Johns Hopkins Develops AI to Detect COVID-19 in Lung Ultrasounds Using Simulated Training Data
-
Johns Hopkins University highlighted research showing AI can spot Covid in lung ultrasound images by detecting bright, vertical anomalies called B-lines that suggest inflammation.
-
The AI system was trained on a mix of computer-generated images and actual ultrasounds of patients, including some who sought care at Johns Hopkins.
-
Early in the pandemic there were not enough real COVID-19 ultrasound images to fully train the deep neural networks.
-
Researchers taught the AI using a combination of real and simulated ultrasound data to identify COVID-19 abnormalities.
-
The tool is a deep neural network designed to recognize patterns like the interconnected neurons in the human brain.