AI on par with clinicians in the detection of Tuberculosis

Tuberculosis is a leading cause of death worldwide. It is a huge threat to public health translating into a crisis for those with deficient immunities e.g., for patients with HIV. According to the WHO for the year 2019 reported tuberculosis cases reached 10 million with 1.4 million losing their lives to the infection. Unsurprisingly, the burden of tuberculosis is not evenly distributed among countries with developing states being affected the most largely owing to limited health expertise and lack of medical equipment. The emergence of multi-drug-resistant tuberculosis has exacerbated the problem rendering the control of the disease more challenging. Especially when it comes to the early identification and diagnosis of resistant strains.

Medical professionals worldwide have been extending their efforts to define the solutions to the fight against tuberculosis. Significant improvements have been made in the direction of lab diagnostics as well as standardizing treatment schemes. In addition to these, more innovative solutions such as computer-aided diagnostics involving artificial intelligence have emerged, overcoming the drawbacks of existing systems. With AI in play, health professionals have access to more sophisticated algorithms that can aid more focused diagnosis and detect a more extensive variety of TB features. As a result, clinicians can deliver their work with greater precision.

AI systems can detect tuberculosis in chest X-rays using deep learning. To train the algorithm, researchers used over hundreds of thousands of chest X-rays of many patients from different countries. As a result, AI was taught to categorize images into 3 options: normal, TB, and abnormal but not TB. This way, automated detection of TB cases and sometimes even other health conditions became possible. There are several different algorithms designed for the detection of TB including the one by Google. WHO recommends the use of AI to maximize the use of resources.

While previous machine learning solutions were as accurate as 80%, with deep learning AI performance is on par with a radiologist. Having such strength, AI can be applied in large-scale TB screening programs, especially in those areas where radiologic resources and health personnel are limited. The automated process provides a cost-effective option as AI can save resources by filtering those patients that later require to receive more expensive sputum tests or NAAT to confirm the diagnosis. Earlier detection of the disease can also dictate better health outcomes for infected patients with shorter treatment times.

Stay tuned to Risalat Social Pages 

LinkedIn | Twitter | Facebook | YouTube