AI tools are quickly creating new opportunities for innovation in the medical world, which could provide much-needed support to specialities like radiology where clinicians suffer from systemic burnout and workforce shortages. But in order for imaging AI developers to build successful products, it’s not enough to show hospital leadership and IT the financial and operational value of a tool. For radiologists to get excited and fully embrace a product, developers also need to prove that they understand how these specialists work by creating tools that work where they do.
Breaking down a radiologist’s workflow
The core of a radiologist’s workflow takes place on several platforms, including the picture archiving and communication system (PACS), report dictation software, electronic health records, and Radiology Information System. These programs are spread across multiple monitors, allowing radiologists to access the current case, prior medical history, worklist, reports, and dictation software. However, looking back and forth at the different monitors creates workflow disruption, which is why many radiologists are loathe to add in AI tools to the already fractured set-up.
AI tools often increase the disruption that radiologists already experience from using non-integrated platforms. Opening a web browser or separate application takes time and disrupts the main workflow. Extra clicks equal more time spent reviewing each case, and spending all day switching in and out of programs depending on the type of scan can increase cognitive stress, not reduce it. In some cases, the radiologist may have to manually enter information from an AI tool into their report, which takes up even more time.
Non-integrative tech creates risk
Creating imaging AI tools that integrate with existing workflows isn’t just about reducing clinician frustration; it’s also about reducing the risk of mistakes and making sure that “findings” from AI tools are actually usable to clinicians. For example, AI tools that operate as widgets create pop ups on the screen, actually cluttering the radiologist’s view and interfering with the diagnostic process. Additionally, tools that help prioritize urgent cases may be helpful in some situations with many radiologists reading from a common list, but not as much in departments like the ER where nearly every case is urgent.
Another problem with non-integrative tools is that they may generate a static report, often a PDF, that remains permanently in the patient’s files, even if the radiologist disagrees with the AI’s “findings.” Because it’s in a non-editable format, the radiologist can’t interact with the report or respond to it, creating confusion that may linger in the patient’s health record for years to come.
Anticipate localization needs
Research has shown that the accuracy of imaging AI tools often drops when applied to new datasets because of differences between the training data and local clinic populations. To build a tool that radiology practices can rely on and build their practices around, developers need to offer secure methods for users to incorporate their own data for localized training.
Doing so allows practices to customize AI tools for their specific needs, which can include integration with a clinic’s unique procedures and workflows. It also gives them the ability to ensure the tool is accurate to the clinic’s patient population and what kinds of surgeries and abnormalities are commonly in their system.
For radiologists, this means a tool that not only performs well on average but performs well for their patients, making it something they can genuinely trust and build their practice around.
Plan for integration from day one
While developing an imaging AI tool that is widely adopted by radiologist practices isn’t easy, there are platforms finding success by focusing on providing tools that work where the clinicians do. Tools that take dictated findings and turn them into report drafts are popular, as are programs that pre-populate measurements.
For detecting anomalies in scans, successful platforms are those that act like an intuitive layer on top of the existing PACS, sending measurements and notes straight to the report. Understanding the importance of integration, some PACS developers have even begun incorporating built-in AI tools rather than wait for third-party developers.
Medical AI development requires significant time for HIPAA compliance and clinical validation. However, meeting these safety standards does not guarantee that a tool will be useful in practice. To address this, some developers use AI foundry tools to automate or accelerate the standard parts of the development process. By reducing the time spent on these technical tasks, developers can focus on how the platform will function for the radiologist. The success of an imaging AI platform depends on this balance between technical compliance and practical integration.
The golden question
Even if a tool looks good to hospital and IT leadership, widespread adoption will only happen if the clinicians find the tool genuinely helpful and easy to use. Asking, “Would a radiologist actually use this?” is one of the most important steps developers can take to ensure their product’s success.
Photo: ismagilov, Getty Images
Dr. Roger Boodoo, Radiologist and Medical Director of AI, HOPPR, is a double board-certified diagnostic radiologist and clinical informaticist known for his pioneering contributions to healthcare innovation and military medicine. Trained at Walter Reed National Military Medical Center and the University of Illinois Chicago, he has held key leadership roles within the Department of Defense, including Chief of Innovation and Lead Imaging Informaticist at the Defense Health Agency, and Chief of Radiology at Fort Belvoir. He is also a retired Commander in the U.S. Navy.
During his tenure at the Defense Health Agency, he spearheaded the Military Health System’s Enterprise Imaging and modern EHR initiative, successfully overseeing the deployment of imaging solutions across 138 military hospitals and clinics, facilitating over 5 million exams annually. He also founded two innovative startups: a radiation tracking app and an AI-driven image labeling platform designed to create curated datasets for pathology applications.
This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.
