Detection of Acute Leukaemia Cells Using Variety of Features and Neural Networks

Abstract

This paper presents the application of features combination and Multilayer Perceptron (MLP) neural network for classification of individual white blood cells (WBC) inside the normal and acute leukaemia blood samples. The WBC will be classified as either normal or abnormal for the purpose of screening process. There are total 17 main features that consist of size, shape and colour based features had been extracted from segmented nucleus of both types of blood samples and used as the neural network inputs for the classification process. In order to determine the applicability of the MLP network, two different training algorithms namely Levenberg- Marquardt and Bayesian Regulation algorithms were employed to train the MLP network. Overall, the results represent good classification performance by employing the size, shape and colour based features on both training algorithms. However, the MLP network trained using Bayesian Regulation algorithm has proved to be slightly better with classification performance of 94.51% for overall proposed features. Thus, the result significantly demonstrates the suitability of the proposed features and classification using MLP network for acute leukaemia cells detection in blood sample.