News | Artificial Intelligence | January 31, 2018

Machine Learning Techniques Generate Clinical Labels of Medical Scans

New study’s findings will help train artificial intelligence to diagnose diseases

Machine Learning Techniques Generate Clinical Labels of Medical Scans

January 31, 2018 — Researchers used machine learning techniques, including natural language processing algorithms, to identify clinical concepts in radiologist reports for computed tomography (CT) scans, according to a new study. The study was conducted at the Icahn School of Medicine at Mount Sinai and published in the journal Radiology. The technology is an important first step in the development of artificial intelligence (AI) that could interpret scans and diagnose conditions.

From an ATM reading handwriting on a check to Facebook suggesting a photo tag for a friend, computer vision powered by artificial intelligence is increasingly common in daily life. AI could one day help radiologists interpret X-rays, CT scans and magnetic resonance imaging (MRI) studies. But for the technology to be effective in the medical arena, computer software must be taught the difference between a normal study and abnormal findings.

This study aimed to train this technology how to understand text reports written by radiologists. Researchers created a series of algorithms to teach the computer clusters of phrases. Examples of terminology included words like phospholipid, heartburn and colonoscopy.

Researchers trained the computer software using 96,303 radiologist reports associated with head CT scans performed at The Mount Sinai Hospital and Mount Sinai Queens between 2010 and 2016. To characterize the “lexical complexity” of radiologist reports, researchers calculated metrics that reflected the variety of language used in these reports and compared these to other large collections of text: thousands of books, Reuters news stories, inpatient physician notes and Amazon product reviews.

“The language used in radiology has a natural structure, which makes it amenable to machine learning,” said senior author Eric Oermann, M.D., instructor in the Department of Neurosurgery at the Icahn School of Medicine at Mount Sinai.  “Machine learning models built upon massive radiological text datasets can facilitate the training of future artificial intelligence-based systems for analyzing radiological images.”

Deep learning describes a subcategory of machine learning that uses multiple layers of neural networks (computer systems that learn progressively) to perform inference, requiring large amounts of training data to achieve high accuracy. Techniques used in this study led to an accuracy of 91 percent, demonstrating that it is possible to automatically identify concepts in text from the complex domain of radiology.

"The ultimate goal is to create algorithms that help doctors accurately diagnose patients,” said first author John Zech, a medical student at the Icahn School of Medicine at Mount Sinai.  “Deep learning has many potential applications in radiology — triaging to identify studies that require immediate evaluation, flagging abnormal parts of cross-sectional imaging for further review, characterizing masses concerning for malignancy — and those applications will require many labeled training examples."

“Research like this turns big data into useful data and is the critical first step in harnessing the power of artificial intelligence to help patients,” said study co-author Joshua Bederson, M.D., professor and system chair for the Department of Neurosurgery at Mount Sinai Health System and clinical director of the Neurosurgery Simulation Core.

Researchers at Boston University and Verily Life Sciences collaborated on the study.

For more information: www.mountsinai.org

 

Related Content

Anatomage Releases Anatomage Cloud Platform
News | Remote Viewing Systems | July 16, 2019
Anatomage Inc. released an update to the Anatomage Cloud platform that allows medical and dental professionals to...
Example of an intentionally truncated CT image

Figure 1: Example of an intentionally truncated CT image. The truncation percentage was calculated as the ratio of the patient border touching the field of view to the total patient border (red/(read+blue)). Image courtesy of Qaelum.

Feature | Radiation Dose Management | July 15, 2019 | Niki Fitousi, Ph.D., and An Dedulle
One of the main benefits of a radiation dose management system is the possibility to automatically generate alerts when...
Graphic courtesy Pixabay

Graphic courtesy Pixabay

Feature | Artificial Intelligence | July 15, 2019 | By Greg Freiherr
Siemens has long focused on automation as a way to make diagnostic equipment faster and more efficient.
Videos | Artificial Intelligence | July 12, 2019
Khan Siddiqui, M.D., founder and CEO of HOPPR, discusses the economic advantages and costs presented by...
Routine scan of abdomen pelvis taken with the UW-Madison’s Revolution 256 CT scanner using the FDA-cleared reconstruction algorithm, called TrueFidelity.

Routine scan of abdomen pelvis taken with the UW-Madison’s Revolution 256 CT scanner using the FDA-cleared reconstruction algorithm, called TrueFidelity. UW-Madison was the first site in the U.S. to get this technology. Its use is now being integrated into UW CT protocols. Image courtesy of Timothy P. Szczykutowicz

Feature | Computed Tomography (CT) | July 12, 2019 | By Greg Freiherr
When providers develop their own imaging protocols, they are wasting time and money, according to...
Videos | Digital Pathology | July 11, 2019
Toby Cornish, M.D., Ph.D., associate professor and medical director of informatics at the University of Colorado Scho
FDA Clears Koios DS Breast 2.0 AI-based Software
News | Ultrasound Women's Health | July 11, 2019
Koios Medical announced its second 510(k) clearance from the U.S. Food and Drug Administration (FDA).
Mednax National Cardiac Centers of Excellence Program Highlighted at SCCT 2019
News | CT Angiography (CTA) | July 11, 2019
Mednax Inc. and Mednax Radiology Solutions announced that Chief Medical Officer Ricardo C. Cury, M.D., FSCCT, will...
SimonMed Imaging Implements ProFound AI for 3-D Tomosynthesis
News | Mammography | July 10, 2019
Arizona-based SimonMed Imaging announced their implementation of the first U.S. Food and Drug Administration (FDA)-...