Comparative Study on Different Color Spaces for Segmentation of Acute Leukemia using Automatic Otsu Clustering

• Published in: 2022 IEEE 18th International Colloquium on Signal Processing and Applications, CSPA 2022 - Proceeding
• Main Author: Mohd Zairy S.M.; Hazwani Abd Halim N.; Sulaiman M.S.; Saad Z.
• Format: Conference paper
• Language: English
• Published: Institute of Electrical and Electronics Engineers Inc. 2022
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85132739664&doi=10.1109%2fCSPA55076.2022.9781962&partnerID=40&md5=bece5a9f456ee4f1013957945ddb4ed4

Leukemia is common cancer that occurs due to the abnormality in the production of white blood cells in the human blood. This disease which is likely to affect youngsters below the age of 15 years old can be detected by analysing bone marrow samples. However, the process of extracting bone marrow samples is complicated and this could cause patients to experience discomfort, hence blood samples could serve as alternatives since it also contains white blood cell (WBC) information that was needed in determining acute lymphocytic leukemia (ALL). As the development of ALL is quite fast, the early detection of the diseases is vital; hence this segmentation method was developed. The blood images of ALL patients will first be converted from RGB to HSV and Lab color space before the grayscale images produced used as the input to Otsu by clustering technique resulting in black background images while the cells will be staying in white color. Next, the watershed ridge technique has been used to remove the overlapping cells before retrieving the colored version of the WBC images. The results from testing the image segmentation method using a different component from different color spaces showed that the images applied with component b from the Lab color space proved to produce the clearest image of ALL subtypes compared to the other color components applied as the accuracy produced for B cell, T cell and normal cell were at 99.17%, 99.88%, and 99.92% respectively. © 2022 IEEE.