A Blog by Jonathan Low

 

Jan 23, 2019

AI System Analyzes and Prioritizes XRays By Degree Of Urgency

A matter, literally, of life and death. JL

Kyle Wiggers reports in Venture Beat:

The electromagnetic scans account for 40% of all diagnostic imaging worldwide. In the U.K. alone, there are 330,000 x-rays at any time that have waited more than a month for a report. The system can prioritize x-rays.“There are no systematic, automated ways to triage chest x-rays and bring those with critical findings to the top of the pile.” A computer vision algorithm was trained using labeled images to predict priority from visual information only, not text. When tested, the AI system sort (ed) abnormal x-rays from normal with “high accuracy.” X-rays with “critical” designations received a radiologist opinion in 2.7 days, compared with the current 11.2-day average.
Radiologists are drowning in x-rays. The electromagnetic scans account for a whopping 40 percent of all diagnostic imaging worldwide, and in the U.K. alone, there are an estimated 330,000 x-rays at any given time that have waited more than a month for a report.
Fortunately, artificial intelligence (AI) promises to shrink the backlog substantially. In a new study published in the journal Radiology, scientists at the University of Warwick describe a system that can automatically prioritize x-rays, picking out scans in urgent need of attention.
“Currently there are no systematic and automated ways to triage chest x-rays and bring those with critical and urgent findings to the top of the reporting pile,” Giovanni Montana, who coauthored the study, said in a statement.
The researchers sourced a database of 470,388 adult chest x-rays that had been stripped of identifying information and annotated. Specific abnormalities visible on each x-ray were labeled and fed to a natural language processing system, which used them to categorize each scan as “critical,” “urgent,” “non-urgent,” or “normal.”
Next, the x-rays were fed into a computer vision algorithm trained using the labeled images to predict priority from visual information only — not text. When tested using an independent set of 15,887 images, the AI system was able to sort abnormal x-rays from normal scans with “high accuracy.”
The team’s simulations show that x-rays with “critical” designations received a radiologist opinion in 2.7 days on average, compared with the current 11.2-day average. They leave to future work training the system on a larger sample size, and deploying “more complex” algorithms that might improve performance.
“The initial results reported here are exciting as they demonstrate that an AI system can be successfully trained using a very large database of routinely acquired radiologic data,” Dr. Montana said. “With further clinical validation, this technology is expected to reduce a radiologist’s workload by a significant amount by detecting all the normal exams so more time can be spent on those requiring more attention.”
It’s not the first AI system that can recognize abnormalities from chest x-rays, it’s worth noting.
Qure.ai, an AI health care startup headquartered in Mumbai, secured CE certification in Europe last year for qXR, a chest x-ray product that can identify 15 of the most common chest x-ray abnormalities. And in a paper published in the journal Nature Biomedical Engineering earlier this month, researchers at Massachusetts General Hospital in Boston describe a deep learning algorithm that can detect acute intracerebral hemorrhages, or ICHs, with a high degree of accuracy.

1 comments:

Post a Comment