Detection of pulmonary tuberculosis from chest X-Ray images using multimodal ensemble method

Abstract

Tuberculosis (TB) is one of the deadliest diseases nowadays, and it is caused by the mycobacterium tuberculosis bacterium that generally attacks the lungs. Following artificial intelligence implementation in the field of computer vision, especially deep learning, many computer-based diagnostic systems have been proposed to help detect TB from chest X-Ray images. It can produce better and faster accuracy and consistency of the diagnosis results. However, many radiology applications based on the deep learning method consider only images as input sources. In modern medical practice, non-imaging data from patient medical record history or patient demographics may influence disease detection and provide more data for radiologists to obtain additional insights in a clinical context. This study proposed a multimodal model that uses images and patient demographics to answer the need. The evaluation results show that our approach leads to high accuracy and can improve the area under curve (AUC) value by 0.0213 compared to the unimodal model. Additionally, this model successfully outperformed the previous state-of-the-art multimodal model by a 0.0075 (0.0213 vs. 0.0138, respectively) increase in AUC.