Patients Prefer AI Breast Imaging Summaries

Researchers compared patient responses to radiologist-written breast imaging reports and summaries generated by ChatGPT-4 in a prospective, multicentre survey study. Ten anonymized reports—5 ultrasound and 5 mammography, covering BI-RADS categories 1–5 for each modality—were simplified to no more than 50 words. A panel of 20 physicians assessed the reports before 314 patients evaluated paired versions, producing 2,965 evaluations. Patients rated simplicity, comprehensibility and empathy on a 5-point Likert scale and chose a preferred report; the report’s source was hidden, and pair order was randomized.

Patients preferred the AI-generated version in 63.3% of evaluations (1,876/2,965). Ratings of 4–5 for simplicity were more common for AI summaries than radiologist-written reports (70.0% versus 43.4%); the same pattern held for comprehensibility (68.1% versus 40.4%).

Mixed-effects logistic regression found comprehensibility to be the strongest predictor of preference in both directions. A comprehensibility score of 4 for an AI report was associated with higher odds of preferring it (adjusted OR 1.87, 95% CI 1.08–3.23, p=0.025), while a score of 4 for a radiologist-written report was associated with lower odds (adjusted OR 0.45, 95% CI 0.29–0.69, p<0.001). Preference for AI summaries showed a non-significant upward trend with education level. The authors say the findings support supervised, complementary use of AI-generated summaries in clinical breast imaging.

Researchers compared patient responses to radiologist-written breast imaging reports and summaries generated by ChatGPT-4 in a prospective, multicentre survey study. Ten anonymized reports—5 ultrasound and 5 mammography, covering BI-RADS categories 1–5 for each modality—were simplified to no more than 50 words. A panel of 20 physicians assessed the reports before 314 patients evaluated paired versions, producing 2,965 evaluations. Patients rated simplicity, comprehensibility and empathy on a 5-point Likert scale and chose a preferred report; the report’s source was hidden, and pair order was randomized.

Patients preferred the AI-generated version in 63.3% of evaluations (1,876/2,965). Ratings of 4–5 for simplicity were more common for AI summaries than radiologist-written reports (70.0% versus 43.4%); the same pattern held for comprehensibility (68.1% versus 40.4%).

Mixed-effects logistic regression found comprehensibility to be the strongest predictor of preference in both directions. A comprehensibility score of 4 for an AI report was associated with higher odds of preferring it (adjusted OR 1.87, 95% CI 1.08–3.23, p=0.025), while a score of 4 for a radiologist-written report was associated with lower odds (adjusted OR 0.45, 95% CI 0.29–0.69, p<0.001). Preference for AI summaries showed a non-significant upward trend with education level. The authors say the findings support supervised, complementary use of AI-generated summaries in clinical breast imaging.