# Clinical technology and semantic AI | ESPEIR

> Discover ESPEIR’s technology foundations: clinical NLP, contextualization, terminologies, multilingual support, traceability and CPU-based processing.

Web version: https://espeir.com/en/technology
Languages: [fr](https://espeir.com/md/fr/technology) · [en](https://espeir.com/md/en/technology) · [de](https://espeir.com/md/de/technology)

## Technology and science · Clinical intelligence deserves evidence.

Specialized models to understand and structure medical information and link it to its sources.

Clinical understanding technologies built for health systems: precise concepts, preserved context, semantic interoperability and controlled compute costs.

## Capabilities · Understand clinical meaning, not just recognize words.

Confirmed, suspected, negated or family history: the same term carries different information.

- **Clinical extraction.** Identify the medical concepts in supported documents.
- **Contextualization.** Distinguish affirmed, negated and hypothetical mentions, and other situations covered by the models.
- **Terminologies.** Map clinical wording to supported reference terminologies, including SNOMED CT.
- **Multilingual.** Link medical expressions across validated languages and terminologies.
- **Social determinants.** Documented social information, structured when the models cover it.
- **Traceability.** Link results to the documents and passages that contributed to their extraction.

## Clinical information keeps its context.

1. **Identify.** Detect medical mentions and concepts in the analyzed content.
2. **Contextualize.** Understand their meaning and map them to shared reference terminologies.
3. **Deliver.** Produce structured results, traceable to their source documents.

## Architecture · Medical memory, built on your existing IT system.

The HL7 feeds already in place populate a FHIR server. CLARYS Medical Memory unlocks their value, for care and beyond.

Secure IT system · on-premises or cloud

Contextual call to CLARYS Medical Memory

### Deliver · Care and data

Existing applications

- Admissions
- Radiology
- Laboratory
- X
- Y
- EHR

- Identity
- Movement
- Activity

Documents

Application data: logical target

Application data

### Standardize · Secure

FHIR server

Sovereign medical memory

Standardized data

- Secure external access
- Mobile access
- Community care link
- Single point of access

Only ADT and document feeds, which everyone has mastered for years.

With your usual interoperability partners or your own teams, setup is simple: your existing feeds are all it takes.

## Performance · Measured performance.

Scale, speed, coverage: CLARYS Intelligence is benchmarked on realistic volumes and against an open-source reference.

### At the scale of a healthcare facility

10,000 patients, 100 documents per patient on average

- **response time.** 120 ms
- **for full ingestion*.** < 2 min
- **60 GB of documents reduced to 600 MB after ingestion.** ÷ 100

### Compared with MedCAT, the open-source reference

228 clinical reports, same SNOMED CT and ICD-10 scope

- **more distinct concepts identified (2,118 vs 1,590).** +33%
- **more ICD-10 concepts recognized (319 vs 39).** ×8
- **of concepts found by MedCAT also detected.** 94%
- **faster: 8.1 ms vs 12 ms per document.** 1.5×

Internal ESPEIR measurements. Deterministic, reproducible extraction, with no generative model. Results specific to the stated corpora and scopes. * Ingestion excluding OCR, on documents already available as text.

## Standards and interoperability · Structured data that works with existing systems.

An exchange format is not enough: clinical information must be identified, contextualized and properly represented.

- **FHIR R4.** Standardized resources, depending on available profiles.
- **International Patient Summary.** The patient’s key information, based on available data.
- **Clinical terminologies.** Concepts mapped to the supported coding systems.
- **Traceability.** The origin of every result, always traceable.

## Governance and national frameworks · Prepare for data reuse without blurring purposes.

Providing care does not automatically authorize reuse: research and management follow their own governance and access rules.

The EHDS in Europe and comparable programs elsewhere are gradually strengthening these frameworks. ESPEIR builds capabilities that support these changes, but cannot by itself guarantee regulatory compliance.

Comparable frameworks, country by country

---

## Site pages

- [Home](https://espeir.com/md/en)
- [CLARYS Medical Memory](https://espeir.com/md/en/clarys/medical-memory)
- [CLARYS Intelligence](https://espeir.com/md/en/clarys/intelligence)
- [IT convergence and modernization](https://espeir.com/md/en/solutions/modernisation)
- [Archives and medical heritage](https://espeir.com/md/en/solutions/archives)
- [Clinical data warehouses](https://espeir.com/md/en/solutions/clinical-data-warehouse)
- [Health and social care continuity](https://espeir.com/md/en/solutions/care-continuity)
- [Software vendors and integrators](https://espeir.com/md/en/solutions/software-vendors)
- [Trajectories and populations](https://espeir.com/md/en/solutions/trajectories-populations)
- [Technology and science](https://espeir.com/md/en/technology)
- [Partners](https://espeir.com/md/en/partners)
- [ESPEIR](https://espeir.com/md/en/espeir)
- [Contact](https://espeir.com/md/en/contact)
- [Legal notice](https://espeir.com/md/en/legals)
- [Privacy policy](https://espeir.com/md/en/confidentiality)

## Contact

ESPEIR, 66 avenue des Champs-Élysées, 75008 Paris, France
contact@espeir.com