Tuberculosis is the leading cause of deaths worldwide from an infectious disease, second only to COVID-19. Vietnam ranks 11^th among the world's 30 highest TB burden countries. However, according to the National TB Control Program, about 62% of people with active, transmissible TB disease would have been missed if they had been screened using the standard diagnostic algorithm, which relied on verbal screen for prolonged cough.

With the determination to improve the current situation of TB diagnosis and treatment to save more lives, VinBrain and FIT have collaborated to write and release a white paper on "Utilizing AI in reading Chest X-rays for TB screening in Vietnam".

The white paper provides an overview of the current situation of TB diagnosis and screening in Vietnam, and introduces the AI model developed by VinBrain which helps improve the accuracy in reading Chest X-rays for TB diagnosis and screening. The solution has been developed via several steps including system design, data collection and labeling, pre-processing and model training, model evaluation, model deployment and monitoring.

To develop the AI model, VinBrain utilized the Cloud Computing platform of Microsoft Windows Azure. On Windows Azure, VinBrain has built, deployed, and managed the model applications through Microsoft's global datacenters. In total, more than 500,000 TB X-ray images from Vietnam, India and China were collected. After filtering out the poor-quality images, 185,486 images were cleaned for model training, evaluation and testing. The AI model v1 for TB diagnosis and screening developed by VinBrain (launched on January 5, 2021) achieved a sensitivity of 86% and a specificity of up to 96.1 %.

Tuberculosis Classification Approach

"The white paper on utilizing AI in TB diagnosis and screening has harnessed new tools and technologies to improve the current situation of the diagnosis and treatment of the disease, which has been considered to be 'incurable'. Technological solutions will help Vietnam to achieve its national goal of ending TB by 2030. This not only reflects Vingroup's determination to build a better society, but also affirms its right strategy of investing in industry and high technology," - Mr. Steven Truong, CEO of VinBrain, said.

VinBrain has integrated the AI model for TB diagnosis and screening into DrAid - its "Doctor Assistant". DrAid is an AI-powered product which can detect 21 abnormalities and diseases of heart - lung - bone with an average accuracy of over 88% within 5 seconds, and auto-generate medical reports. Thanks to this integration, the AI model can diagnose TB disease simultaneously with other lung diseases and abnormalities.

Doctors can now use DrAid to detect and screen TB by uploading X-ray images directly from the X-ray machines to the system and receiving diagnostic suggestions by AI.

Chest X-ray (CXR) is an effective and cost-effecient tool for TB diagnostic and classification. Hence, utilizing AI in reading CXR will help find and treat more persons with TB, and thereby cut down community transmission./.