For years, the debate around AI in radiology has been framed in the biggest possible terms. Will it replace radiologists? Will it outperform specialists? Will it make diagnostic imaging faster, cheaper, and more accurate?
Those questions matter. But they can distract from a more immediate and practical opportunity. The first truly scalable use of AI in imaging may not be autonomous diagnosis. It may be helping patients understand the imaging information they already receive.
That is a larger problem than many healthcare leaders realize.
Patients increasingly see their results before they speak with a clinician. The 21st Century Cures Act accelerated that shift by requiring immediate electronic access to many kinds of health information. In a survey of 8,139 patients, 96% said they wanted to continue receiving immediately released test results online even if a clinician had not yet reviewed them. That is a powerful vote for transparency. But transparency alone does not create understanding.
Radiology reports were never written for lay readers. A study found that only a very small share of radiology reports in a large U.S. health system were readable at the eighth-grade level, which is roughly the reading level of the average American adult. The rest were filled with dense terminology, acronyms, sequence names, and compressed clinical reasoning. That may work for specialist-to-specialist communication. It does not work nearly as well for patients trying to understand what is happening inside their own bodies.
CT and MRI make this challenge even harder. These reports often combine anatomy, technical imaging language, incidental findings, and a differential diagnosis in a format that assumes medical training. For many patients, the result is not clarity but confusion.
None of this happens in an emotional vacuum. Waiting for imaging results is stressful, especially in oncology, neurology, and other high-stakes settings where a scan may shape a major treatment decision. One study found that patients generally expected outpatient imaging results within one to three days, and 45% reported an emotional change while waiting, most commonly anxiety. In practice, the patient experience is not simply access to results. It is access to complex results, under stress, often before explanation arrives.
That is the context in which patient-facing imaging AI becomes important.
What has changed is not that AI is now ready to diagnose patients independently. It is not. But open medical models are becoming capable enough to act as translation layers between technical imaging data and patient understanding. Google’s MedGemma 1.5 model card is notable here because it expands support for interpreting three-dimensional CT and MRI volume representations, not just 2D medical images. It can also generate text outputs such as answers, image analysis, and summaries. On Google’s published benchmarks, MedGemma 1.5 4B improved over earlier versions on internal CT and MRI classification tasks.
That is a meaningful shift. It suggests that medical AI is moving beyond generic chatbot behavior toward modality-aware imaging support.
This does not mean AI is ready to replace radiologists, or that it should. In fact, the more useful question is different: should healthcare organizations build a patient explanation layer around imaging, with appropriate guardrails?
Used well, that layer could do several things the current system does poorly. It could translate jargon into plain language. It could explain what a sequence is and why it matters. It could answer follow-up questions about anatomy, report structure, and common next steps. It could help patients prepare for more informed conversations with their doctors. Most importantly, it could meet patients at the exact moment when portals deliver information but workflows do not yet deliver context.
That matters for clinicians, too. Immediate release of test results has consequences for workflow. In one analysis, daily patient messages sent within six hours of result review increased substantially after transition to Cures Act compliance. If health systems are going to continue opening the front door to raw results, they also need to invest in the explanatory infrastructure behind that door.
The key is to use AI in the right role. Healthcare does not need to hand diagnosis over to AI for this to be valuable. A safer approach is to use AI as a patient education layer that sits alongside, not instead of, the official radiology report. That means telling patients clearly when the system is uncertain, avoiding treatment advice, flagging findings that need human follow-up, and reviewing how the tool performs in the narrow role it is meant to serve: helping people understand their imaging, not making clinical decisions for them.
Healthcare has spent too much time asking whether AI can read scans like a doctor. The more urgent question is whether it can help patients understand what their portal already shows them.
That is not a small problem. It is the one patients feel first.
About Peter Nemeth
Peter Nemeth is the founder of ReadYourLab.com, an enterprise software architect, and a cancer survivor. His work focuses on how AI can help patients better understand complex CT and MRI findings and prepare for more informed conversations with their doctors.
