What Criteria Do You Need to Adopt AI Into Your Radiology Practice?

Posted by Alyssa Watanabe, MD, FACR on Nov 20, 2019 2:24:54 PM
Alyssa Watanabe, MD, FACR

Artificial Intelligence (AI) has arrived in radiology and is quickly moving from the lab to clinical use. A recent survey shows that healthcare executives are not only embracing the use of artificial intelligence, they’re making budgetary plans to bring the technology into their organizations.

More than 500 senior U.S. healthcare industry executives participated in the inaugural OptumIQ Annual Survey on AI in Health Care. The survey found that healthcare organizations will invest an average of $32.4 million on AI implementation over the next five years.

The healthcare executives cited more accurate diagnosis and increased efficiency as AI’s top two benefits, and most expect AI to lead to clinical and financial improvements in health care. The survey reflects what many providing direct patient care have already recognized – artificial intelligence is fast becoming a game changer in health care.  

Nowhere is this more evident than in radiology, where its vast potential has resulted in a plethora of new products and technologies coming onto the market. But with so many to choose from, what criteria do radiologists need to select and incorporate AI into their practices? To help radiologists and practice executives prepare for the RSNA 2019 - Radiological Society of North America Annual Meeting, we share key business and clinical factors to consider in evaluating and selecting AI partners for radiology.

Five Key Factors for Selecting AI Radiology Solutions

In choosing an AI product for your radiology program, keep in mind how the technology will affect such issues as quality, workflow, patient care and efficiency. The first and most important questions to ask are, “Why are you buying AI? What direct benefits does a solution offer?” Here are some other key questions to consider:

  • Improve accuracies. Does the AI technology demonstrate improved sensitivity and specificity in image analysis to help radiologists enhance disease detection? Is it going to strengthen your clinical performance?
  • Improve efficiencies in readings. Does the AI product optimize the radiologist’s workflow by giving highest priority to the most suspicious cases rather than the first in, first out approach currently in use? Is it going to make your clinical day more efficient and effective?
  • Improve patient care – and faster patient diagnosis. Does the AI product aid radiologists in making faster and more accurate readings that will expedite patient diagnosis, and as a result, improve patient care? Is it going to help you attract and retain patients?
  • Help identify false positives. Has the technology been shown, in a statistically significant way, to help radiologists reduce the rate of false positive readings? Is it going to reduce your liability and increase your patients’ satisfaction?

Technical Aspects Are Also Important

AI needs to be more than just an algorithm to be an effective commercial product. Algorithms alone may make for great papers and publications, but commercial products are solutions that you can rely upon in your practice. To evaluate if a solution is a science project or a commercial product, consider the following questions:

  • Clinical Validation and FDA approval - Is the product FDA cleared? Does the clearance cover only the algorithm or the entire technical architecture for the product?
  • Algorithm Testing - Has the AI product been tested across populations? Do the results generalize “in the wild” and match what is reported “in the lab?”
  • Security Measures – Is the solution cyber-secure? Is it HIPAA compliant? Does your patients’ PHI leave your clinic?
  • Technical Infrastructure – Is the solution cloud-based or server-based? If cloud-based, is it a reputable cloud brand such as Amazon Web Services or a self-managed cloud from the vendor?
  • Installation and Setup – Is the solution easy to install and setup or will it require countless hours from your already over-worked IT team and PACS Admin?
  • Customer Service - Does the vendor have a service plan to provide technical and user support?
  • System Integration - Does the AI technology work with multiple manufacturers so that it can seamlessly integrate with existing medical center technology?

Addressing Other AI Concerns

In talking with radiologists about adopting AI technology, some have expressed fear about AI products helping them in their diagnostic role. Underlying this apprehension are concerns that AI products will ultimately take over the radiology function – leaving no need for the knowledge and experience of highly trained radiologists. Such a scenario doesn’t seem likely. Rather, many in the industry see AI as an asset - and one that is fast becoming an important tool that will aid radiologists, not replace them.

AI also offers the potential for increased efficiencies, better diagnostics and improved patient care, making it attractive to health care leaders.

Radiology AI Focuses on Image Analysis and Workflow Efficiency

A number of AI-powered products for medical image analysis are emerging in radiology and cover a wide variety of disease areas. These include applications that help radiologists in analyzing brain structures, cardiac conditions, lung diseases and a variety of cancers.

Another type of emerging AI technology looks at improving the radiologist’s ability to focus their workflow. For instance, in the area of breast cancer, cmTriage™, the first FDA-cleared AI triage software in the U.S. for mammography, changes the “first in first out” protocol in mammograph readings by prioritizing the most suspicious cases to the top of the radiologist’s worklist. Studies have demonstrated up to 30% faster read time for radiologists using cmTriage.

CureMetrix is also currently conducting studies across the globe to expand its AI solutions to also help identify, mark and score anomalies in breast cancer screening. In studies published in the Journal of Digital Imaging, CureMetrix cmAssist® was able to demonstrate the ability to find cancers up to six years before first detection, and help radiologists improve their breast cancer detection rate on average 27% without increasing recall rates.

  • These are just a few examples of the AI applications that have arrived in radiology. They add further credence to AI’s potential and demonstrate that the use of artificial intelligence is not something “down the road.” AI is happening now and will continue to reshape and improve our industry.

To learn how Radiology + AI = Powerful Results for your practice, contact us at info@CureMetrix


Topics: News, radiology, artifical intelligence