Leukemia Detection using Object Oriented Method

Sai Sreeja Alapati, Min Bai2020 · DOI

TL;DR

This paper proposes an object-oriented approach for detecting leukemia from microscopic blood cell images. By segmenting images using RGB, CMYK, and YCbCr color spaces and analyzing cell counts, the system aims to provide a more accurate and cost-effective alternative to traditional blood tests.

What Problem Does This Paper Address?

Blood cancers, particularly leukemia, disrupt the normal production and function of blood cells, often leading to uncontrolled growth of abnormal cells. Early and accurate detection is crucial for effective treatment. Traditional methods can be time-consuming and expensive. This paper addresses the need for an automated, objective, and potentially more accessible method for leukemia detection by leveraging image processing techniques on microscopic blood cell images.

Methods

The core of the proposed method is an object-oriented approach that processes microscopic blood cell images through a series of modules. The process begins with an Image Module to acquire the input image. Subsequently, the image is processed through three distinct color space modules:

  • RGB Color Space Module: Analyzes the image in the Red, Green, Blue color space to differentiate between red blood cells (RBCs) and white blood cells (WBCs).
  • CMYK Color Space Module: Processes the image using the Cyan, Magenta, Yellow, Key color model, which is typically used in printing.
  • YCbCr Color Space Module: Utilizes the YCbCr color space, commonly employed in digital photography and videography, for further analysis.

Each color space module aims to extract cell counts. The Result Calculation Module then aggregates the findings from these three modules. It calculates the percentage of RBCs and WBCs relative to the total detected blood cells. A key criterion for potential leukemia detection is a significant difference (greater than 20%) in the percentage of WBCs compared to RBCs. The system also includes an "Export Data" feature to save results.

Datasets

The paper mentions using the ALL-IDB database for testing their system. This database contains images of acute lymphoblastic leukemia. The authors note that if images are not from this specific database, they might require preprocessing (exposure and highlights adjustment) for optimal results.

Key Insights and Ideas

  • Multi-Color Space Analysis for Robustness: The paper's central idea is that analyzing blood cell images across multiple color spaces (RGB, CMYK, YCbCr) can enhance detection accuracy. This approach aims to capture different visual characteristics of cells that might be missed by a single color space.
  • Object-Oriented Modular Design: The system is structured into distinct modules, making it potentially easier to understand, maintain, and extend.
  • Quantitative Cell Count for Diagnosis: The method relies on quantifying the number of RBCs and WBCs, moving beyond subjective visual inspection to a more objective, data-driven diagnosis.
  • Cost-Effectiveness: The authors suggest this automated method is cost-optimized compared to traditional blood tests.

Key Results

The paper demonstrates the system's functionality through figures showing the user interface and output. The results section indicates that the system was tested on a sample image from the ALL-IDB database. The output shows the processed images from the RGB, CMYK, and YCbCr modules, along with the calculated cell counts and percentages. The paper states that the final conclusion is drawn by considering the majority opinion across all three color spaces. For instance, one output example shows "POTENTIAL LEUKEMIA DETECTED" with specific cell counts and percentages, while another shows "Normal." The paper claims the method is effective and can be digitalized.

Limitations and Future Work

  • Limited Dataset Mention: While the ALL-IDB is mentioned, the paper doesn't detail the extent of its use or if other datasets were employed for training or validation. The reliance on a single database might limit generalizability.
  • Preprocessing Requirement: For images not from the ALL-IDB, preprocessing is noted as necessary, implying the system's performance might be sensitive to image quality and acquisition parameters.
  • Generalization to Other Cancers: The paper mentions testing for "all five highly occurring cancers" but only explicitly shows an example related to plasma cell leukemia and acute lymphoblastic leukemia. Further details on the system's performance across a broader range of blood cancers would be beneficial.
  • Future Work: The paper suggests that the testing can be digitalized, implying a potential for further automation and integration into digital pathology workflows.

Summary updated

Abstract

Blood cancers affect the production and function of your blood cells. In most blood cancers, the normal blood cell development process is interrupted by uncontrolled growth of an abnormal type of blood cell. This paper will present an approach for the detection and tracking of Leukemia. This paper aims at identifying leukemia, by using image segmentation. A microscopic image of blood sample is taken, the cell count is extracted using image processing module by module. The purpose is to increase the accuracy using three different approaches object wise.
BibTeX
@article{alapati2020leukemia,
  author = {Sai Sreeja Alapati and Min Bai},
  title = {{Leukemia Detection using Object Oriented Method}},
  journal = {International Journal of Engineering Research and},
  year = {2020},
  doi = {10.17577/ijertv9is050374},
}
Figure