# TBNet:Pulmonary Tuberculosis Diagnosing System using Deep Neural Networks

**Authors:** Ram Srivatsav Ghorakavi  
**arXiv:** [1902.08897](https://arxiv.org/abs/1902.08897)  
**Date:** 2019-02-24  
**Summary page:** https://www.summarizepaper.com/en/paper/1902.08897v1/

## Key points

- Challenges associated with detecting tuberculosis
- Lack of proper technology and reliance on outdated methods
- Existing methods for tuberculosis detection are not up to commendable standards
- Proposal of a deep-learning based system for diagnosing tuberculosis using Chest X-rays
- Acceleration of the tuberculosis diagnosis process, eliminating the need for time-consuming sputum-based testing methods
- Utilization of state-of-the-art ResNet architecture combined with traditional robust features like Haar and LBP
- Significant enhancement in the rate of tuberculosis detection compared to previous methods
- Training of the network using a publicly available pulmonary chest X-ray dataset
- Development of an intuitive feature-based data augmentation pipeline to improve accuracy
- 10% boost in performance accuracy observed by utilizing these techniques
- TBNet as a deep neural network-based system for diagnosing pulmonary tuberculosis
- Efficient and accurate alternative to traditional diagnostic methods
- Potential to revolutionize tuberculosis detection and improve healthcare outcomes worldwide

## Summary

In the paper titled "TBNet: Pulmonary Tuberculosis Diagnosing System using Deep Neural Networks," authored by Ram Srivatsav Ghorakavi, the author addresses the challenges associated with detecting tuberculosis. This deadly infectious disease is prevalent worldwide and its early detection is hindered by the lack of proper technology and reliance on outdated methods. Existing methods for tuberculosis detection are not up to commendable standards as they rely on unnecessary features and are obsolete for reliable healthcare technology. To overcome these limitations, Ghorakavi proposes a deep-learning based system that diagnoses tuberculosis by analyzing important features in Chest X-rays along with original chest X-rays. By employing this system, the process of tuberculosis diagnosis can be accelerated, eliminating the need for time-consuming sputum-based testing methods such as Diagnostic Microbiology. Unlike previous methods mentioned in the paper (Kant et al., 2018; Melendez et al., 2016), which do not meet satisfactory levels of accuracy, TBNet utilizes state-of-the-art ResNet architecture (He et al., 2016) combined with traditional robust features like Haar (Viola and Jones, 2005; Viola and Jones, 2001) and LBP (Ojala et al., 1994; Ojala et al., 1996). The author observed that this approach significantly enhances the rate of tuberculosis detection to a highly satisfactory level. The author trained their network using a publicly available pulmonary chest X-ray dataset (Jaeger et al., 2014). However, due to its small size, this dataset alone does not provide optimal accuracy. To address this issue, an intuitive feature-based data augmentation pipeline was devised. This approach helps the deep neural network focus its training on tuberculosis-affected regions, making it more robust and accurate compared to conventional methods that rely on procedures like mirroring and rotation. The author observed a 10% boost in performance accuracy by utilizing these simple yet powerful techniques. In summary, Ghorakavi's paper presents TBNet, a deep neural network-based system for diagnosing pulmonary tuberculosis. By leveraging important features in Chest X-rays and employing data augmentation techniques, this system offers an efficient and accurate alternative to traditional diagnostic methods. The proposed approach has the potential to revolutionize tuberculosis detection and improve healthcare outcomes worldwide.