According to a writeup in Radiology Business, a novel artificial intelligence (AI) system is able to analyze x-rays in only 10 seconds, which is 120 times faster than the average doctor, who takes around 20 minutes to do the same.
The technology was initially developed at Stanford and is called CheXpert. It was found, through a study run in conjunction by Intermountain Healthcare and Stanford, to be able to swiftly and accurately analyze chest x-rays from emergency room (ER) patients who have pneumonia; specialists feel that CheXpert’s capabilities could have a twofold benefit: reduce wait times in the ER and accelerate treatment.
When a patient presents to the ER with symptoms that point to pneumonia, they typically undergo an X-ray, after which their images are put in a queue with others waiting to be analyzed by radiologists with jam-packed schedules. This takes an average of 20 minutes, but can sometimes take significantly longer, which delays the initiation of necessary antibiotic treatment.
The Stanford Machine Learning Group taught the AI by inputting 188,000 images into the system, instructing it on what is and isn’t pneumonia. They then added an additional 7,000 images from Intermountain in order to “fine-tune the model.”
For this study, radiologists at Intermountain categorized close to 500 chest images as “unlikely,” “unlikely-uncertain,” “likely-uncertain,” or “likely” to have pneumonia. These experts disagreed amongst themselves about half of the time; on the other hand, CheXpert’s performance was on par with that of the quintessential medical specialist.
Nathan Dean was the study’s lead investigator. He’s also the section chief of pulmonary and critical care medicine at Intermountain. In a statement, he said that “CheXpert is going to be faster and as accurate as radiologists viewing the studies. It’s an exciting new way of thinking about diagnosing and treating patients to provide the very best care possible.”
In the future, the investigators hope to test the AI out in a liver ER setting at Intermountain. They plan to do so this fall.
