By Raj Chopra

I am and have always been a believer in the potential for good that AI can bring to radiology. I am a practicing radiologist myself; I know too well how tedious it is to stare at scan after scan, either in a windowless reading room or in my home office, and how radiologists like me would be served by an automated tool that can speed up that process.

But despite the market need for AI tools in radiology and the inescapable buzz around it, for years, going to event after event and talking to vendor after vendor, I found myself increasingly disappointed and skeptical that anyone would ever crack the code on AI for radiology. Forget about the concerns of whether AI would replace radiologists’ jobs; it never seemed like AI would even get far enough to help support the jobs that already existed.
That all changed for me at RSNA 2025, when, for the first time, my skepticism was fully brushed aside. What I saw there was a true inflection point in the future of AI for radiology and the steps forward we can and must take to make sure we’re not just using AI for the sake of it, but actually solving the imaging crisis we all face.

The paradigm shift to efficiency AI is real and overdue
What was really clear at that conference was the shift away from pixel AI to efficiency AI. For years, pixel AI applications received all the talk and investment for AI in imaging. I was not a fan of this trend; I’ve always thought that efficiency is where AI can deliver innovation for radiologists like me. Others have talked about that, but 2025 was the year when the money finally seemed to be moving in that direction.

What does that mean for clinical teams, like the radiologists and practices that I work with? It means faster, quality, and reducing burnout. Efficiency AI applications have the potential to finally address and resolve so many of the issues that my colleagues deal with on a daily basis – from reducing the number of clicks a radiologist has to make per image, to decreasing scan teams, to extrapolating out images that could effectively mean shrinking five or six MRI sequences into just two or three.

Across the country, it’s taking longer and longer for patients to get MRIs done and have their results back. That’s not fair to the patient, and it’s not the expectation they sign up for when they have a test like that performed; they’re concerned and want answers fast. But radiologists are overrun with demands for exams and the sheer quantity of images being generated by those exams.

Efficiency AI has the potential to solve both problems. AI-driven advances in scanner technology are making it easier and faster to get patients imaged in the first place. We are moving toward a world where scans can be done with fewer sequences, less time, and potentially fewer technologists, while still extracting the same or better clinical information. That has meaningful implications for deploying imaging more easily in rural or underserved areas and for increasing throughput at existing sites.

Artificial Intelligence Homepage