Swarm and federated learning
Train models jointly while each hospital retains control of its patient data.
Saldanha Lab · Junior Research Group
Distributed Learning & Multimodal AI
We develop methods to train medical AI across hospitals while keeping patient data at the institution where they were collected.
Research
Learning across hospitals
Our work spans swarm and federated learning in radiology, pathology, single-cell data, and surgical video.
We train multimodal models that combine these sources and develop the infrastructure needed for reliable distributed training across institutions.
Patient data remain at each participating site. Models learn from local data and exchange only the information needed for collaborative training.
Methods
Distributed learning as a research problem
Hospital networks differ in their patients, cohort sizes, data systems, and technical environments. We study how these differences affect training and reproducibility.
Alongside overall network performance, we examine the models produced at individual sites. This makes local variation visible instead of reporting pooled averages alone.
Multimodal models
Combine radiology, pathology, single-cell measurements, and surgical video.
Site-level evaluation
Measure performance at every participating site as well as across the full network.
Reliable infrastructure
Build reproducible training workflows that operate across different hospital IT environments.
Research highlights
Selected work and collaborations
Our work combines methodological research, open infrastructure, and multi-institutional studies.
Nature Medicine · 2022
Swarm learning for decentralized artificial intelligence in cancer histopathology
Nature Biomedical Engineering · 2026
Benchmarking foundation models as feature extractors for weakly supervised computational pathology
NEJM AI · 2026
Privacy-Preserving Surgical Video Analysis with Swarm Learning
Horizon Europe
ODELIA
Open-source swarm learning for medical AI across a pan-European network.
German Cancer Aid
DECADE
Distributed AI for diagnosis, prognosis, and response prediction in colorectal cancer.
International network
Collaborations across more than ten countries
Multi-site studies in Europe and beyond.
Team
Group members
Dr. Oliver Lester Saldanha
Junior Principal Investigator
Karthik Durai
PhD student
Open position
Research group member
Open position
Research group member
Open position
Research group member
Collaborate
Start a multi-site study
We work with clinical departments, research groups, and multicentre consortia and welcome new partner sites. Participation does not require patient data to be transferred to another institution.
contact@kather.ai