Lymphoblast cell morphology identification to detect Acute Lymphoblastic Leukemia (ALL) using various color segmentation

Abstract Acute Lymphoblastic Leukemia (ALL) is a disease that is defined by the uncontrollable growth of abnormal cell of lymphocyte, which is called lymphoblast. ALL patients cannot be left untreated as it can be fatal, and hence, early detection is very crucial for proper treatment suggestion. Conventionally, ALL analysis is manually done, which is time-consuming and highly dependent on the pathologist’s skills. Furthermore, it will be hard to pathologist as the number of sample increases, and also there are other cells inside the blood smear image, which will create confusion to them. Haematology counter is great to assist the process, but unfortunately, the cost is unbearable for some countries. For that reason, in this project, a computer-aided ALL detection with fast response is proposed. The system consists of five main modules which are image acquisition, pre-processing, segmentation, feature extraction and classification. For the process flow, firstly the color space correction based on l*a*b* color space is applied to standardize the color of the input image. Next, WBC segmentation is employed to locate the WBC region and consequently divided it into two parts which are the nucleus and cytoplasm based on a combination of color space analysis and Otsu thresholding. However, the segmented image contains noises and hence, is eliminated by using a combination of morphological filter and Connected Component Labelling (CCL). Then, the feature extraction process is made to study the nature of each individual cell using features derive from color, texture and geometrical properties. Lastly, lymphoblast classification is incurred to categorized the lymphoblast and the non-lymphoblast cell by employing Support Vector Machine (SVM) with a linear kernel function. As mentioned before, WBC segmentation is divided into two parts, and nucleus detection accuracy is higher than cytoplasm detection accuracy, which is 98.87% and 74.12% respectively. The presence of color space correction is analyzed, and the result is better with 96.92% accuracy for the presence of color space correction compared to 93.55% accuracy for without color space correction result. Classification performance is able to achieve 98.72% of accuracy.