AI Sprawl in Radiology Demands Integration Over Deployment

The Problem With Piling On AI Tools

A radiologist reviewing a chest CT scan encounters three separate AI algorithms flagging the same pulmonary nodule. One writes findings directly into the PACS system. Another generates a risk score in its own worklist. A third requires a separate login. Each tool defines "urgent" differently, and none delivers results at the same point in the reading workflow.

This scenario reflects a growing challenge in medical imaging departments, according to Lior Eshel, founder and CEO of TestDynamics. With more than 1,500 AI algorithms cleared by the FDA—most targeting medical imaging—health systems are accumulating what Eshel calls "vertical AI solutions" designed to solve specific problems. But deploying these tools independently creates friction that undermines their value.

"The first tool changed something real," Eshel explains. "The second is where workflows break, the gain is smaller, and the cost is not only the license, but it's another integration, another stream of alerts, and another confidence score."

Why it matters

Radiology departments risk wasting significant AI investments if clinicians abandon tools that disrupt workflow. As health systems move beyond pilot projects, the operational challenge shifts from selecting individual algorithms to orchestrating multiple AI outputs into coherent, clinically useful information—a capability that will determine which organizations extract real value from AI.

From Point Solutions to Platform Thinking

By the third or fourth AI deployment, health systems must shift their approach, Eshel says. The focus moves from evaluating individual products to managing how multiple AI outputs are "amalgamated" and surfaced to users in relevant ways.

Vendor-neutral platforms like TestDynamics' Satori address this integration challenge by consolidating AI findings within existing clinical workflows. The platform uses radiologist-validated templates to present AI results in proper clinical context, giving clinicians control over which findings they see. If a backend algorithm needs replacement due to poor performance, the interface remains consistent for physicians.

"Radiologists like the solution because it saves them a lot of time and makes everything concise," Eshel notes.

Workflow-First Selection and Post-Deployment Monitoring

Eshel recommends radiology leaders adopt a workflow-first approach when evaluating AI tools, asking how to enhance existing processes without forcing clinicians to change viewers, worklists, or reporting systems. Rather than presenting departments with lists of hundreds of AI vendors, he suggests focusing on a single clinical use case and identifying tools that address it effectively.

Post-deployment governance presents another critical gap. While the FDA clears algorithms based on performance at a specific point in time, it doesn't account for drift that occurs as local populations, scanner protocols, and clinical conditions evolve.

"You have to have post-market surveillance data on the AI and how it performs in real life and in clinical setup moving forward," Eshel says. Platforms like Satori monitor algorithm performance over time, triggering alerts when accuracy begins to decay or when end users consistently disagree with AI recommendations.

Measuring What Matters

Successful AI-enabled imaging departments won't be defined by the number of algorithms they deploy, Eshel argues. Instead, they'll measure performance locally rather than accepting external validation, maintain the ability to remove tools without rebuilding workflows, and separate algorithm vendors from those responsible for evaluating them.

These details were first reported by ITN Online.