Online sequential extreme learning machine for classification of mycobacterium tuberculosis in ziehl-neelsen stained tissue

The application of image processing and artificial intelligence for computer-aided tuberculosis (TB) diagnosis has received considerable attention over the past several years and still is an active research area. Several approaches have been proposed to improve the diagnostic performance in term of diagnostic accuracy and processing efficiency. This paper studies the performance of a recent training algorithm called Online Sequential Extreme Learning Machine (OS-ELM) for detection and classification of TB bacilli in tissue specimens. The algorithm is used to train a single hidden layer feedforward network (SLFN) using a set of data consists of simple geometrical features, such as area, perimeter, eccentricity and shape factor as feature vectors. All of these features are extracted from tissue images which consist of TB bacilli and further classified into three types; TB, overlapped TB and non-TB. Promising result with 91.33% of testing accuracy has been achieved for the OS-ELM using sigmoid activation function and 40-by-40 learning mode.