“CALM” AI: Technology with Less Friction
Less-intrusive approach benefits patients as well as doctors
The principles of continuous, ambient, low-friction (CALM) technology emphasize minimal interruption, clear communication and background support that surfaces only when needed. Embedded seamlessly into PACS and reporting systems, CALM AI can reduce cognitive load without adding steps or alerts.
“If I had to summarize CALM in one sentence, I’d say the best AI may ultimately be the AI that the radiologist doesn't have to think about using,” said Shlomit Goldberg-Stein, MD, professor of radiology at the Zucker-Hofstra School of Medicine and director of AI in the Department of Radiology at Northwell Health in New York.
Radiologists in her practice use AI tools not only for detection and triage, but throughout the imaging workflow. “We have built tools that synthesize the relevant clinical history for the radiologist. Instead of searching through multiple notes, labs and prior encounters, the goal is to bring the clinical context to the radiologist at the time of interpretation,” Dr. Goldberg-Stein explained. “Those are all areas where AI can support the radiologist, ideally without adding another task or another piece of technology that we must actively manage.”
More broadly, the goal is not just efficiency, but reducing fragmentation across systems, tasks, and handoffs, she explained. A major opportunity lies in closing the loop to ensure the recommended next steps actually happen.
“Increasingly, we're also using AI during the interpretation workflow to suggest next steps in care. Getting patients to the right specialist sooner, accelerating appropriate treatment and preventing patients from being lost to follow-up are meaningful outcomes that can make a difference at a larger scale,” Dr. Goldberg-Stein said.
Ideally, she said, those recommendations should be patient-centered and grounded in the full clinical context, including the patient’s existing conditions, acuity and prognosis.
“Radiologists are the most direct beneficiaries when AI reduces tedious, repetitive work, improves efficiency and lowers cognitive burden,” Dr. Goldberg-Stein noted. “Ultimately, however, the most important beneficiary should be the patient.”
AI Tools That Keep Radiologists in the Driver’s Seat
Paul J. Chang, MD, professor and vice-chair of radiology informatics in the Department of Radiology at the University of Chicago, compared the ideal relationship between radiologists and CALM AI to driving a car.
“I like driving my car, but I don't like fiddling with the buttons if someone thinks they want more air conditioning,” Dr. Chang explained. “To me, the optimized human-AI orchestration is that I'm still driving the car, but my AI agent is a very willing, helpful passenger that's not nagging me, but actually helping me do what I want to do so I can still enjoy the driving part.”
For that partnership to work, Dr. Chang said AI must understand what the radiologist is doing and what information is needed in the moment. “Without real-time data interoperability, state awareness and optimized human-machine workflow orchestration, radiologists could become increasingly removed from decision-making,” he warned.
For Dr. Chang, that means AI must be able to access and interpret information from multiple systems in real time and recognize where the radiologist is in the workflow. Rather than functioning as a separate tool, AI should understand the patient’s clinical history, prior imaging, current examination and the radiologist’s actions during interpretation, allowing it to provide relevant support without disrupting decision-making.
Technology intended to alleviate radiologists’ cognitive burden can instead increase it if it is not integrated effectively into existing systems. “If radiologists have to open another application, look at another screen, acknowledge or reconcile multiple results, then we've potentially worsened the cognitive load rather than improved it,” Dr. Goldberg-Stein explained.
Dr. Chang said radiologists cannot fully realize the benefits of CALM AI until it is more deeply integrated into clinical workflows. “It's not that the radiologist doesn't want to work with AI; it's because our existing systems—our electronic medical records and PACS—treat AI as a peripheral and won't allow deep integration, drastically limiting AI’s potential,” he said.
Greater integration also raises important considerations about data governance, privacy and cybersecurity, challenges that many health systems must address as AI becomes more deeply embedded in clinical workflows.
According to Dr. Chang, effective integration requires AI to have the same clinical context radiologists use to interpret examinations, including patient history, imaging findings and relevant electronic medical record data. “It has to know what I know right now,” Dr. Chang explained. “To work well, I have to also be able to share that orchestration context with the AI agent.”
That context extends beyond clinical data to include where the radiologist is in the workflow and what task is being performed.
Dr. Chang added that AI must also have access to real-time workflow context that shapes radiologists’ decisions. “Sharing data alone is not enough if the technology cannot recognize what the radiologist is doing as the interpretation unfolds,” he said.
By understanding both the patient information and radiologist’s actions in real time, AI can coordinate information and support across systems without disrupting the interpretation process.
Looking ahead, Dr. Goldberg-Stein sees broad potential for CALM AI. “Ultimately, I think the future is an intelligent layer around the radiologist that finds information, organizes it, learns from what happens afterward, and brings the right information to the right person at the right time,” she said.
For More Information
Read previous RSNA News stories on radiology workflow optimization: