On 5 November, Oliver Diaz of the University of Barcelona represented the EU-funded RadioVal project at a webinar that brought together ODELIA, RadioVal and BreastSCan to explore how AI can support breast cancer screening, diagnosis and care.
In his talk, The Pain Behind Trustworthy AI: From Data to Clinical Applications, Diaz highlighted what it takes to develop and validate AI tools for use across different healthcare settings. He shared three takeaways from RadioVal’s work.
- Trustworthy AI starts with trustworthy data. Teams need to know where their data came from, assess its quality and check how well it represents the patients the tool is intended to serve.
- Multicentre data still needs to be checked for representativeness. Adding hospitals and patients can broaden a dataset, but teams still need to examine which patient groups are represented and evaluate how the model performs across clinical settings.
- Trustworthiness needs to be built in from the start. Data governance, standardisation, quality control and clinical validation should be planned from the earliest stages of development.
The webinar was moderated by Derya Yakar of University Medical Center Groningen. Carlos Telleria Orriols of the Instituto Aragonés de Ciencias de la Salud discussed federated infrastructure for breast cancer screening, and Jakob Nikolas Kather of University Hospital Heidelberg spoke about building AI models for breast MRI that generalise across settings.
You can watch recording of the webinar here