S335
Clinical - Breast
ESTRO 2026
Eline E.F. Verreck 1 , Wilma D. Heemsbergen 2 , Thijs van Dalen 1 , Sandra Windhorst 3 , Linda de Munck 3 , Femke van der Leij 4 , Liesbeth J. Boersma 5 , Femke E. Froklage 2 1 Department of Surgical Oncology, Erasmus University Medical Centre, Rotterdam, Netherlands. 2 Department of Radiation Oncology, Erasmus University Medical Centre, Rotterdam, Netherlands. 3 Department of Research and Development, Netherlands Comprehensive Cancer Organisation, Utrecht, Netherlands. 4 Department of Radiation Oncology, University Medical Centre Utrecht, Utrecht, Netherlands. 5 Department of Radiation Oncology Maastro, 5Grow Research Institute for Oncology and Reproduction- Maastricht University Medical Centre+, Maastricht, Netherlands
(low risk) or positive (medium or high risk). All positive cases (as determined by human or AI readings) underwent further investigation at specialized
facilities. Results:
There was concordance between human and AI readings for 2,339 breasts (88%), predominantly for negative findings (2,268 cases; 97%). Discordance occurred in 12% of cases, with human positive/AI negative readings in 40% and human negative/AI positive readings in 60%.For concordant positive readings (71 breasts), ultrasound was suspicious in 26 cases (37%). Biopsies were performed in 27 patients, including 21 with suspicious ultrasound findings (18 positive biopsies) and six with non-suspicious ultrasound but high-risk AI findings (four positive biopsies).For discordant readings with human positive/AI negative results (124 cases), seven ultrasounds were suspicious, leading to four biopsies, all negative. For human negative/AI positive results (189 cases), 14 ultrasounds were suspicious. Biopsies were performed in all 14 cases, with nine yielding positive results. Conclusion: AI serves as a complementary tool in mammographic interpretation, demonstrating a high negative predictive value (no positive biopsies in low-risk AI cases). Moreover, AI identified breast cancer in 15 women who would not have undergone biopsies based on human interpretation alone, highlighting its potential to enhance diagnostic accuracy in breast cancer screening programs. Keywords: Breast cancer screening, IA Towards objective cosmetic self-evaluation: machine-learning on photographs from breast conserving treatment with validation on external trial dataset Morten Sahlertz 1,2 , Ida R. Johannsen 1,3 , Mette H. Nielsen 4 , Else Maae 5 , Marie L. H. Milo 6 , Mechthild Krause 7 , Tine Engberg Damsgaard 8,9 , Mette E. Brunbjerg 10 , Mark A. Sydenham 11 , Fay H. Cafferty 11 , Jaymini Patel 11 , Lucy Kilburn 11 , Judith M. Bliss 11 , Charlotte E. Coles 12 , Adrian M. Brunt 11,13 , Cliona Kirwan 14 , Anna M. Kirby 15 , Birgitte V. Offersen 3,16 , Jasper Nijkamp 1,2 1 Department of Clinical Medicine, Aarhus University, Aarhus, Denmark. 2 Danish Centre for Particle Therapy, Aarhus University Hospital, Aarhus, Denmark. 3 Experimental Clinical Oncology, Aarhus University Hospital, Aarhus, Denmark. 4 Department of Oncology, Odense University Hospital, Odense, Denmark. 5 Department of Oncology, University Hospital of Southern Denmark, Vejle, Denmark. 6 Department of Proffered Paper 4746
Digital Poster Highlight 4743 Boosting breast cancer detection: The impact of AI in mammography screening Jamel Daoud 1 , Wafa Mnejja 1 , Rim Trigui 1 , Mariem Frikha 1 , Nejla Fourati 1 , Mounir Frikha 2 1 Radiotherapy Department Habib Bourguiba Hospital, Faculty of Medicine University of Sfax, Sfax, Tunisia. 2 Medical Oncology, Private practice, Sfax, Tunisia Purpose/Objective: Implementing a mass organized breast cancer screening program in a developing country presents numerous challenges, particularly financial and human resource constraints. Over recent years, the integration of artificial intelligence (AI) in medicine, and specifically in radiology, has grown significantly. This study aims to assess the contribution of AI-assisted mammography interpretation in breast cancer screening. Material/Methods: The study included 1,326 women (2,652 breasts) who underwent mammographic screening for breast cancer between January and December 2023. Mammograms were analyzed independently by two radiologists and an AI software. Based on the radiologists' assessment, tests were classified as negative (ACR 1 or ACR 2) or positive (ACR 0, ACR 3, or ACR 4). Using AI, tests were categorized as negative
Made with FlippingBook - Share PDF online