Para brindar el contenido más relevante a los lectores de HealthCare Business News, le pedimos que comparta un poco de información sobre quién es usted (tarda dos segundos y listo).
Registro
Mount Sinai researchers train machine learning tool to understand radiology reports
Dr. Eric Oermann of the Icahn School of Medicine at Mount Sinai believes that the future of radiology will involve artificial neural networks that assist physicians in performing daily tasks such as interpreting imaging.
He and his team are using machine learning techniques including natural language processing algorithms to identify clinical concepts in radiology reports for CT scans. Their research was recently published in the journal Radiology.
“The tools currently available on the market and in development by other groups in academia and industry for the most part employ a similar set of algorithms, with the research tools having a bias toward more advanced techniques in deep learning,” Oermann told HCB News.
Final days to save an extra 10% on Imaging, Ultrasound, and Biomed parts web prices.* Unlimited use now through September 30 with code AANIV10 (*certain restrictions apply)
His research team is what makes this tool different from the others. Many of the team members are practicing physicians and medical students who are competent machine learning researchers.
“I like to think that at Mount Sinai we're starting from the perspective of physicians and attempting to solve medical problems, rather than starting with technical solutions and trying to fit medical problems to them,” said Oermann.
His team is working on training the tool to understand text reports written by radiologists. They created a series of algorithms to teach it certain terminology, including words like colonoscopy and heartburn.
The training data involved 96,303 radiologist reports on head CT exams that were performed at The Mount Sinai Hospital and Mount Sinai Queens between 2010 and 2016.
The team found that the techniques used in this study resulted in a 91 percent accuracy rate. That demonstrates that it is possible to automatically identify concepts in text from the complex domain of radiology.
When asked whether machine learning techniques are something any hospital and radiology practice will have the financial means to deploy, Oermann replied that algorithms are generally cheaper to deploy than most things in health care.
However, he did note that the final cost will largely depend upon how the techniques are deployed.
