A Novel Method for Convolution Neural Network in Deep Learning for Detecting Tuberculosis Disease

Abstract

The common ancient disease widely prevailed in the world which is one of the top 10 main causing of crucial death is widely known as Tuberculosis. The chest x-ray which is the modest way for finding out tuberculosis is the model preparation of the main objective in this paper. Classification is done in this model to achieve high accuracy with attaining normal and abnormal classes infected by Tuberculosis. The techniques of deep neural networks is used in our study of approach in order to obtain and improve high accuracy of the model to meet the goal objective in classifying the new chest x- ray which is given as the input. Pre-processing of images are needed for achieving good accuracy and is obtained through the datasets achieved from both datasets of Shenzhen and Montgomery and together totally there are 800 chest x rays. Data augmentation is done over the 680 training set images and normalized them. It is followed by giving these pre- processed images as inputs to our models for supervised training. Then we performed model testing over the set of 120 images out of 800. In this paper we used two models. baseline CNN model and pre-trained VGG16 model and gave pre- processed images as inputs to these both models and evaluated the models to see which performed accurately better results. On comparing the results using different performance metrics like accuracy rate, specificity, sensitivity, high precision and f1- score. Finally it is depicted through using graphs and tables, where baseline CNN model gave an accuracy of 82% and the model VGG16 gave an accuracy of 90%.