Open AccessAn Automated Intelligent Identification and Counting System Procedure for Tuberculosis
Author(s) -
Fatin Atiqah Rosli,
Mohd Yusoff Mashor,
Siti Suraiya Md Noor
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1372/1/012079
Subject(s) - artificial intelligence , pattern recognition (psychology) , bacilli , computer vision , tuberculosis , sputum , computer science , feature (linguistics) , image processing , histogram , medicine , image (mathematics) , pathology , biology , linguistics , philosophy , genetics , bacteria
Tuberculosis (TB) is an infectious disease caused by Mycobacterium Tuberculosis or TB Bacilli. Currently, the classification of TB bacilli is carried out by microbiologist by using Ziehl-Nielsen (ZN) stained smear sputum slide under a light microscopy. However, the manual evaluation is time-consuming and lead to slow decision. Furthermore, the sensitivity is less due to incline of human error which lead to inaccurate conclusion. Therefore, this study proposes an intelligence identification and counting system to detect the presence of TB bacilli in the ZN-stained smear sputum image. This system is designed to identify the presence of TB bacilli and count the number of TB bacilli by applying digital image processing and artificial intelligence techniques. In image acquisition, there are 70 samples images of ZN-stained smear sputum image were collected from Hospital Universiti Sains Malaysia (HUSM) Kubang Kerian, Kota Bharu, Kelantan, Malaysia. The image processing technique consists of contrast enhancement, segmentation, and feature extraction. The contrast of original image was enhanced by the combination of global enhancement, local enhancement and Contrast Limited Adaptive Histogram Equalization (CLAHE). Then, the enhanced image was segmented using color thresholding and the features were extracted consists of on 18 colour features, 15 shape features and 5 texture features. Afterward, the features underwent feature selection to select the relevant features by using Neighborhood Component Analysis (NCA) and ReliefF Analysis. The study showed that there are relevant features were chosen by ReliefF at feature weight more than 0.004 including (8 colour features, 11 shape feature and 3 texture features) for improving the performance and accuracy of Multilayer Perceptron (MLP) trained by Scaled Conjugate Gradient (SCG). For classification process, MLP, k-Nearest Neighborhood (k-NN) and Support Vector Machine (SVM) are used with 6 folds cross-validation. It was found that MLP has the highest of accuracy, sensitivity and specificity with 93.8%, 93.4% and 94.1% respectively.