Artificial intelligence in healthcare must be more than accurate, it must be interpretable, equitable, and safe. In this work, I present real-world implementations of responsible machine learning models designed to support, not replace, clinical decision-making in ophthalmology.
Across three clinical contexts: (1) neonatal screening for retinopathy of prematurity (ROP), (2) early detection of vision abnormalities in children with Neurofibromatosis type 1 (NF1) using OCT imaging, and (3) population-level prediction of proliferative diabetic retinopathy (PDR) from routine lab tests, we developed transparent, locally validated models that prioritize safety and fairness. Our ROP model preserved 100% sensitivity while reducing unnecessary examinations by 80%. In NF1, OCT-derived features predicted vision abnormalities with AUC 0.97, enabling earlier detection in children who struggle with behavioral vision testing. In diabetic retinopathy, we identified differences in risk patterns across sex and ethnicity, highlighting the importance of equity-aware modeling.
This work demonstrates how data science can move beyond performance metrics toward responsible, interpretable clinical decision support, reducing burden, improving early intervention, and advancing more personalized and equitable care.
Dr. Ayelet Goldstein is a Senior Lecturer in Computer Science and Director of the M.Sc. Program at the Jerusalem Multidisciplinary College. Her research focuses on interpretable machine learning for clinical decision support in Medicine. She collaborates with medical centers in Israel, Spain, and the United States to translate AI models that improve early detection, risk stratification, and equitable care. Her work bridges data science and medicine to translate algorithms into real clinical impact.
Keynote session: Hadas Grossmon Ella
Break
Lightning talks session
Roundtable closing
Talk by Hila Paz
Talk by Dr. Moran Mizrahi
Closing remarks
End
