Deep Learning AI effectively diagnoses Tuberculosis

Doctors from the Thomas Jefferson University Hospital in Philadelphia have developed highly accurate artificial intelligence programs to identify tuberculosis from X-rays of patients’ chests. The new method could be implemented to diagnose tuberculosis in remote areas lacking expert radiologists.

Co-author of the study, Paras Lakhani, MD, said: “An artificial intelligence solution that could interpret radiographs for presence of tuberculosis in a cost-effective way, could expand the reach of early identification and treatment in developing nations.”

The computer programs use ‘deep learning’, a type of artificial intelligence which can be taught how to perform tasks based on existing data. The researchers trained two different programs, AlexNet and GoogleNet, to diagnose tuberculosis by showing them more than one thousand chest X-rays from healthy and sick patients. When used together, the programs could accurately diagnose tuberculosis in 96 percent of the cases.

“In the past, other machine learning approaches could only get to a certain accuracy level of around 80 percent,” Dr Lakhani added. “However, with deep learning, there is potential for more accurate solutions, as this research has shown.”

Tuberculosis is an infectious disease that normally affects the lungs. Although it can be treated, it is still among the top ten causes of death worldwide, according to the World Health Organization. Last year, about 10.4 million people got tuberculosis, and 1.8 million of them died. Now, artificial intelligence can be implemented to improve screening and lead to more cases being treated successfully.

The researchers are working on improving the accuracy of their programs even further by training the computers with more X-ray scans and new deep learning methods.

Bruno Martin is studying for an MSc in Science Communication at Imperial College London

Banner image: chest x-ray, AkeSak

Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *