vRad trains AI to help identify intracranial hemorrhaging

Teleradiology company vRad (Virtual Radiologic) announced this week that it has developed a new algorithm that can review CT images and look for potential intracranial hemorrhaging (IH).

The breakthrough came as a result of the company’s continued commitment to advancing technology through “Deep Learning” artificial intelligence (AI). vRad announced an ongoing collaboration with AI company MetaMind back in June.

vRad is working to gain the proper regulatory approval by the end of 2015.

“Once approved and implemented into vRad’s telemedicine platform, the deep learning algorithm can immediately identify—or ‘recall’—a potential IH and automatically prioritize that patient’s study so that it is reviewed by the most appropriate radiologist more quickly,” Shannon Werb, vRad chief information officer, said in a statement. “While many are talking about machine-assisted diagnostics, like IBM’s Watson and Google, vRad will be the first to leverage deep learning in a real-time practice environment. Our radiologists will have an additional tool to help maximize the speed at which they can get ‘eyes on images’ to determine if there is a diagnosis of IH. This milestone shows how the right clinical and technical collaboration can empower radiologists, increase their time being doctors and diagnosticians—and ultimately improve patient outcomes.”

vRad physicians were able to train the AI software to “flag” a patient’s study for potential IH and automatically make it a priority. With vRad interpreting approximately 90,000 head CT images in a month, flagging possible cases of IH helps patients who need immediate care get treatment as soon as possible.

“The combination of deep learning technology with our large clinical datasets and expertise serves as a model of how cutting-edge technology can be used to complement—not supplant—clinicians and improve care,” Benjamin Strong, MD, vRad chief medical officer, said in the statement. “We are encouraged by the algorithm’s precision performance to date in the test environment and will continue to focus on continuous improvement of the algorithm’s recall levels of IH so that once it is implemented, we can optimize the study distribution workflow. We look forward to extending deep learning to additional life-threatening abnormalities so vRad’s clinicians can deliver high-quality, accurate diagnoses to referring physicians as quickly as possible.”

The company estimates that this new process could help identify more than 5,000 cases of potential IH in 2016.