Abstract PS1-13-30: Concordance of Mammographic Studies Between Artificial Intelligence and Expert Breast Radiologists

Abstract

Abstract Background. Various artificial intelligence (AI) programs have been specifically designed and trained for breast cancer screening and prediction, and their use is steadily increasing. However, tools such as ChatGPT—an AI not originally developed for imaging analysis—are being increasingly employed for this purpose, particularly by non-medical personnel. This situation may generate mistrust and uncertainty among patients when comparing AI-generated reports with those of medical specialists. Therefore, the proper use of these tools requires awareness of their intended scope and limitations. Methods. Random mammograms performed in August 2025 on asymptomatic women undergoing preventive screening were included. All studies were interpreted and reported by expert breast radiologists. For AI evaluation, images were uploaded using an iPhone 16 Pro Max into ChatGPT, which generated a structured mammography report. Concordance was assessed across three parameters: breast density, BI-RADS classification, and recommendations. Results A total of 50 mammograms met the inclusion criteria. Concordance for breast density was 88%, with a predominance of pattern C in 65% of cases. BI-RADS classification showed 93% concordance, most commonly BI-RADS 2. Recommendations demonstrated 90% concordance, particularly regarding complementary breast ultrasound and annual follow-up when clinically indicated. Conclusions. Our findings suggest that ChatGPT, despite not being specifically designed for imaging interpretation, demonstrated a high level of concordance with expert breast radiologists. These results should be confirmed in larger case series to validate the potential role of non-imaging AI tools in breast cancer screening support. Citation Format: R. S. LIMON. Concordance of Mammographic Studies Between Artificial Intelligence and Expert Breast Radiologists [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS1-13-30.

Ähnliche Arbeiten