Women Don't Want AI as Sole Reader of Breast Cancer Screenings

 

• Less that 5% of survey respondents accepted the idea of stand-alone mammogram AI interpretation, while 71% supported AI as a second reader.
• About three-quarters of patients had minimal or no knowledge of AI.
• Patient race and education level were significantly associated with acceptance of AI.

Few women wanted artificial intelligence (AI) as the sole reader of their screening mammograms, according to a survey, though most respondents generally supported some use of the technology.

Among more than 500 women who presented for a screening mammography during a 6-month period in 2023, just 4% of survey respondents accepted the idea of stand-alone AI interpretation, whereas 71% were comfortable with AI as a second reader, reported Basak Dogan, MD, of the University of Texas Southwestern Medical Center in Dallas, and colleagues.

However, most patients reported that they had little or no knowledge of AI.

"Although there is an acceptance of AI, especially when used alongside a radiologist, there remain important concerns about its efficacy, potential bias, and the implications for patient-doctor interactions," they wrote in Radiology: Imaging Canceropens in a new tab or window. "As AI technology advances, it is necessary to continue engaging with patients to understand their evolving views and address any concernsopens in a new tab or window through education, transparency, and robust AI validation processes."

Overall, 518 patients who underwent screening at the University of Texas Southwestern Medical Center responded to the 29-question survey.

Most respondents (73%) were ages 40 to 69, two-thirds had attained college-level education or higher, and about half identified as non-Hispanic white. The vast majority (81%) of participants spoke English as their primary language, and 43% had a yearly income of more than $100,000.

Of the participants, 29% were premenopausal, 27% had at least one first-degree relative with breast cancer, 5% were previously diagnosed with breast cancer, 36% reported abnormal mammography results, and 23% had undergone a breast biopsy.

The survey showed that most patients reported they had minimal or no knowledge of AI, with 44% reporting they know "a little bit" about AI and 32% having no knowledge of AI.

About three-quarters of respondents (74%) said they believed consent was necessary before using AI for mammogram interpretationopens in a new tab or window. Participants with a yearly income exceeding $99,000 were less likely to consider permission necessary compared with those with a lower income.

When assessing AI's efficacy compared with radiologistsopens in a new tab or window, 5% believed AI to be "much worse," 21% said "worse," 43% said "same," 14% said "better," and 1% said "much better." There was no difference in the perception of AI's diagnostic accuracy versus a radiologist's based on respondents' age, race and ethnicity, educational attainment, or menopausal status.

Participants were approximately two times more likely to agree with AI usage in reading their mammogram if they had more than a college degree (OR 2.05, 95% CI 1.31-3.20, P=0.002) or a higher knowledge about AI (OR 2.31, 95% CI 1.51-3.53, P<0.001).

Concerning data privacy, 30% of surveyed individuals expressed moderate concern, and 36% were very or extremely concerned.

However, concerns about data privacy and potential AI biases were significantly higher among participants who were Hispanic (ORs of 2.41 and 3.32, respectively) and non-Hispanic Black (ORs of 2.87 and 4.31).

In addition, compared with non-Hispanic white participants, non-Hispanic Black participants were less likely to agree with the use of AI (OR 0.38, 95% CI 0.22-0.65, P=0.005).

"Our findings highlight the critical need to address racial and ethnic disparities within AI healthcare applications, revealing that participants, like researchers, are concerned about potential biases," the study authors observed.

The authors acknowledged several limitations to the study, including the fact that the survey was conducted at an academic center where most patients were covered by private insurance. Attitudes toward AI could vary depending on access to insurance and healthcare accessibility.

Additionally, Dogan and colleagues noted that AI is an inherently complex subject and that respondents might not fully have understood the concept, which could have influenced survey responses.

"However, we believe that the responses we obtained reflect patients' opinions in the absence of detailed prior knowledge, thus capturing perspectives representative of the general public," they said.

Disclosures

Dogan reported relationships with the NCI/NIH, Bracco Medical, Cancer Prevention Institute of Texas, Elsevier, Seno Medical, Massachusetts General Brigham Women's Hospital, and Mount Sinai Icahn Medical School, and for a patent related to predicting breast cancer metastasis.

Primary Source

Radiology: Imaging Cancer

Source Reference: opens in a new tab or windowOzcan BB, et al "Patient perception of artificial intelligence use in interpretation of screening mammograms: A survey study" Radiol Imaging Cancer 2025; DOI: 10.1148/rycan.240290.

https://www.medpagetoday.com/radiology/diagnosticradiology/115213