Performance Evaluation of Binary Classification of Tuberculosis through Unsharp Masking and Deep Learning Technique

The latest World Health Organization (WHO) study in 2018 shows that about 1.5 million people died and around 10 million people are infected with tuberculosis (TBC) each year. Moreover, more than 4,000 people die every day from TBC. Important work can be found in automating the diagnosis by applying techniques of deep learning (DL) to the medical image. DL requires a large number of high-quality training samples to reach better performance. Due to the low contrast of TBC x-ray images, the image obtained is poor in quality. Our work assesses the effect of image enhancement on the performance of the DL technique based on this problem. An image enhancement algorithm will highlight the overall or local characteristics of the images, and highlight some interesting features. Specifically, an image enhancement algorithm called Unsharp Masking (UM), is evaluated. The enhanced image samples are then fed to the pre-trained ResNet model for transfer learning. In a TB image dataset, we achieve 88.69% and 96.15% of classification accuracy and AUC scores, respectively. All the results are obtained using the Shenzhen dataset which is available in the public domain.