Identification the number of Mycobacterium tuberculosis based on sputum image using local linear estimator

Nur Chamidah,Yolanda Swastika Yonani,Elly Ana,Budi Lestari

doi:10.11591/eei.v9i5.2021

Abstract

Infectious disease caused by infection of Mycobacterium tuberculosis is called tuberculosis (TB). A common method in detecting TB is by identifying number of mycobacterium TB in sputum manually. Unfortunately, manually calculation by pathologists take a relatively long time. Previous researches on TB bacteria were still limited to detect the absence or presence of mycobacterium TB in images of sputum. This research aims are identifying number of mycobacterium TB and determining accuracy of classification TB severity by approaching nonparametric Poisson regression model and applying an estimator namely local linear. Steps include processing of image, reducing of dimension by applying partial least square and discrete wavelet transformation, and then identifying the number of mycobacterium TB by using the proposed model approach. In this research, we get deviance values of 28.410 for nonparametric and 93.029 for parametric approaches and the average of classification accuracy values for 4 iterations of 92.75% for nonparametric and 85.5% for parametric approaches. Thus, for identifying many of mycobacterium TB met in images of sputum and classifying of TB severity, the proposed identifying method gives higher accuracy and shorter time in identifying number of mycobacterium TB than parametric linear regression method.

Highlights

Mycobacterium tuberculosis can cause a direct infectious disease namely tuberculosis (TB).Identification of TB through microscopic observation using sputum smear samples has greatly helped prevent TB disease [1,2,3,4,5,6,7]
In this research, we propose a statistical model approach called Poisson additive nonparametric regression model using local linear estimator to identify how many TB bacteria that are in sputum images of TB patients
Based on 100 observations, we have 2048 predictor variables that is reduced to 5 predictor variables through image processing by using discrete wavelet transformation (DWT) and partial least square (PLS) methods

Summary

Introduction

Identification of TB through microscopic observation (screening) using sputum smear samples has greatly helped prevent TB disease [1,2,3,4,5,6,7]. As a result of the long identification process that requires high accuracy, statistical modeling and software assistance are needed to identify TB disease from sputum samples of patients using processing of images. This process is one of processing digital images that is a discipline of study about digitally techniques to proceed images [8,9,10]. Researchers [11,12,13] used meta analysis, [14] used self organizing map and [15] used

Objectives

Methods

Results