This study surveyed 249 medical students at West China Hospital to understand their attitudes toward using artificial intelligence in diagnosing kidney cancer. The students came from two different specialties: urology (which deals with the urinary system and kidneys) and pathology (which involves examining tissue samples under the microscope).
The researchers were especially interested in how students viewed the recently updated WHO 2022 classification of kidney cancers, which introduced new subtypes and changed how several cancers are defined. Understanding both the new classification and AI tools will be important for the next generation of doctors treating kidney cancer patients.
Published in BMC Medical Education in 2025, this paper provides a rare look at how future clinicians perceive AI technology and highlights important gaps in medical education that need to be addressed as AI becomes more common in hospitals.
Kidney cancer is not a single disease. There are many different subtypes, each with distinct biology, behavior, and response to treatment. The WHO 2022 classification updated and expanded the previous list of recognized kidney cancer types, adding new subtypes such as those defined by FH-deficient RCC, SMARCB1-deficient RCC, and tumors driven by TFE3 gene rearrangements.
These new categories are important because they help doctors match patients to the most appropriate treatments, including targeted therapies and immunotherapies that work for specific tumor types. Without understanding which subtype a patient has, treatment decisions may be less precise.
For medical students learning to diagnose and treat kidney cancer, staying current with these rapidly evolving classification systems is challenging. Adding AI tools to the mix makes this even more complex, since AI is now being used to help identify these subtypes from imaging and tissue samples.
The research team developed a questionnaire containing 16 items measured on a Likert scale, where students rated their agreement with each statement from strongly disagree to strongly agree. The questionnaire covered four main areas: knowledge of AI in medical imaging, attitudes toward AI in clinical practice, concerns about relying on AI, and views on incorporating AI into the medical curriculum.
All 249 students were enrolled at West China Hospital, one of China's largest and most respected academic medical centers. The two groups, urology students and pathology students, were compared to see whether their specialty training influenced their views on AI and kidney cancer diagnosis.
The survey was conducted in 2024 and analyzed using standard statistical methods to compare responses between the two groups and identify patterns in student attitudes across different areas of concern.
Among urology students, 74.1% reported being familiar with AI tools used in medical imaging. This relatively high awareness likely reflects the fact that AI-assisted CT and MRI analysis is already becoming part of urological practice. In contrast, pathology students showed slightly more cautious attitudes, particularly around whether AI has been adequately validated for clinical use.
More than 60% of students across both groups expressed concern about over-reliance on AI, a phenomenon sometimes called automation bias, where clinicians trust an AI output without applying their own critical judgment. This concern is shared by many experts in medical AI and is considered one of the most important risks to manage as these tools are adopted.
Nearly 70% of students said they wanted AI to be integrated into their medical curriculum. This strong demand signals that future doctors are eager to learn how these tools work, how to interpret their outputs, and when to question them. Students appear to recognize that AI literacy will be an essential professional skill, even if they have reservations about how AI is currently being used.
While both groups showed broadly positive interest in AI, their specific concerns differed in meaningful ways. Pathology students were more likely to question whether AI systems had been rigorously validated before being used to analyze tissue samples. This skepticism is understandable given that pathology involves subtle, expert-level interpretation of cellular features, and errors can directly affect cancer diagnosis and treatment.
Urology students were more likely to view AI as a practical tool that could help with imaging interpretation, suggesting that exposure to radiology-based AI in urology training shapes a more pragmatic attitude. However, even in this group, concerns about accountability, meaning who is responsible when an AI makes an error, were frequently raised.
These differences highlight that AI education may need to be tailored to the specific clinical context of each specialty, rather than offering a one-size-fits-all curriculum. The risks, benefits, and appropriate uses of AI differ between reading a scan and examining a biopsy slide.
The findings make a strong case that AI literacy should be a formal part of medical training, not just an optional topic. As AI tools become embedded in radiology software, pathology platforms, and clinical decision support systems, doctors who have never been taught how these systems work will be poorly equipped to use them safely.
Teaching students not just how to use AI but how to critically evaluate its outputs is particularly important. Understanding when an AI tool is likely to be reliable and when it might fail, for example in rare cancer subtypes underrepresented in training data, requires specific knowledge that goes beyond simply learning to click buttons in a software interface.
For patients, this research matters because the quality of AI-assisted diagnosis ultimately depends on the clinicians interpreting the results. If future doctors are well-trained in both the new WHO 2022 kidney cancer classification and the AI tools designed to help identify these subtypes, patients with kidney cancer are more likely to receive accurate diagnoses and appropriate treatments tailored to their specific tumor biology.