Caen–Nagoya Alliance for AI in Health Data — a Franco-Japanese federated-learning pilot predicting cardiovascular adverse events of hormone therapy in prostate cancer, using the hospital data warehouses at Caen University Hospital (France) and Nagoya University Hospital (Japan). The pilot is designed to seed a wider federated network of additional partner hospitals.
Only model parameters cross the border. Patient data never moves. FedBioMed · OMOP-CDM · designed for GDPR + APPI conformity.
Public site: https://canal-ai.org/ — trilingual EN / 日本語 / FR.
Research-use-only pilot. No patient data has been processed to date. Caen Scientific & Ethics Committee favourable opinion (30 March 2026); Nagoya IRB and GDPR DPIA in progress.
| Framework | FedBioMed — INRIA / Université Côte d'Azur |
| Primary model | PyTorch MLP on OMOP tabular features, trained with FedAvg (FedProx fallback) |
| Baseline | Federated logistic regression via FedBioMed's scikit-learn training plan (FedSGDClassifier) |
| Privacy | Joye-Libert secure aggregation; OPACUS differential privacy evaluated |
| Endpoint | 4-point MACE within 12 months of ADT initiation (HERO definition) |
| Evaluation | Leave-one-site-out — AUROC, AUPRC, calibration (Brier, ICI), DCA |
| Funder | SECOM (first supporter) |
CANAL-AI/
├── README.md ← you are here
├── LICENSE ← code MIT + docs CC BY 4.0
├── .gitignore ← patient data + internal docs + rosters
├── CANAL-AI.Rproj ← RStudio project file
│
├── site/ ← public GitHub Pages site (HTML + CSS + JS)
│ ├── index.html
│ ├── sitemap.xml
│ ├── robots.txt
│ └── assets/
│ ├── style.css
│ ├── i18n.json ← EN / JA / FR dictionary
│ ├── i18n.js ← vanilla-JS language runtime
│ ├── photos/ ← hero banner, team portraits (PNG + WebP)
│ └── logos/ ← institutional logos (PNG + WebP)
│
├── .github/workflows/
│ └── pages.yml ← deploys site/ to GitHub Pages
│
├── docs/ ← INTERNAL — gitignored, not public
├── admin/ ← INTERNAL — gitignored, not public
│
├── R/ ← R code for LOCAL analysis at each node
│ ├── 00_setup.R
│ ├── 01_data_import.R
│ ├── 02_harmonization_omop.R
│ ├── 03_cohort_definition.R
│ ├── 04_descriptive_table1.R
│ ├── 05_local_eval.R
│ ├── 99_utils.R
│ └── README.md
│
├── python/ ← FedBioMed side (Python-based)
│ └── README.md
│
├── data/ ← NEVER commit patient data
│ ├── README.md
│ └── .gitkeep
│
└── output/ ← generated artefacts
├── README.md
└── .gitkeep
| Window | Milestones |
|---|---|
| Q2–Q3 2026 | Nagoya IRB submission · DPIA · Caen data-access convention · FedBioMed kickoff with INRIA/UCA |
| Q4 2026 | OMOP variable list locked · Cohort definition frozen · Synthetic-data prototype end-to-end |
| Q1–Q2 2027 | First federated round · MLP baseline with secure aggregation · Sample-size and power report |
| Q3–Q4 2027 | Model comparison · Methods paper submitted · APPI–GDPR blueprint published (CC BY 4.0) |
| 2028 | External validation at a third site · Clinical paper · Open-source release of full pipeline |
- Dr Charles Dolladille — MD, PhD, cardiologist and pharmacologist. MCU-PH, Caen University Hospital; INSERM U1086 ANTICIPE. ORCID · LinkedIn.
- Dr Basile Chrétien — PharmD, MPH, MSc, pharmacologist. PhD candidate, Nagoya University Graduate School of Medicine. On leave from Caen University Hospital. ORCID · LinkedIn.
- Dr Kazuki Nishida — MD, PhD, biostatistician with machine-learning expertise, Nagoya University. ORCID · LinkedIn.
- Nagoya University — Graduate School of Medicine (Japanese side).
- Nagoya University Hospital — clinical and data-warehouse partner.
- Caen University Hospital — Entrepôt de Données de Santé, Department of Pharmacology, INSERM U1086 ANTICIPE.
- Université de Caen Normandie — French institutional partner.
- FedBioMed — INRIA / Université Côte d'Azur (methodology, in-kind).
Further collaborators are acknowledged on the public site; internal planning is tracked in gitignored documents.
The trilingual landing page is deployed by .github/workflows/pages.yml
on every push to main that touches site/**.
To enable after first push:
- Settings → Pages
- Source: GitHub Actions
- Push to
main— the workflow publishes to https://canal-ai.org/.
Local preview:
python -m http.server --directory site 8080
# http://localhost:8080Language override via URL hash: …#lang=ja or …#lang=fr. Preference
is persisted in localStorage.
- Open
CANAL-AI.Rprojin RStudio. - Install R dependencies — see
R/00_setup.R. - Internal planning documents (charter, team roster, status, timeline,
governance, protocol, glossary) live in
docs/andadmin/and are not tracked by git. Contact the project PI for access.
- Funders — see https://canal-ai.org/#fund. Target pilot budget ≈ €450k over 24 months. SECOM has committed the first tranche; additional funders are warmly welcome.
- Partner hospitals — see https://canal-ai.org/#join. Requirements: a structured data warehouse, willingness to harmonise to OMOP-CDM, local IRB support, willingness to participate in federated learning, and a workstation capable of running a FedBioMed node.
- Code (everything under
R/andpython/): MIT — seeLICENSE. - Documentation and site content: CC BY 4.0.
- Portraits of individuals displayed on the site are shown with the subject's permission; photograph rights are reserved by the subject and the photographer.
Enquiries are best addressed through the single "Email the team" button on the public site (https://canal-ai.org/#contact), which reaches the three principal investigators in one message. For institutional correspondence, write to Nagoya University's Graduate School of Medicine or to Caen University Hospital's Department of Pharmacology.