International

Journal of Cardiovascular Imaging 39 (1): 145–52.

Burch, Jane, and Sera Tort. (2020). “What Is the Accuracy

of Changes in White Blood Cell (WBC) Count for

Diagnosing COVID-19?” Cochrane Clinical Answers.

https://doi.org/10.1002/cca.3398.

Casler, Kelly Small. (2022). “White Blood Cells and the

Differential Count.” Laboratory Screening and

Diagnostic Evaluation.

https://doi.org/10.1891/9780826188434.0165.

“Classification of Acute Lymphoblastic Leukemia on

White Blood Cell Microscopy Images Based on

Instance Segmentation Using Mask R-CNN.” (2022).

International Journal of Intelligent Engineering and

Systems. https://doi.org/10.22266/ijies2022.1031.54.

Deshpande, Nilkanth Mukund, Shilpa Gite, Biswajeet

Pradhan, Ketan Kotecha, and Abdullah Alamri. (2022).

“Improved Otsu and Kapur Approach for White Blood

Comparison of Superpixel and Correlation Based Segmentation for Improved WBC Segmentation in Microscopic Images

Cell Segmentation Based on LebTLBO Optimization

for the Detection of Leukemia.” Mathematical

Biosciences and Engineering: MBE 19 (2): 1970–2001.

Fatichah, Chastine, Diana Purwitasari, Victor Hariadi, and

Faried Effendy. (2022). “Overlapping White Blood

Cell Segmentation And Counting On Microscopic

Blood Cell Images.” International Journal on Smart

Sensing and Intelligent Systems.

https://doi.org/10.21307/ijssis-2017-705.

Khan, Rehan, and Junaid Mir. (2022). “White Blood Cells

Segmentation and Classification Using U-Net CNN and

Hand-Crafted Features.” 2022 International

Conference on IT and Industrial Technologies (ICIT).

https://doi.org/10.1109/icit56493.2022.9988955.

Nanmaran, R., Srimathi, S., Yamuna, G., Thanigaivel, S.,

Vickram, A. S., Priya, A. K., ... & Muhibbullah, M.

(2022). Investigating the role of image fusion in brain

tumor classification models based on machine learning

algorithm for personalized medicine. Computational

and Mathematical Methods in Medicine, 2022.

Nithyaa1, A. N. Nithyaa A. N., A. N. Nithyaa A. N.

Nithyaa, Kumar R. Prem, Gokul M Gokul, and Geetha

Aananthi C. (2021). “Matlab Based Potent Algorithm

for WBc Cancer Detection and Classification.”

Biomedical and Pharmacology Journal.

https://doi.org/10.13005/bpj/2328.

Ramalakshmi, M., & Vidhyalakshmi, S. (2021). GRS

bridge abutments under cyclic lateral push. Materials

Today: Proceedings, 43, 1089-1092.

Ramkumar, G. et al. (2021). “Energetic Glaucoma

Segmentation and Classification Strategies Using

Depth Optimized Machine Learning Strategies”

Contrast Media & Molecular Imaging Volume 2021,

Article ID 5709257, 11 Pages, 2021

https://doi.org/10.1155/2021/5709257

S. K. Sarangi, Pallamravi, N. R. Das, N. B. Madhavi, P.

Naveen, and A. T. A. K. Kumar, “Disease Prediction

Using Novel Deep Learning Mechanisms,” J. Pharm.

Negat. Results, vol. 13, no. 9, pp. 4267–4275, (2022),

doi: 10.47750/pnr.2022.13.S09.530

Shahzad, Muhammad, Arif Iqbal Umar, Muazzam A. Khan,

Syed Hamad Shirazi, Zakir Khan, and Waqas Yousaf.

(2020). “Robust Method for Semantic Segmentation of

Whole-Slide Blood Cell Microscopic Images.”

Computational and Mathematical Methods in Medicine

2020 (January): 4015323.

V, Arsha P., P. V. Arsha, and Pillai Praveen Thulasidharan.

(2018). “A Survey Of White Blood Cell Segmentation

In Medical Image Analysis.” International Journal of

Computer Sciences and Engineering.

https://doi.org/10.26438/ijcse/v6si6.9194.

Verma, J. P. (2013). “Data Analysis in Management with

SPSS Software.” https://doi.org/10.1007/978-81-322-

0786-3.

Vickram, A. S., Samad, H. A., Latheef, S. K., Chakraborty,

S., Dhama, K., Sridharan, T. B., ... & Gulothungan, G.

(2020). Human prostasomes an extracellular vesicle–

Biomarkers for male infertility and prostrate cancer:

The journey from identification to current knowledge.

International journal of biological macromolecules,

146, 946-958.

Khan, Rehan, and Junaid Mir. (2022). “White Blood Cells

Segmentation and Classification Using”

V. P. Parandhaman, "A Secured Mobile Payment

Transaction Handling System using Internet of Things

with Novel Cipher Policies," (2023) International

Conference on Advances in Computing,

Communication and Applied Informatics (ACCAI),

Chennai, India, 2023, pp. 1-8, doi:

10.1109/ACCAI58221.2023.10200255.

U-Net CNN and Hand-Crafted Features.”

(2022)International Conference on IT and Industrial

Technologies (ICIT).

https://doi.org/10.1109/icit56493.2022.9988955.

Ware, Alisha D. (2020). “The Complete Blood Count and

White Blood Cell Differential.”

Contemporary Practice in Clinical Chemistry.

https://doi.org/10.1016/b978-0-12-815499-1.00025-9.

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics