Apply Learning Models to recognize the case of Tuberculosis Using X-Ray Images

Tuberculosis is a type of chronic diseases, thus needs long term treatment. Each year, over 2 lakh people in India are affected by TB and on average around 20,000 people are infected every month in each state. According to WHO, primary statistics collated from 84 countries, approximately less than 1.4 million persons received tuberculosis (TB) treatment in 2020 than in 2019, (i.e. 21% decline from 2019). An early detection can significantly shorten the treatment period. In this study, we’ve investigated on a dataset of 4195 images (699 of Normal Cases, 3496 of Tuberculosis Cases). The methodology adopted is different than other available detection algorithms, as here, a dataset is created in (.csv) format from the collection of X-ray Images. The main objective is to train model such that it can detect the presence of Tuberculosis from the image data. Different types of libraries are used like Pillow, scikit-image, sciPy, which can easily extract and transform the image data to csv dataset. The best classification accuracy is 98.06% attained by KNN model. With a truth loss of 0.98% and false loss of 0.11%. Thus, our approach brings a solution to quantify/test classifications models appropriately.