Andrew MacLean

Why Radiology Departments Must Insist on AI Platforms

The clinical AI hype has spread like wildfire. The crowd of skeptics have largely converted to believers, and we’ve seen conversations evolve toward best practices for AI governance structures and which stakeholders ought to play a role in creating a lasting, system-wide AI strategy. 

Radiologists, however, are already adept when it comes to deep dive AI discussions. Having long realized the potential for AI to drastically impact workflows, whether that be reading time reduction or worklist reprioritization to name a few, the reading room has been the de facto starting point for many new forms of healthcare technology, AI being no exception. 

As the radiology department remains a crucial stakeholder and facilitator of a sustainable AI strategy, here are three key things to consider no matter where you are in your AI journey:

Championing the Enterprise-Wide AI Potential

AI algorithms can be transformative for their designated use cases, and facilities have found success in adding one-offs as part of an immediate need to solve real clinical problems. With great success in AI, the question doesn’t become whether AI is right, but whether, as constructed, this solution remains stable and scalable. 

Will adding a new algorithm from a separate vendor cause unforeseen complications? Will they work in unison or against one another? Will the protocols for the new algorithm conflict with our current one? These are just some of the key questions radiology leaders consider when scaling AI beyond a single use case.

Let’s take an example of an ED patient following a car accident. They enter the ED and are given a chest and abdomen contrast-enhanced CT. How would you answer the following questions about your AI?

Which algorithms will be orchestrated to run on each exam?

Is your AI always on in the background acting as an extra layer of intelligence? As the scans come in, how does your system “orchestrate” or decide which algorithms to run on each exam? The beauty of AI is that systems can run multiple algorithms, in parallel, on each exam looking for the expected but also the unexpected pathologies. The tough part is how do you configure and maintain this orchestration of exams, and monitor the systems performance so that as changes happen at your site, your orchestration remains optimally tuned.

What is the radiologist’s experience?

Say your health system is deeply invested in radiology AI and you’ve been able to expand your pool of algorithms to 20. How will you know the status of each algorithm being run on an exam? Are they all finished processing, making it safe to read the exam, or do you still need to wait for the algorithm to run? How will the system handle prioritization of urgent acute findings such as pulmonary embolism or stroke? What if you want two or three of the AI results to be automatically inserted into the report – how is that handled? With a proper AI platform, a radiologist should be presented the AI status and results in a unified interface, in the most non-intrusive way possible.

Let’s refer to our aforementioned ED patient. You could decide to run obvious algorithms looking for pathologies associated to the car accident like rib fracture, vertebral compression fractures, etc., but would you also run other unrelated algorithms looking for pulmonary nodules, a pulmonary embolism and perform automated measurements of the aortic diameters? When it comes to an enterprise-wide platform, having a single platform that orchestrates all of the different algorithms based on the scan type and anatomy present in an image enables you to not just find what’s expected to be wrong with the patient, but also find the unexpected – enabling better patient outcomes.

This is part and parcel of the AI vision moving forward. With industry experts forecasting consolidation among AI vendors, it’s vital for health systems to exercise extreme scrutiny when evaluating their potential partners. Things like data normalization, single interfaces and workflow integrations acting as a single point of contact for healthcare systems is essential to guarantee a long-lasting return on your AI investment, both clinically and financially.

Explore the Latest AI Insights, Trends and Research

Andrew MacLean