Hybrid multilayered perceptron network trained by modified recursive prediction error-extreme learning machine for tuberculosis bacilli detection

Abstract

In this paper, image processing technique and artificial neural network are used to detect and classify the tuberculosis (TB) bacilli in tissue slide images. The tissue sections consisting of TB bacilli are stained using the Ziehl-Neelsen method and their images are acquired using a digital camera mounted on a light microscope. Colour image segmentation is applied to remove the remove undesired artefacts and background. Then affine moment invariants are extracted to represent the segmented regions. Finally, the study proposes a method that integrates both Modified Recursive Prediction Error (MRPE) algorithm and Extreme Learning Machine, called MRPE-ELM to train Hybrid Multilayered Perceptron (HMLP) network. The network is used to classify the segmented regions into three classes: TB', overlapped TB' and non-TB'. The classification performance of the HMLP network trained by the MRPE-ELM is compared with the HMLP trained by the MRPE algorithm and single layer feedforward neural network (SLFNN) trained by the ELM. The results indicated that the proposed MRPE-ELM has slightly improves the classification performance and reduces the number of epochs required in the training process compared to the MRPE algorithm. © 2011 Springer-Verlag.