By Dr Amrita Kumar, Consultant Radiologist and AI Clinical Lead, Medica Group
AI is no longer arriving in radiology; it is already embedded across the entire workflow. It is optimising image acquisition and reconstruction at the scanner. It is triaging worklists, flagging critical findings and prioritising suspected bleeds and fractures before a radiologist opens a case. It is detecting, segmenting and quantifying pathology, including nodules, lesions, bone density and tumour burden, with tools that have regulatory clearance and real clinical deployment. Increasingly, it is also moving into the reporting process itself, structuring templates, drafting preliminary impressions and, in some settings, generating narrative reports that a radiologist then reviews and authorises. Generative AI is beginning to reshape not just what we look at, but what we say about it.
This is the landscape radiologists are already operating in; not a future scenario, but a present reality across NHS, private and international practice. The question is not whether you are ready for AI. It is whether you understand what your role has actually become within it.
The role has become more complex, not simpler
There is a comfortable narrative that AI will handle the routine, freeing radiologists for the complex. The reality is more nuanced. When an AI tool prioritises a worklist, quantifies a lesion or drafts a report impression, and you review that output and sign it off, you are not simply confirming a machine’s work. You are exercising clinical judgement in a system where the probabilities have already been shaped before you looked. That is a different cognitive task from reporting cold.
Thriving in this environment requires what I would call governance instinct: the ability to ask not just “is this finding right?” but “should I trust this tool’s output in this patient, in this clinical context, given what I know about how it performs?” In practice, this might mean recognising that a CT head triage tool is generating a pattern of false-positive bleeds in a specific patient cohort, such as elderly patients with chronic small vessel disease, where the AI has not been adequately trained to distinguish acute from chronic change, and escalating that concern before it erodes clinical trust in the system. That is not a technical skill. It is a clinical one, built from experience and a willingness to interrogate AI outputs rather than accept them.
Accountability is the radiologist’s strength
One of the most important things AI has clarified for our profession is something that was always true: the radiologist is accountable. AI systems do not hold clinical responsibility. We do. That is not a burden; it is the foundation of our professional authority.
This means understanding that regulatory approval is not the same as real-world clinical safety. A tool validated on a research dataset may perform very differently in your patient population, on your scanner hardware and within your reporting workflow. Real deployment demands active surveillance. In practice, this can mean identifying that the decision thresholds set by a vendor at the point of approval are not optimised for your population, and that recalibrating those thresholds for chest X-ray AI or mammography AI is what actually delivers the sensitivity and specificity your patients need. That kind of post-deployment adjustment is not unusual. It is necessary, and it requires a radiologist who understands both the clinical stakes and the performance metrics well enough to drive that conversation.
Communicating this to colleagues, clinical leads and multidisciplinary teams is increasingly part of the role. AI-supported findings need to be explained, contextualised and sometimes defended. That requires confidence in both the technology and its limits.
The mindset that matters most
The radiologists who will thrive are not necessarily those with the deepest technical knowledge. They are those who remain genuinely curious, who adapt workflows without surrendering clinical judgement and who understand that AI is not a destination; it is a continuously evolving system that requires continuous engagement.
Lifelong learning in this context means more than CPD checkboxes. It means staying close enough to how tools actually perform in practice to know when something is wrong, and having the professional confidence to say so.
The question is no longer whether AI will define radiology’s future. It will. The question is whether radiologists will be the ones shaping how it is used, or simply the ones reviewing its outputs. That choice, ultimately, is ours.
Explore our upcoming events to deepen your understanding of AI in radiology:
The Future AI Radiologist:
|
Making AI Work in Radiology:
|
AI in Healthcare:
|
 |
 |
 |
|
5 November 2026 |
Tuesday, 28 July 2026 |
On-demand |
| – Register here | – Register here | – Register here |