Three Channels for Gray Level Co-occurrence Matrix (GLCM) to detect Acute Lymphoblastic Leukemia (ALL) Images
Arif Muntasa, Muhammad Yusuf · 2020
Leukemia is one of the dangerous diseases causing the death of people. Based on our literature review, numerous methods have been utilized for leukemia detection with various performances. Therefore, this research aims to employ the three channels for GLCM to detect ALL based on the Lymphoblastic Leukemia Image Database (ALL-IDB2) dataset. This research has some novelties. Firstly, the proposed method related to segmentation has separated the main object and background and removed the noises. Secondly, Feature extraction of the proposed method has depicted the object in three-dimensional form, so that the object can be visualized in more details than one-dimensional. Therefore, this research contributes by proposing a novel method of GLCM to detect ALL. We have employed 260 images of ALL-IDB2 and conducted twenty experiments for each scenario. Our proposed method has proven that three channels of GLCM and testing sets using Manhattan and Euclidean Distance achieved 91.54% of maximum accuracy. It has better accuracy than a combination of the GLCMshape-based feature-SVM classifier method. Furthermore, we have demonstrated that the quantity of training sets has given an impact on accuracy performance. Our experiments show higher numbers of the training sets obtained higher accuracy. Besides accuracy, we also calculated the false-negative and false-positive. The average of the false-negative results contributes to stability. It means that the difference of the results among scenarios is insignificant and almost similar. Meanwhile, false-negative tends to decrease in percentage. This is inversely proportional to the increase in the amount of the training set used. Additionally, the use of large amounts of training sets can reduce misclassification for the false-negative rate.