Different reliance on ER signalling for replication and disease heterogeneity

A big thank you to #LeoCeli and @mitcriticaldata.bsky.social for their consistent support in our research on AI #bias and disparities in breast cancer-related subfields. 🙏🧠

-> Model accuracy is limited in time and space. AI should be local, iterative, reflective and reflexive. The business model should be assisting health systems with creating, maintaining, curating, evaluating their data pipeline, and most importantly, their "learning pipeline".

3.we also observed #gender imbalance among first and last authors (male predominance), with #affiliations predominantly from high-income regions.

2.nearly all patient cohorts originated from high- #income countries, with no representation from low-income settings.

1.despite a ~300% increase in #AI-related publications between these periods, we found that only 0-25% of studies reported #race / #ethnicity data, with the majority of patients identified as Caucasian.

In our latest study, we identified 264 studies on AI-powered #mammography interpretation for #breastcancer detection out of 5774 in the two-year periods 2017-2018 and 2022-2023: