Tuberculosis is a contagious disease caused by Mycobacterium Tuberculosis and is one of the leading causes of death worldwide, especially in third world countries such as Uganda. There are several ways to diagnose Tuberculosis but of these, sputum smear microscopy is the commonest method practised in low and medium income areas. However, this method can be error prone and also requires trained medical personnel who are not always readily available because of their few numbers in developing countries. It has therefore become apparent that automating the diagnosis process would go a long a way in alleviating this problem. In this research, we use two pre-trained Convolutional Neural Networks, VGGNet and GoogLeNet Inception v3, to diagnose tuberculosis from Ziehl-Neelsen stained sputum smear microscopic images. These networks are used in three di↵erent scenarios, namely, fast feature extraction without data augmentation, fast feature extraction with data augmentation and fine-tuning. A dataset of 148 images from multiple sources and with di↵erent backgrounds was used. VGGNet achieved accuracies of 76.7%, 80% and 79.6% for the 3 scenarios while Inception v3 achieved accuracies of 86.7%, 77.2% and 76.8% respectively. Additionally, in all three scenarios, Inception v3 computed faster than VGGNet.

show more