Classification of bone marrow acute leukemia cells using multilayer perceptron network

Abstract

The leukemia classification based on bone marrow samples highly benefits the doctors in the confirmation of leukemia occurrence from blood test. This paper focuses on the classification of bone marrow acute leukemia cells into three groups namely normal, acute promyelocytic leukemia subtype (M3) and other acute leukemia subtypes. The images are implemented with a series of digital image processing technique such as image enhancement, median filtering and feature extraction. Thirteen features are extracted on whole image, inclusive of color and geometrical based features of the cells. Multilayer Perceptron neural network trained using Levenberg Marquardt training algorithm is used for classification purpose. The classification performances are evaluated by comparing the accuracy rate between standard and hierarchical MLP network. Results show that the hierarchical networks have managed to outperform the accuracy of standard network with an average accuracy of 100% on training data and 97.55% on testing data. Results also show that color features play an important role in obtaining good classification performance.