Through a collaboration between Philips and NVIDIA, imaging AI is advancing toward anticipating what an MR scan will reveal before the exam is even finished — bringing error detection earlier in the workflow and making it more actionable.

Think of a “predictive preview,” says Sathish Kumar Balakrishnan, head of global research and development for MRI at Philips. “It’s being designed to give operators an early, AI-driven estimate of what an MR image is likely to look like,” he says. Instead of waiting for a scan to finish, the system will generate a near-real-time preview at the console, drawing on early scan data and patterns learned from prior imaging.

These capabilities will allow technologists to assess whether positioning, sequence selection, and coverage are correct without waiting for the full scan to complete. The goal is straightforward: surface issues sooner, make adjustments in the moment, and avoid costly rework later.
Connecting the workflow
Image preview capabilities are part of a broader effort between the two companies to bring the MR workflow into closer alignment. Philips is also developing an MR foundation model designed to work alongside NVIDIA tools such as NV-Generate, NV-Segment, and NV-Reason so that planning, acquisition, and interpretation can be more tightly bound together.

Over time, according to Balakrishnan, these capabilities “may also play a role in enabling more automated scan workflows,” where systems help optimize parameters proactively while still keeping technologists in control.

Support, not replace
For now, Philips’ technology remains in the research phase, and Balakrishnan is careful to emphasize its limits. “The preview is an estimate, not a diagnostic image,” he says, “and it is not intended to replace the fully acquired scan.”

As with any AI system, performance depends on the data behind it, he notes. That means validation across different anatomies, patient populations, and clinical scenarios, along with attention to real-world scan conditions.

Philips is taking a stepwise approach to validation, combining quantitative testing with clinical feedback to assess how closely previews align with final images. “Defining appropriate reliability thresholds and intended use is an important part of the ongoing research and development process,” he notes.

Clear communication is part of that effort. “We are considering how to communicate the nature of the preview to users,” Balakrishnan says. That work involves framing the preview as a supportive input while keeping final decisions firmly in the hands of the MR technologist.

By shifting feedback earlier in the process, predictive preview aims to tighten the loop between insight and action. If it delivers, the impact won’t come as a single breakthrough—but as a steady reduction in friction: fewer repeat scans, more consistent results, and less second-guessing at the console.

Artificial Intelligence Homepage