Detecting emergent findings in x-rays has just become easier and faster thanks to Dr. Alvin Rajkomar, an assistant professor at the University of California, San Francisco Medical Center, who was able to train a deep learning neural network to automatically detect life-threatening abnormalities in chest X-rays.
When Dr. Rajkomar was a medical resident, he encountered a patient he suspected had a life-threatening pneumothorax. He immediately ordered a stat x-ray, and it so happened that he was standing next to the digital x-ray machine as the x-ray was being taken.” Seeing the finding in real-time, I was able to immediately thrust a needle into his chest to evacuate the air, saving his life. I wondered if we could create an algorithm that could identify emergent findings so that radiographs don’t sit dormant in a database, waiting for a doctor to finally read the study and contact someone to take action,” said Dr. Rajkomar.
In order to train this algorithm to detect life-threatening abnormalities, Dr. Rajkomar and his team had to collect many radiology images. They also used more than 1 million color images from the ImageNet public database (including images of everyday objects), and then retrained the network by showing it a portion of the photos in grayscale.
In the end, by mixing thousands of radiology images with millions of random images, the team was able to create a smart algorithm that is able to detect alarming abnormalities in chest x-rays.
The team now hopes they will be able to generate algorithms that will allow the creation of immediate and automatic red flags of radiographs with critical findings. This would enable better healthcare in general, but most importantly, patients with life-threatening conditions would no longer need to wait.
Digital Trends (http://www.digitaltrends.com/computing/cat-photos-xrays/)