<![CDATA[AML is one type of cancer of the blood and spinal cord. AML has a number of subtypes including M0 and M1. Both subtypes are distinguished by the dominant blast cell type in the WBC, the myeloblast cells, promyelocyte, and myelocyte. This makes the diagnosis process of leukemia subtype requires identification of blast cells in WBC. Automatic blast cell identification is widely developed but is constrained by the lack of data availability, and uneven distribution for each type of blast cell, resulting in problems of data imbalance. This makes the system developed has poor performance. This study aims to classify blast cell types in WBC identified AML-M0 and AML-M1. The method used is divided into two stages, first pre-processing, image segmentation and feature extraction. The second stage, perform resample, which is continued over sampling with SMOTE. The process is done until the amount of data obtained is relatively the same for each blast cell, then the process of elimination of data duplication, randomize, classification and performance measurement. The validation method used is k-fold cross-validation with k=10. Performance parameters used are sensitivity, specifyicity, accuracy, and AUC. The average performance resulting from classification of cell types in AML with Random Forest algorithm obtained 82.9% sensitivity, 92.1% specificity, 89.6% accuracy and 87.5% AUC. These results indicate a significant improvement compared to the system model without using SMOTE. The performance generated by reference to the AUC value, the proposed system model belongs to either category, so it can be used for further stages of leukemia subtype AML-M0 and AML-M1.]]>
Copyrights © 2019
