Leukemia Detection Algorithms: Comparative study
No Thumbnail Available
Date
2022
Authors
riad, Sada
wazani, Maryam
Abuzant, Aleen
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Our project is about Leukemia Detection using two algorithms of Image Processing techniques. Leukemia is a group of blood cancers that usually begins in the bone marrow and is caused by a high number of abnormal blood cells.
We can summarize our contribution in the following points:
• Prepare a set of images to be used as a data set.
• Extracting a set of features from the cells of the data set (CKC Algorithm).
• Implementing expert knowledge in detecting Leukemia cancer and categorizing images (data set) accordingly (ALLS Algorithm).
• Use Weka 1 to categorize the data set.
• Compare between all methods using a set of related measures.
A. Color K-means clustering (CKC) :This algorithm is based on analyzing several microscopic images of white blood cells to detect the presence or absence of Leukemia using these steps:
1-acquire digital data microscopic test images.
2-Image Pre-processing and morphological operations
3-Image segmentation using color K-means clustering algorithms
4-features extraction
5-image classification
Acute Lymphoblastic Leukemia Subtypes (ALLS) :In this algorithm, we relied on the visible characteristics in the image of the blood sample for the three types: L1, L2 and L3 as an expert recognizes them. To do this, we investigated a set of features with a pathologist who illustrates to us what features he searches for inside an image to decide if the image is infected or not. Type L1 is distinguished by the presence of similarities between cells in terms of size and shape, while L2 is characterized by the presence of small nuclei inside the nucleus and irregular cell shapes. Finally, L3 is most distinguished by the presence of vacuoles in the cytoplasm.
The following steps are implemented:
1- Image Pre-processing
2-Nuclei Separation
3-Cytoplasm Separation
4-Morphological operations on segmented nuclei and cytoplasm
5- features extraction