Researchers from Thomas Jefferson University Hospital in Philadelphia are training deep learning models to identify tuberculosis (TB) in an effort to help patients in regions with limited access to radiologists.
TB is one of the top ten causes of death worldwide with nearly two million deaths in 2016. TB can be identified on chest imaging, but the areas where TB is most common lack the expertise needed to diagnose the disease.
“There is a tremendous interest in artificial intelligence, both inside and outside the field of medicine,” said study co-author Paras Lakhani, M.D., from Thomas Jefferson University Hospital (TJUH) in Philadelphia. “An artificial intelligence solution that could interpret radiographs for presence of TB in a cost-effective way could expand the reach of early identification and treatment in developing nations.”
Using CUDA, TITAN X GPUs and cuDNN with the Caffe deep learning framework, the researchers trained their deep convolutional neural networks on 1,007 X-rays of patients with and without active TB. The cases consisted of multiple chest X-ray datasets from the National Institutes of Health, the Belarus Tuberculosis Portal, and TJUH. The datasets were split into training (68.0 percent), validation (17.1 percent), and test (14.9 percent).
Their best model achieved an accuracy of 96 percent – but when an expert radiologist was included in the process, they had a greater accuracy of close to 99 percent.
“We hope to prospectively apply this in a real world environment,” Dr. Lakhani said. “An artificial intelligence solution using chest imaging can play a big role in tackling TB.”