Main project

SPEARHEAD

Federated learning for antibiotic resistance prediction

• Federated learning
• Healthcare
• Life sciences

Federated learning: models are trained locally at each institution and only model updates are shared with a coordinating server.

Challenge

Antimicrobial resistance makes many treatments ineffective against infections that were previously easy to treat. A more appropriate use of antibiotics, only when necessary and with narrow-spectrum drugs whenever possible, is one of the main measures to limit its spread. Clinical data that could support this decision are distributed across hospitals and cannot easily be pooled.

Approach

SPEARHEAD brings together a large consortium to apply the lessons learned during the COVID-19 response to antimicrobial resistance. For the data-driven part, focused on bacterial urinary tract infections, our group develops machine learning models trained with federated learning through FLOTTA, the federated learning framework developed at IDSIA. Each hospital keeps its data on site and only model updates are exchanged.

Outcome

The models are intended to be accessible through a web interface, where clinicians enter patient characteristics and receive suggestions on whether to prescribe an antibiotic and, if so, which ones are most appropriate.

Publications

• Flotta: a Secure and Flexible Spark-inspired Federated Learning Framework

  C. Bonesana, D. Malpetti, S. Mitrović, F. Mangili, L. Azzimonti

  IEEE International Conference on Federated Learning Technologies and Applications (FLTA 2024), 156–161, 2024PreprintCode

• Global Outlier Detection in a Federated Learning Setting with Isolation Forest

  D. Malpetti, L. Azzimonti

  IEEE International Conference on Federated Learning Technologies and Applications (FLTA 2024), 251–258, 2024Preprint

Software

• FLOTTA

  Secure and flexible federated learning framework, inspired by Apache Spark, for training models across institutions without sharing data.

  GitHubDocumentationPaper