![\"Radiology](\"https:\/\/rtmedical.com.br\/wp-content\/uploads\/2026\/05\/residentes-ia-radiologia.jpg\")
Among the 401 respondents, 13% ranked radiology as their top specialty choice. Only 2.5% said AI was “extremely influential” in shaping that decision. Another 57% acknowledged a “slight or moderate” impact, and 35% reported no influence at all. The picture shifts dramatically when filtered by interest in radiology: 91% of those interested in the field said AI influenced their thinking, compared with just 54% of those who were not.<\/p>\n
Polarization: Threat or Opportunity?<\/h2>\n
The most revealing finding is not the average, but the distribution. Among students drawn to radiology, 33% said AI discouraged<\/strong> them, 13% felt encouraged<\/strong>, and 33% reported no AI influence at all. The authors described this pattern as a “growing polarization”: part of the cohort sees AI as a job-security risk, part sees it as a clinical innovation opportunity.<\/p>\n One quantitative effect stands out: students who believe<\/em> AI will reduce demand for radiologists are 50% less likely to be interested in the specialty. The belief alone \u2014 independent of whether it reflects reality \u2014 is enough to redirect career paths. For a field already facing chronic workforce shortages in North America, Europe and Latin America, that is a strategic signal medical schools and professional societies cannot afford to ignore.<\/p>\n When IBM Watson made its high-profile debut at RSNA 2015, expectations were that AI would automate large parts of radiology within a few years. U.S. radiology residency applications dipped in the immediate aftermath. The trend then reversed: adoption proved far slower than predicted, very few hospitals turned core reading tasks over to algorithms, and commercial tools settled into a decision-support role rather than replacement. The renewed interest in radiology reflects that lived experience \u2014 and the survey numbers track that adjustment.<\/p>\n This evolution echoes ongoing discussions in our own coverage. In a recent study on uneven AI performance in chest radiography<\/a>, we showed that current algorithms still carry bias and generalization limits that make human radiologists indispensable in many real-world scenarios. We also explored the strategic questions service leaders must ask before adopting AI in our piece on five critical questions every radiology director should ask before deploying AI<\/a>.<\/p>\n For residency program directors and medical school deans, the study delivers a practical message: how AI is framed \u2014 as a tool versus as a replacement \u2014 directly shapes the talent pipeline. Curricula that show students how AI is integrated into real workflows in PACS, with human validation and clinical governance, tend to convert discouragement into technical curiosity. The authors recommend that schools include dedicated modules on applied imaging AI, addressing both capabilities and limitations.<\/p>\n The same logic applies internationally. In countries struggling with radiologist shortages, teleradiology expansion and rising demand for cross-sectional imaging, the next generation of clinicians must be fluent in both medicine and machine learning. Hiding AI from training only feeds the fear. Putting it in context creates the leaders the next decade will need.<\/p>\n The study is restricted to Canadian students, carrying cultural and structural assumptions specific to that health system. Replications in Brazil, the United States, Europe and Latin American countries would help confirm whether the polarization pattern is universal. Another caveat: measuring intention is not the same as measuring final career decisions \u2014 some respondents now discouraged may change course as they encounter imaging firsthand during clerkships.<\/p>\n Still, the signal is clear. The generation now entering medicine has lived with AI since high school. For them, the question is no longer “will AI exist?” but “how will I work with it?”. The radiology programs that answer that question convincingly \u2014 in lectures and at workstations \u2014 will attract the right students. The ones that don’t will only amplify the fear the study describes.<\/p>\n Department chairs and academic radiologists should treat the survey as an early warning rather than a confirmation that the fight is over. The 33% who say AI discouraged them are not a fringe \u2014 they are a third of the most motivated candidates, the ones most likely to consume technical content and shape future cohorts. Reaching them requires concrete artifacts: documented case studies where AI augmented rather than replaced the radiologist, transparent reporting of AI failures and false positives, and faculty role models who comfortably use AI tools in daily practice without anxiety.<\/p>\n Vendor partnerships also need to be structured to give residents access to validated real cases, not just commercial demonstrations. Training environments that let trainees see how AI alerts are triaged, accepted or rejected by an attending radiologist deliver a far more grounded view of the profession’s future than any sales pitch can. The choice facing radiology educators is not whether to discuss AI, but how honestly and how early to do so. The students are already forming opinions; the question is whose data those opinions are based on.<\/p>\nWhat Has Changed Since 2016<\/h2>\n
Implications for Medical Education<\/h2>\n
Limitations and Outlook<\/h2>\n
What Departments Should Do Next<\/h2>\n