A new convolutional network model for detection of tuberculosis disease

Tuberculosis (TB) disease is one of the contagious and dangerous diseases that result in death in many parts of the world. For tuberculosis, specialist radiologists diagnose it by looking at chest X-rays. However, radiologists can sometimes misdiagnose and diagnose because they examine a large number of chest X-rays while making this diagnosis. In this case, computer-assisted analyzes that can make faster and more accurate decisions are required instead of a manual diagnosis. The aim of this thesis is to create a model that will autonomously detect and classify tuberculosis and healthy images from lung x-ray data using artificial intelligence methods. Artificial intelligence is to create machines that work similarly to the working principle of human intelligence and that can continuously improve themselves. With the development of technology, it is widely used in many fields such as voice processing, cyber security, social security, e-commerce, robotics, natural language processing, advice systems, health, military, in order to improve business areas and make life easier with the help of computers. It is frequently used in the diagnosis of many diseases such as cancer, especially in the field of health, in the examinations made by doctors, in the diagnosis and treatment performed with devices. Because it is very difficult for doctors to manually reveal the correct diagnosis and correct diagnosis of any disease in terms of both time and cost. Because correct diagnosis and diagnosis is very important in the treatment of the disease. In this thesis, the solution of this problem is discussed and Convolutional Neural Networks (CNN), which is a subset of artificial intelligence, which is a deep learning method, has been used to perform computer-aided analysis and detection of tuberculosis disease, which has been going on for many years. With the proposed CNN model, easier and faster computer aided analysis was performed. In the first stage of the proposed model, 1000 images obtained from lung x-ray films were preprocessed and labeled as diseased and healthy. In order to diagnose images accurately, quickly and with minimum cost, a different and new network structure was created and important features in the data were determined. The network structure has a deep structure consisting of 3 convolutions, 3 pooling, 1 smoothing, 3 density and 2 dropout layers. In addition, the proposed CNN model, and the Artificial Neural Network (ANN) model, which is widely used in many areas in the literature, were compared in detail using different evaluation metrics such as accuracy, sensitivity, precision and F1-Score. In addition, performance analyzes were carried out for both models. The classification success of the CNN and ANN models used in the thesis was found to be 98.91% and 90.41%, respectively. The proposed CNN model gave more successful results in the correct diagnosis and classification of tuberculosis disease compared to the ANN model. In addition, the suggested VGG19+CNN model was applied to the tuberculosis image data. This model consists of feature extraction and classification steps and includes 2 convolutions, 1 batch, pooling, smoothing and 2 density layers. The hybrid model gave 100% and 99.66% success in training and test images, respectively.