Glossary

Terms used throughout this documentation and in the CVDLink interfaces. Where a term has a particular meaning in this project that differs from its ordinary use, that meaning is given here.

Terms that are easily confused

Three words are used in more than one sense across the system. The intended sense is normally clear from the surrounding text, but the distinctions are worth stating.

Node means a participant in a network, and CVDLink has two separate networks. In the federation pages a node is a P2P node exchanging dataset announcements between institutions. In the Metadata application a node is a mesh node reporting its status over the message broker. A site typically runs both.

Site, node and peer. A site is an institution taking part, with its own installation and its own data. A node is the software that represents it on one of the networks. A peer is any site other than the one being described.

Resource, dataset and workspace are distinct records in the metadata schema and are not interchangeable. Their relationship is given under their individual entries below.

A

Announcement. A message published by a P2P node telling other sites that a dataset is available, carrying a description rather than the data itself. Receiving sites store the description and acknowledge it.

Application host. The machine running the Metadata application, the ETL service, the P2P nodes and Keycloak. Used throughout this documentation in place of an address, because addresses change between environments. The current address is in Hosts and ports.

Audit log. A record of user actions in the Metadata application, kept for accountability. Enabled by AUDIT_ENABLED.

C

Central deployment. The profile in which components connect to shared infrastructure: a common database, a shared message broker, and Keycloak for authentication. Used for the project's own environments. Compare standalone deployment.

Cohort. The set of patients selected for a study, defined by stated criteria so that every site selects on the same basis.

D

Data host. The machine running the shared PostgreSQL database and the MQTT message broker. As with the application host, referred to by role rather than address.

Dataset. A named collection of files within a workspace, and the unit that is described by metadata and announced to other sites. A workspace contains datasets; a dataset contains files.

DHT, distributed hash table. The mechanism P2P nodes use to find which site holds a given dataset, without a central directory. Operates over UDP on port 8468.

DICOM. Digital Imaging and Communications in Medicine, the standard format for medical imaging and its accompanying metadata.

Dublin Core. A small set of general-purpose metadata fields, such as title, creator and date. CVDLink exposes metadata in this form alongside the clinical formats, so that records can be read by systems that do not understand DICOM or FHIR.

E

EDF. European Data Format, used for recordings of physiological signals.

Embedding. A numerical representation of a piece of text that captures its meaning in a form that can be compared mathematically. The assistant features store embeddings so that material relevant to a question can be found by similarity rather than by exact match. See pgvector.

ETL. Extract, transform, load. The component that reads clinical and imaging files, extracts structured metadata, and writes it to the database.

EUCAIM. The European Federation for Cancer Images, whose metadata model CVDLink can convert records into, so that they can be shared with that federation.

F

Federated study. An analysis carried out across several sites at once, where each site computes over its own data and only results are exchanged. The patient data itself does not leave the institution that holds it.

FeatureCloud. The platform CVDLink uses to run analyses across sites. See FeatureCloud.

FHIR. Fast Healthcare Interoperability Resources, a standard for exchanging healthcare records. Often supplied as NDJSON, one record per line.

G

Gossip. The protocol by which P2P nodes pass announcements and acknowledgements to the peers they know, each node relaying to others so that a message reaches the whole network without every node contacting every other. Operates over TCP, published on port 4447.

H

HDF5. A file format for large scientific datasets, used in CVDLink for signal recordings.

HL7. Health Level Seven, a standard for messages exchanged between hospital systems. Distinct from FHIR, which is a newer standard from the same organisation.

K

Keycloak. The identity provider used in central deployments. It authenticates users and issues the tokens that carry their roles.

L

LEAF. A node role in the mesh network, meaning a node that participates without coordinating others. Set by MESH_NODE_ROLE.

M

Mesh network. The status network in the Metadata application, over which nodes publish heartbeats and messages via the MQTT broker. Distinct from the P2P federation, which exchanges dataset metadata.

Metadata. Information describing data rather than the data itself: what a dataset contains, who produced it, when, in what format, and under what conditions it may be used. CVDLink shares metadata between sites while the underlying data stays local.

Model app. An analysis packaged so that it can be distributed to participating sites and executed there. See Model apps.

MQTT. A lightweight publish and subscribe messaging protocol, used for mesh node status. The broker listens on port 1883.

N

Node identifier, node seed. The value identifying a site within a network, set by MESH_NODE_SEED or NODE_ID. Must be unique across all participating sites; two sites sharing one cannot be told apart.

O

OIDC, OpenID Connect. The protocol by which the Metadata application authenticates users against Keycloak.

OMOP CDM. The Observational Medical Outcomes Partnership Common Data Model, a standard structure for health records. Putting data into this form lets different institutions run the same analysis against their own records.

P

P2P. The CVDLink federation service, which announces datasets to other sites and maintains the shared catalogue that makes federated search possible. See P2P installation.

pgvector. A PostgreSQL extension adding a column type for embeddings and an index for searching them by similarity. Required by the assistant features.

PKCE. Proof Key for Code Exchange, an addition to the OIDC sign-in flow that allows an application with no stored secret to authenticate safely.

PTB-XL. A published collection of electrocardiogram recordings, supported by the ETL as a data source.

R

RAG, retrieval augmented generation. The technique behind the assistant features: material relevant to a question is retrieved from the database and supplied to a language model as context, so that answers are grounded in the project's own documents rather than the model's general knowledge.

Realm. A Keycloak term for a self-contained set of users, roles and clients. CVDLink uses the realm named cvdlink.

Resource. A record in the metadata schema describing one item of data, with its format-specific details held in related tables. The term appears in the metadata APIs, where a resource identifier addresses a single described item.

Role. A permission granted to a user and carried in their Keycloak token. CVDLink uses data-manager, data-steward and platform-administrator. In standalone deployments roles are not applied and every authenticated user has full access.

S

Standalone deployment. The profile in which every component runs on one machine with no external dependencies, using local accounts rather than Keycloak. Used by partner institutions and for demonstrations. Compare central deployment.

Stratification. Selecting and balancing a site's data so that the training dataset it contributes is representative and comparable with those of other sites. See Stratification.

V

VCF. Variant Call Format, a text format for genomic variants.

W

WFDB. Waveform Database, a format for physiological signal recordings, typically a data file paired with a header describing it.

Workspace. The top level of the data hierarchy in the Metadata application: a named area belonging to a user or group, containing datasets.