Pneumonia is one of the most common infections that can lead to hospitalization, and depending on the population, there are often variations in imaging diagnosis accuracy. A group of Stanford University researchers are working to maximize the rate of correct diagnosis with CheXNet, a machine learning algorithm that can detect pneumonia from chest x-rays better than a radiologist, according to a non-peer-reviewed paper.
The radiology community is embarking upon an exciting era of AI-driven diagnosis. Back in August, researchers at the Thomas Jefferson University Hospital developed a way to use deep learning to identify TB on chest exams using images from the National Institutes of Health (NIH). Similarly, this Stanford team used 100,000 images provided by the NIH to create an algorithm that “exceeds the performance of radiologists in detecting pneumonia from frontal-view chest X-ray images.”
The researchers compared CheXNet to the work of four academic radiologists and found that the algorithm was more accurate than the radiologists in terms of sensitivity and specificity. They also evaluated 14 different pneumonia pathology classes, including hernia, edema, fibrosis, and emphysema and found that the algorithm produced fewer false positives and false negatives than the work that was done by the NIH researchers.
“With automation at the level of experts, we hope that this technology can improve healthcare delivery and increase access to medical imaging expertise in parts of the world where access to skilled radiologists is limited,” write the researchers.