After training two deep-learning models to identify tuberculosis, researchers at Thomas Jefferson University in Philadelphia have gotten their human-free method to nail the disease with 96 percent accuracy, according to a study published online in Radiology.
Paras Lakhani, MD, and Baskaran Sundaram, MD, split 1,007 x-rays of patients with and without active TB into training, validation and test datasets.
They used the cases to train the AlexNet and GoogLeNet deep convolutional neural networks (DCNNs) to distinguish between TB-positive and TB-negative x-rays.
The authors then tested the networks’ accuracy on 150 cases that were excluded from the training and validation datasets.
It was a combination of the two DCNNs that achieved 96 percent accuracy.
“A radiologist-augmented approach for cases where there was disagreement among the classifiers further improved accuracy,” Lakhani and Sundaram report.
In a press release sent by Radiology publisher RSNA, Lakhani suggests the AI-for-TB method might be used to help with screening and evaluation in developing areas of the world that are dogged by TB and short on radiologists.
“In the past, other machine learning approaches could only get to a certain accuracy level of around 80 percent,” Lakhani says. “However, with deep learning, there is potential for more accurate solutions, as this research has shown.”
RSNA has posted the full study for free.
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.