Reimagining radiology: Dr. Paul Chang’s vision for real-time AI collaboration

Radiology has always evolved through disruption. But today’s AI conversation is stuck at the surface. Most solutions operate in parallel to the radiologist, prioritizing worklists or generating secondary captures, without truly integrating into workflow. Algorithms detect nodules, prioritize worklists and generate findings. Yet despite this progress, something fundamental has not changed.

Radiologists are still overwhelmed. Worklists are still growing and burnout remains real. Why?

Because most AI today operates at the margins. It functions as a peripheral running in parallel to the radiologist, touching workflow only at the edges: flagging, suggesting and capturing. But it does not collaborate. Dr. Paul Chang believes this is the critical limitation. AI will not fulfill its promise until it becomes integral to workflow, embedded directly into how radiologists think, navigate and report.

The greatest inefficiencies in imaging are not dramatic. They are microscopic. Let’s think about that:

Stopping to measure a lesion. Switching screens to dictate. Scrolling through prior reports. Repeating structured information manually. These are just few examples of what Dr. Chang calls “micro-inefficiencies.” Think about that: each interruption may take seconds. But multiplied across hundreds of cases, they accumulate into fatigue, cognitive load and frustration. Radiologists do not dislike interpreting images. They dislike the friction surrounding interpretation. “I enjoy being a radiologist,” Dr. Chang says. “I just don’t like the micro-inefficiencies.” The future of AI, in his view, is not about replacing radiologists. It is about eliminating this friction at scale.

Dr. Chang’s inspiration comes from academic radiology. When an attending radiologist works with a resident, efficiency increases dramatically. Even a first-year resident, new to radiology, can measure lesions, compare studies, document findings and prepare structured drafts. The attending remains in control while the resident acts as an “agent”: this is the model Dr. Chang believes AI should follow, which means a true collaboration between parties and not an autonomous replacement. To achieve this, incremental change is not enough. Today’s AI solutions typically operate asynchronously. They run independently of the radiologist’s actions. They produce results that must be reviewed separately. They add data, but often also add work.

This kind of transformation is not incremental but architectural. It requires re-engineering the entire radiology reading workflow to enable real-time, context-aware human–AI collaboration. Delivering this vision requires more than a traditional PACS. It depends on a deeply integrated platform that unifies diagnostic viewing, worklist orchestration and reporting into a single workflow environment. It requires deep expertise across enterprise imaging, workflow orchestration, cloud, interoperability and clinical practice. It is, undeniably, a heavy lift. If AI remains superficially bolted onto existing systems rather than built into them, then innovation will advance outside the radiologist’s control. Foundation models will evolve. Automation will accelerate and radiologists may slowly shift from decision-makers to supervisors.

That is not the future Dr. Chang envisions because in his view, radiologists must define how AI integrates into their work. They must remain in the driver’s seat while intelligent agents eliminate friction, anticipate needs and execute micro-tasks in real time.

In this model, AI does not sit beside workflow, but it moves within it. Radiology has always adapted to disruption. Each technological leap has ultimately strengthened the field. AI is no different if it is integrated thoughtfully, deeply and collaboratively.

The future of radiology is not defined by simply combining human and AI. It will be defined by how seamlessly AI is embedded into the radiologist’s workflow—eliminating friction, reducing unnecessary steps, and enabling truly efficient, real-time collaboration. In this new model, human–AI collaboration is not an add-on. It is a fully integrated experience—designed to deliver meaningful efficiency and a better, more intuitive way to read.