Grant: CIHR Fellowship

Although it is outside the global spotlight, tuberculosis (TB) rages on as one of the top infectious disease killers world-wide, with 1.5 million people dying from TB each year. Despite this staggering mortality, every year an estimated 3 million people who fall ill with TB never receive a proper case notification; as a result, they never receive correct treatment. Conventional TB tests are expensive, as they require trained personnel to run them, or are slow, so those individuals who do seek testing are often lost to follow-up as they await results. The COVID-19 pandemic has worsened this problem, as lockdowns, resource diversion, and patient fears of accessing healthcare have led to a drop-off in TB case notifications in 2020 and 2021. TB testing solutions that quickly produce results and require minimal hands-on time by healthcare workers could be useful tools for finding the missing TB cases, but first they need to be properly validated. Here, we propose a validation study of an artificial intelligence application called Computer-Aided Detection for TB (CAD4TB). CAD4TB can read digital chest x-rays without help from a human in 30 seconds. We plan to evaluate CAD4TB's ability to detect TB on digital chest x-rays in the rural province of Ca Mau, Vietnam. As well as determining CAD4TB's diagnostic accuracy, we will also investigate its time-to-test-result and the proportion of TB cases it identifies compared to conventional TB tests. Compared to the conventional tests, we expect that the CAD4TB may be less accurate and may detect slightly fewer TB cases, but that the time-to-test-result will be much shorter. Using rapid and portable technology like CAD4TB to test for TB would be convenient for healthcare workers and patients, especially those living in very remote areas. Easy, rapid testing solutions will be necessary to find the millions of missing TB cases, particularly as the on-going COVID-19 pandemic continues to stretch healthcare systems to the limit.