Detection of Tuberculosis using a Multi-model Classification Approach on CXR Images

Tuberculosis is known as one of the most dangerous infectious diseases in the world. TB is treatable with proper diagnosis and preventive care. They are known as an” air-borne disease,” as they spread through sneezes and coughs. Usually, sarcoidosis, pneumonia, and lung cancer are misdiagnosed as TB or vice versa. TB diagnosis can be done with chest X-rays, sputum tests, blood tests, and the tuberculin skin test (TST). Generally, in clinical practice, CXRs are used by radiologists for the detection of TB. But in low-resource countries, there is also a lack of trained radiologists, which leads to improper medication and an increased fatality rate. To overcome this, many models have been developed with the help of machine learning and deep learning to automate the detection of tuberculosis using CXRs. The dataset used is a combination of two publicly available datasets: the Montgomery dataset and the Shenzhen dataset. This research is based on experimental processes such as resizing, filtering, feature extraction, and several classification algorithms that include both ML and DL. Then more than one classification algorithm is used to compare the model outcome with their respective accuracies. The algorithms with higher accuracies were combined together to perform ensemble technique. This enhances the accuracy that will be used to develop a model in the future.