Tuberculosis with Algorithms to detect

0
244

Not enough x-ray equipment and not enough Doctors that can diagnose tuberculosis. This is the Situation in South Africa. Want to improve young scientists from Germany.

A mobile clinic currently drives through KwaZulu Natal, to investigate about 120,000 people on tuberculosis. The lung disease is highly contagious and can be fatal. You must be detected early and treated. That is precisely the problem in South Africa. Therefore, the African Institute for health research in Durban has initiated a large-scale Initiative for the detection of tuberculosis cases.

X-rays are only the first step in the investigation of tuberculosis. Finally, someone must evaluate the images, so a tuberculosis recognize. But there is a lack of Doctors who can analyze the x-rays.

Algorithm detects tuberculosis

Support comes from the Hasso-Plattner-Institute in Potsdam. There, scientists are working on a computer program designed to simplify the diagnosis of tuberculosis and speed up. Christoph Lippert is head of the chair for digital health and machine Learning on the HPI. “To be able to the Algorithms, we need x-rays of people with tuberculosis have or have had and of people without tuberculosis,” explains Lippert.

The x-ray pictures taken in South Africa, show different stages of the disease. If the program works, be able to recognize the Deep-Learning Algorithms to these differences. A further aim is, says Lippert, that the Algorithms show exactly where the tuberculosis occurs in the lungs. “We program the Computer so that you can recognize the patterns that suggest a case of tuberculosis.”

The tuberculosis studies are mainly carried out in remote villages

The System is still learning

For almost a year, the project is running. A Phase was completed last year. The x-ray images of 1200 people were evaluated. “Our goal is, of 19,500 people in this study,” explains Lippert. The x-ray images will be anonymized and sent from South Africa to Potsdam. Each month, the scientists can get access to additional databases, and new recordings. “The results that are created by using the Algorithms, should be the assessment of the radiologists were compared. Can bring his experience,” says Lippert.

Ultimately must also be tested, how reliable is the data of the computer program. There is much that is not yet Mature. The Algorithms must be adapted to the respective groups of people. “If you would feed the Algorithms, for example, only with data of Europeans, then the Computer tuberculosis among Europeans learn to find, but not for Africans,” says Lippert. Therefore, ethnic differences need to be taken into account. This project is exclusively focused on South Africa.

Tuberculosis is highly contagious

Yet it happens that the Algorithms have overlooked cases of Tuberculosis. But gradually the System will be better. Conventional Tests, however, it is not. If an algorithm shows that there is a x-ray signs of tuberculosis, must be made for additional laboratory tests.

Christoph Lippert hopes that the tuberculosis-a computer program is soon to be about 90 percent correct. Until then, be perfected, the Algorithms until they can be used in mobile clinics. By the end of 2019, it will be ready.