Recent survey results show that patients largely support the use of artificial intelligence (AI) to help radiologists interpret mammograms. The research results published in Breast cancer research and treatmentalso showed that acceptance varied in relation to factors such as age, race, and education.
“This is the first study to measure patient perspectives on AI in mammography in different hospital settings,” said the corresponding author. Dr. Basak DoganEugene P. Frenkel, MD, Clinical Medical Scientist. Professor of Radiology and Director of Breast Imaging Research. Member of the Harold C. Simmons Comprehensive Cancer Center at UT Southwestern. “We reveal how demographic and socio-economic factors shape acceptance, trust and concern about the integration of AI in breast cancer screening.”
Methods and participant demographics
A 29-item survey on patient perceptions of the use of AI in mammography interpretation was conducted at UT Southwestern and provided to all patients who visited the Parkland Health public safety-net health system between April and June 2024 and the William P. Clements Jr. University Hospital Breast Imaging Clinic between February and August 2023. The researchers analyzed differences in opinion by patient factors using odds ratios (OR).
Survey responses included 518 participants who visited a university hospital and 406 participants who visited a safety net clinic. People who visited university hospitals tended to be older. Mostly non-Hispanic whites. Their income and education level were high. Higher self-reported knowledge of AI (P < .001 on all factors) compared to safety net clinic visitors.
Main findings
Most survey respondents (71.5%) accepted the use of AI in mammogram interpretation, but participants in safety-net health systems were less likely to accept than participants in academic hospitals (OR = 0.71; 95% confidence interval) [CI] = 0.53–0.96; P = .02). Only 6.6% of all participants endorsed AI as the sole reader of mammograms.
Almost 60% of respondents said they would rather wait longer for a radiologist reading than rely on an immediate AI reading, confirming that participants value human oversight. 84% of respondents wanted a radiologist to review mammograms that AI identified as abnormal, while 44% of respondents wanted AI to review scans flagged by a radiologist.
However, participants in safety-net health systems were more likely to request an AI to read their mammogram after a radiologist identified an abnormality on the scan (OR = 1.83; 95% CI = 1.35-2.49; 95% CI = 1.35-2.49; P < .001), rating AI to be as good or better than radiologists in detecting cancer (OR = 1.54; 95% CI = 1.12-2.15; P = .01) and are more concerned about data privacy (OR = 1.87; 95% CI = 1.22–2.93; P = .01) than participants from university hospitals.
A total of 73.8% of survey respondents said they would like to be informed or consented before using AI to read their mammogram. At least 80% of participants reported being concerned about issues such as privacy, bias, accuracy, transparency, and doctor-patient relationships.
“With the increasing use of AI in breast image interpretation, care must be taken to educate patients about the role of AI, obtain consent for its use, and provide safeguards to protect data privacy,” said the study leader. Dr. Emily KnippaShe is an associate professor of radiology and a member of the breast imaging department at UT Southwestern.
Factors associated with higher acceptance of AI use include higher education (OR = 1.99; 95% CI = 1.33-2.99; P < .001) and self-reported knowledge of AI (OR = 1.98; 95% CI = 1.38–2.83; P < .001). On the other hand, non-Hispanic blackness was associated with lower acceptance of AI use (OR = 0.40; 95% CI = 0.25-0.65; P < .001).
Disclosure: This research was funded by Dr. Dougan Eugene P. Frenkel, MD, who received the Clinical Medicine Award from Simmons Cancer Center. For full study author disclosure, please visit link.springer.com.
