Artificial intelligence (AI) has emerged as a transformative technology in healthcare, particularly in radiology. Its ability to analyze large volumes of medical imaging data with speed and precision has positioned AI as a promising adjunct to human radiologists, offering potential improvements in diagnostic accuracy and workflow efficiency. Mammography, with its standardized imaging protocols and abundant structured data, is particularly well-suited for AI integration. However, while technical feasibility and performance metrics have been studied extensively, patient perspectives remain a critical yet underexplored factor influencing successful AI implementation in clinical practice. A recent comparative survey study published in Breast Cancer Research and Treatment by Ogu et al. (2025) sheds light on patient attitudes toward AI in mammography, focusing on differences between academic and safety-net hospital populations.
The study aimed to evaluate and compare patient perceptions of AI use in mammogram interpretation across two healthcare settings: an academic hospital (ACH) and a safety-net hospital (SNH). Safety-net hospitals primarily serve socioeconomically disadvantaged populations and provide care regardless of a patient's insurance status or ability to pay. The research team developed a 29-item survey targeting six core areas: proof of technology, procedural knowledge, competence, efficiency, personal interaction, and accountability. The survey also collected demographic information, including age, race/ethnicity, education level, and income, along with participants' self-reported knowledge of AI and relevant medical history.
Surveys were administered to patients presenting for screening or diagnostic mammograms at the two hospital sites. The academic hospital collected data from February to August 2023, while the safety-net hospital conducted surveys from April to June 2024. Participation was voluntary, anonymous, and available in English and Spanish, with medical translation services provided when necessary.
A total of 924 patients completed the survey, including 518 from ACH (56.1%) and 406 from SNH (43.9%). Participants were predominantly between 40 and 69 years of age, with the majority identifying as Hispanic (33%), Non-Hispanic White (31.3%), or Non-Hispanic Black (21.6%). Educational attainment varied significantly between the two populations; 70.5% of ACH participants had at least a college degree compared to 19.6% at SNH. Income disparities were also notable, with 48.6% of ACH participants earning $100,000 or more annually versus only 3.16% at SNH. Self-reported AI knowledge was higher among ACH participants, and prior experience with mammography differed, with SNH having a higher proportion of first-time patients.
Overall, 71.5% of participants expressed support for AI use, particularly as a secondary reader alongside a radiologist. Only 6.6% favored AI as a stand-alone interpreter, underscoring the strong preference for human oversight. When comparing hospital settings, ACH participants initially showed higher acceptance of AI, but after adjusting for demographic factors such as age, race, education, and AI knowledge, the difference between ACH and SNH populations was not statistically significant. This indicates that demographic variables rather than institutional affiliation largely drive acceptance levels.
Higher education and self-reported knowledge of AI were independently associated with greater acceptance. Conversely, Non-Hispanic Black participants were significantly less likely to support AI use, highlighting the influence of race and potentially reflecting broader concerns regarding equity and trust in medical technology.
Patient perceptions of AI efficacy were nuanced. While 41.9% rated AI as comparable to radiologists, 18.5% rated AI as superior in detecting breast cancer. Notably, SNH participants were more likely to rate AI as same or better than a radiologist, yet they also expressed a greater desire for follow-up readings by AI if a radiologist detected an abnormality. Across both populations, participants overwhelmingly preferred radiologist review for AI-identified abnormalities, reflecting persistent trust in human expertise.
The survey highlighted patient concerns about accountability in cases of diagnostic errors. Over 50% of participants believed the AI developer, hospital, and doctor should all share responsibility for false negatives, while nearly 48% held all parties accountable for false positives. Interestingly, SNH participants were less inclined to include AI developers in assigning blame, indicating nuanced differences in perceived responsibility.
Ethical and privacy concerns were widespread. Between 81.5% and 90.9% of participants expressed worry regarding data privacy, algorithmic bias, technological accuracy, transparency, and potential impacts on the doctor-patient relationship. Data privacy concerns were particularly pronounced in the SNH cohort, and Non-Hispanic Black participants were more likely to report privacy-related apprehensions. These findings suggest that patient trust in AI is closely intertwined with ethical considerations and the need for transparent communication.
The study underscores the importance of integrating patient perspectives into AI deployment strategies. Although there is general support for AI-assisted mammography, patient preferences emphasize human oversight and informed consent. Addressing disparities in acceptance requires culturally sensitive educational interventions, particularly for populations with lower educational attainment or historical mistrust in medical systems. For example, Non-Hispanic Black patients demonstrated lower acceptance and higher privacy concerns, indicating a need for targeted outreach and engagement to build confidence in AI technology.
Moreover, the findings highlight that patient perceptions are influenced by a combination of demographic factors, prior experience with healthcare, and knowledge about AI. Educational initiatives that enhance understanding of AI capabilities, limitations, and safeguards may improve acceptance, while transparent policies regarding accountability and data security can foster trust.
Several limitations of the study merit consideration. The research was conducted at a single academic hospital and one safety-net hospital in the United States, which may limit generalizability. Surveys were administered at different times, potentially introducing temporal bias due to evolving public awareness of AI. The survey instrument was not pre-tested or validated, and some responses were incomplete or inconsistent. Additionally, language limitations may have excluded patients requiring translations beyond English or Spanish. Despite these constraints, the study provides valuable insights into patient attitudes toward AI in mammography across demographically diverse populations.
Ogu et al.'s study highlights strong patient support for AI integration in mammography, particularly as a secondary reader to assist radiologists. Patient concerns regarding privacy, bias, accuracy, and transparency emphasize the need for ethical safeguards, effective communication, and patient-centered educational strategies. Differences in acceptance between demographic groups underscore the importance of culturally tailored interventions to address specific concerns and build trust.
As AI technologies continue to evolve, ongoing research should monitor changes in patient attitudes and expand studies to diverse clinical and geographic settings. Understanding the factors that influence acceptance will be crucial for developers, clinicians, and policymakers aiming to implement AI responsibly and effectively in healthcare. Ultimately, successful AI integration requires a balance between technological advancement and patient-centered care, ensuring that innovations enhance diagnostic accuracy without compromising trust, equity, or autonomy.
Disclaimer
This blog summarizes research findings for informational purposes only and is not intended as medical advice. Patient perspectives reported here are based on a specific survey study and may not represent all populations. Decisions regarding the use of AI in mammography should be made in consultation with qualified healthcare professionals and consider individual clinical circumstances.
Most Accurate Healthcare AI designed for everything from admin workflows to clinical decision support.