Back in 2019, Professor Daniel Rubin [1], a professor of biomedical data science, radiology, and medicine at Stanford University, astutely noted: "In the Gold Rush, the value of gold depended on the market for banks buying gold from miners. In the AI era, the value of these algorithms will depend on the market for radiologists who decide to purchase them."

Radiologists have navigated the integration of AI into routine radiology practice with a blend of eager anticipation and cautious consideration. Initially concerned that AI might replace the interpretation of imaging studies, it soon became evident that AI algorithms could effectively assist and enhance the quality of a radiologist's primary responsibilities. So, why haven't all radiologists embraced AI to reap its benefits in their daily practice?

The key hurdle lies in the absence of clinical evidence demonstrating AI's positive impact on clinical, workflow, and business metrics. The AI business faces a challenging outlook for the upcoming years without this crucial validation, resulting in discouraging deployments in radiology. This creates a detrimental cycle—lack of proven successes leads to slow resource allocations.

What is needed is not just a plethora of AI applications, each specialized and limited to certain pathologies, but an overarching AI application, running in the background of the image viewing processes, that notifies radiologists of any pathology in any study, including incidental findings. This app would become a radiologist's indispensable ally, serving as the catalyst for the clinical and business success of AI in our sector.

Developing an app involves incredibly intricate aspects of research and product development. A commitment to this mission requires not only access to a vast arsenal of academic and financial resources but also a steadfast dedication to a prolonged and continuous deployment of these resources by the university or company taking on this challenge.

The slow pace of AI market deployment, in contrast to the expectations set in 2018, has resulted in limited academic and business studies showcasing the tangible ROI of AI in clinical settings.

If you ask around or spent some time in the AI Showcase at RSNA 2023 in Chicago, you too may have realized the adoption of AI within the radiology community has been sluggish, despite notable successes in emergency departments where AI has demonstrated effectiveness in triage, stroke detection, differentiation between LVO and hemorrhagic strokes, COVID-19 detection, and other critical areas. This evolution has not yet translated to widespread AI integration in the daily practices of radiologists.

At the meeting, numerous imaging informatics vendors demonstrated AI platforms and applications sourced from third-party vendors or specialized AI providers. Larger imaging equipment vendors, recognizing the value of AI, have begun offering application programming interfaces to enable customers to seamlessly integrate their preferred AI applications with the vendor’s software, spanning reporting, PACS, advanced visualization, or EHR systems.

In an attempt to overcome the challenge of limited market potential, many AI vendors have shifted their focus from creating and maintaining an independent sales force to seeking strategic partnerships with radiology equipment vendors. By distributing their AI applications through these established channels, they aim to maximize their reach and impact in the market.

A portion of AI vendors operate in the clinical decision support market, identifying opportunities to enhance CDS applications through AI technologies. However, due to resource constraints, these vendors prefer partnerships with equipment vendors as they lack the capacity to effectively communicate the multitude of benefits and potential return on investment (ROI) associated with widespread AI integration in CDS.

Business intelligence applications can derive significant advantages from the extensive data reservoir that AI can extract from various information systems within and outside of the radiology department, including scheduling, procedure coding and EHRs. However, the utilization of these applications is once again hindered by the initial scarcity of AI applications within the healthcare provider sector.

Uncertainty about regulatory issues in AI is yet another market deterrent for many OEMs.

Taking a big picture view, these market trends are generally unfavorable, signaling a likely consolidation within the AI vendor space. The company that successfully pioneers an AI application—distinctly enhancing the workflow, elevating the effectiveness and efficiency of radiologists, and showcasing undeniable ROI—will forge the elusive "killer app" that establishes an unparalleled competitive edge. Until that innovation emerges, the market will likely linger in uncertainty regarding the true value of AI in radiology.