With the CAD4TB software, computers can automatically analyze chest radiographs for signs of tuberculosis.
We introduce the project with a BBC report on a big health care project in Ghana, where a large number of hospitals are equipped with digital x-ray and our CAD4TB software:
Chest radiography has always been widely used to find TB. Digital chest radiography has made x-ray cheaper and easier to use. No films, chemicals and water are needed to produce a digital radiograph and automatic exposure control ensures high image quality, and images are instanly available. The cost per exam, once a digital unit is operational, is near zero. Digital chest radiography therefore has a large potential to be used in TB case finding and prevalence surveys, as an adjunct test, next to molecular testing. Molecular diagnostics for TB, such as the GeneXpert MTB/RIF test are an exciting new development; they can make a definitive diagnosis of TB in two hours. However, these tests are still much more expensive and time consuming than radiography. Radiography can therefore act as a filter: TB suspects undergo symptom screening and chest radiography, and those with symptoms and/or abnormalities in the chest radiograph undergo further testing.
One issue remains: the lack of human expert readers in countries with a high burden of TB. CAD4TB eliminates this problem by automating the reading process.
Our work in developing software to detect signs of tuberculosis started in 1996 when digital x-ray machines were first entering the market. We work with Delft Imaging Systems, famous as the inventor of the Odelca camera from the 1960s, one of the most widely used systems in x-ray TB screening in the world. Our project resulted in a prototype computer-aided detection system for TB in 2001. At that time however, digital x-ray was not yet widely adopted for TB case finding. In 2007, we attracted a research grant and established the CAD4TB project with partners in South Africa, the University of Cape Town Lung Institute, and in Zambia, the NGO Zambart.
|Left: one of the first digital units that was installed for the CAD4TB project, installed in Kanyama Clinic, in Lusaka, Zambia. In this clinic, more than 10,000 TB suspects are seen every year. Right: mobile digital X-ray unit used in the Gambia for its TB prevalence survey.
The first CAD4TB beta prototype (CAD4TB v0.01) was delivered in January 2010, and field tested in 2010. In 2011, the first official prototype version was released (CAD4TB v1.08), notably improved by training the system with a much larger number of images. Every year a new version of the software has been released: in 2012 v2.09, in 2013 v3.07, and in 2014 v4.10. Version v4.10 was the first version to receive a CE label. In 2016 version 5 was released and early 2018 version 6 is expected, based on deep learning. Algorithm development behind CAD4TB is now carried out at Thirona, a spin-off company from the Diagnostic Image Analysis Group.
CAD4TB is commercially available via Delft Imaging Systems and has over 150 installations in 24 countries worldwide. Our current research focuses on extending the capabilities of the software to do more than outputting a single number related to the probability the subject has TB. We expect that in the near future, the software can recognize and differentiate different types of abnormalities in the chest radiographs.