Computerised Tuberculosis Detection Using Artificial Neural Network

Tuberculosis (TB) is an infectious disease caused by the bacillus Mycobacterium tuberculosis, which typically affects the lungs. Diagnosing TB is still a major challenge. The definitive test for TB is the identification of Mycobacterium tuberculosis in a clinical sputum or pus sample, which is the current gold standard. However, it may take several months to identify this slow-growing organism in the laboratory. Another techniques are sputum smear microscopy, skin tests based on immune response, but they are not always reliable. In an effort to reduce the burden of the disease, develop a computerised tuberculosis detection using artificial neural network. In this thesis, a new method for detecting normal and abnormal TB CXR by using Artificial Neural Network (ANN) classifier is presented to improve the accuracy and performance of the system. An automated system that detect the TB from a set of CXR, the basic procedure behind this system is training and testing. Firstly extract the lung region using an optimization method based on graph min cut, then find the features of the lung region of all normal and abnormal CXR. For each normal and abnormal CXR have object oriented inspired features and CBIR based features. The object oriented inspired features include intensity histogram, gradient magnitude histogram, HOG and LBP. The CBIR based features are colour, edge, fuzzy histogram, Hu moments and autocorrelation. Each feature descriptor is quantized into 8 bin histograms overall number of features is thus 9*8=72. So each input image has a 72 bin feature vector. This 72 bin feature vector of all CXR is one of the input of the ANN classifier for training purpose. During testing extract the lung region and find the all feature vector of the input test CXR, this feature vector is the second set of input to the ANN classifier, which enables the region of all normal and abnormal CXR. For each normal and abnormal CXR have object oriented inspired features and CBIR based features. The object oriented inspired features include intensity histogram, gradient magnitude histogram, HOG and LBP. The CBIR based features are colour, edge, fuzzy histogram, Hu moments and autocorrelation. Each feature descriptor is quantized into 8 bin histograms overall number of features is thus 9*8=72. So each input image has a 72 bin feature vector. This 72 bin feature vector of all CXR is one of the input of the ANN classifier for training purpose. During testing extract the lung region and find the all feature vector of the input test CXR, this feature vector is the second set of input to the ANN classifier, which enables the CXR to be classified normal or abnormal using artificial neural network.