Zühlke – Empowering Ideas

Life Science and Pharmaceutical Industry,  Medical Device and Healthcare

AO Foundation: Federated Learning for Privacy Preserving AI

Federated Learning Pharma
  • Cutting edge solution for major challenges in privacy preserving AI on highly sensitive medical data 
  • Involvement of key stakeholders facilitates future implementation across partner clinics 
  • From start to fully functional prototype in only three months

The AO Foundation and Zühlke codeveloped solutions for the semantic segmentation of spine computed tomographies (CTs) based on cutting edge and privacy preserving AI methodology.

Exploring the opportunities of Federated Learning

The goal of the AO Foundation is to enable best-in-class treatments within surgeries. AI offers a lot of opportunities, but there is one major challenge: to perform well, algorithms require a lot of data, which in the case of health data are often collected in silos (e.g. in different clinics) and furthermore often cannot be shared across clinics due to privacy reasons. Therefore, the AO Foundation teamed up with Zühlke to explore how Federated Learning could provide a solution to this challenge.

Concept and PoC after only three months 

Zühlke and the AO Foundation started with a series of workshops to shape the vision of a Federated Learning platform, its business case as well as the scope of a first prototype. Semantic segmentation of spine computed tomographies (CTs) was chosen as the first use case on the envisioned Federated Learning platform, due to its importance in surgery planning. Stakeholders at partner clinics were involved to establish the foundation for future implementation across clinics. In only three months, Zühlke developed a Federated Learning platform concept as well as a fully functional federated spine segmentation proof of concept based on a proprietary spine CT segmentation dataset.

Platform concept also covering organizational topics 

The proof of concept is based on a special purpose algorithm to train the semantic segmentation model in a federated, decentralized manner, without the need to share private data across clinics. In addition to the technical solution, a platform concept covering organizational topics related to partner clinics has been developed. After the evaluation of the model’s performance, the AO Foundation and Zühlke presented the results at the International Health Summit 2020.

Bardia Zanganeh
Contact person for Switzerland

Bardia M. Zanganeh

Director Business Development

Bardia M. Zanganeh is responsible for the Life Sciences and Healthcare practice in Switzerland. He serves leading healthcare institutions on all technology agenda issues. His primary areas of focus include digital innovation, business model transformation and product innovation. He also serves providers as well as medical technology and pharmaceutical companies. He has a background in engineering, consulting and entrepreneurship and is a lecturer at the University of Applied Sciences in Business Administration in Zurich.

Contact