Top 4 Challenges of Artificial Intelligence Adoption in Enterprise Imaging
As we move into the third decade of the 21st century, radiology—perhaps more than any other medical specialty—is poised for transformation. In a survey conducted by MIT Technology Review Insights, “more than 82% of health-care business leaders report that their AI deployments have already created workflow improvements in their operational and administrative activities—giving clinicians time back to work with their patients more closely, and with more insight.”
Thanks to AI, radiologists foresee a future in which machines enhance patient outcomes and reduce misdiagnosis. Therefore, it is no surprise that AI in healthcare and especially medical imaging presents a promise. A ‘tsunami’ of imaging-focused AI algorithms is flooding the marketplace, and yet caregivers and Health IT professionals face significant challenges when it comes to effectively deploying the algorithms or applying the insights generated by them.
As a Health IT company trusted by top healthcare facilities to simplify enterprise imaging management, Dicom Systems has a unique perspective on the challenges of deploying AI in the clinical workflows. From the lack of understanding of enterprise imaging workflows and industry standards among AI developers to the dependence on the cloud and the inconsistent format of the output, these challenges are significant, yet addressable. Whether through a universal adapter or through improved communication and education, overcoming these obstacles will allow radiology to realize the full potential and reap the benefits of AI.
Insufficient Adoption of Teleradiology Industry Standards by AI Developers
One of the top obstacles in the deployment of AI algorithms is that most algorithms are not developed with imaging industry standards in mind. AI methods require data to optimally train, validate, and test algorithms. Radiology images contain rich metadata, in addition to the actual pixel data. While an algorithm might be effective on JPG images for detection of areas of interest at the pixel level, more often than not it is generally unaware of the metadata. AI developers may simply not be well-versed in DICOM or HL7 data and processes. This gap limits the effectiveness of the AI solution and complicates its adoption within established diagnostic imaging workflows.
To mitigate this challenge, AI vendors need to become sufficiently educated on imaging industry standards, prior to implementing AI. AI developers are often capable of adopting industry standards within their processes as these organizations are staffed with talented software developers and data scientists. While they may not specialize in medical imaging, their collective skills make them natural adopters of imaging industry standards.
Reliance On The Cloud Excludes On-Prem Customers
Some AI algorithms are developed in cloud infrastructure and made available strictly as a cloud solution. While the cloud environment lends itself very well for AI deployments, not all customers can, or are willing to, operate in the cloud due to procedural or security reasons. Healthcare cloud adoption is gaining momentum, with a majority of healthcare organizations reported as being ready to embrace a cloud-first approach for their health IT infrastructure, but “the cloud” is not yet universally adopted. When it comes to radiology, AI algorithms should be “dockerized” in a standard manner. Leveraging standard RESTful APIs would give the caregivers’ organizations the freedom to choose where and how their algorithms are deployed.
Lack Of Enterprise Imaging Workflow Knowledge
AI Results Are Difficult To Access
Our AI Conductor for PACS and EHR was designed to address this formatting gap, among other challenges. We have designed our technology to function as a “universal adapter” for enterprise imaging, allowing imaging professionals to quickly and easily solve interoperability challenges, and create new custom workflows specific to our customers’ needs. Today, our Unifier routing platform can be leveraged as a versatile AI workflow conductor, enabling the deployment of any AI on premises or in the cloud (public, private or hybrid), leveraging industry standards, and allowing AI vendors to uniformly dockerize their technology within the framework.
AI in imaging is still an early-adopter market, with the majority of caregivers waiting for more evidence that AI is safe, effective, easy to implement, and desirable to adopt. Fortunately, the bottlenecks that impact successful AI implementations can be resolved by designing and orchestrating workflows to get the information to the right location at the right time in the right format. Most importantly, AI must perform its function in a discreet, nearly invisible manner, so as not to cause disruption in the workflows of diagnosticians.