The NLP Engine Behind
Radiology Intelligence

Aliri NLP is the entity-extraction and ontology-mapping pipeline at the heart of Aliri AI. It transforms free-text radiology reports into structured, ontology-coded clinical data — every entity, every report, automatically.

Looking for the AI analyst layer? See Aliri+.

Four Stages, One Seamless Flow

From raw text to structured intelligence — here's how every report is processed.

Stage 01

Report Ingestion

Connect to any RIS, PACS, or HL7 feed. Aliri normalizes report structure — handling headers, sections, impressions, and addenda — before NLP processing begins.

  • HL7 v2 / FHIR support
  • Real-time streaming or batch import
  • Section segmentation (Findings, Impression, History)
  • Multi-site, multi-modality support

Stage 02

NLP Entity Extraction

A purpose-built medical NLP pipeline identifies clinical entities from free-text radiology reports. The engine handles negation, uncertainty, laterality, and temporal context.

  • Named entity recognition (NER)
  • Negation and uncertainty detection
  • Relationship extraction (finding → anatomy)
  • Context-aware entity linking

Stage 03

Ontology Mapping

Each extracted entity is mapped to RadLex and SNOMED CT concepts using a combination of dictionary lookup, embedding similarity, and rule-based disambiguation.

  • RadLex (RSNA standard)
  • SNOMED CT clinical terminology
  • ICD-10 cross-reference (optional)
  • Confidence scoring per mapping

Stage 04

Structured Output

The result: a fully structured, queryable representation of every report. Each entity carries ontology codes, section context, negation status, and relationship links.

  • JSON / FHIR-compatible output
  • Searchable entity database
  • Cohort builder integration
  • Analytics-ready structured data

What Aliri Extracts

The NLP engine recognizes and classifies nine core entity types from every radiology report.

Anatomy4,200+

right upper lobe, liver, left kidney

Finding3,800+

nodule, mass, effusion, fracture

Disorder/Disease2,500+

pneumonia, cirrhosis, lymphoma

MeasurementDynamic

5mm, 2.3 x 1.8 cm

Modifier600+

stable, increased, new, unchanged

LateralityAuto-detected

right, left, bilateral

PolarityAuto-detected

positive, negative, no evidence of

CertaintyInferred

definite, probable, possible, suspected

RecommendationRule-based

follow-up CT in 6 months, MRI recommended

RadLex + SNOMED CT, Automatically

Every entity is assigned standardized codes from both RadLex and SNOMED CT, enabling cross-system interoperability and research-grade data quality.

High-confidence mapping

Confidence scores on every mapping. Low-confidence results are flagged for review.

Continuous vocabulary updates

Ontology dictionaries are updated with each RadLex and SNOMED release.

Cross-mapping support

Navigate between RadLex and SNOMED via built-in cross-reference tables.

ontology-output.json
{
  "report_id": "RPT-20260329-4182",
  "entities": [
    {
      "text": "hepatic steatosis",
      "type": "FINDING",
      "negated": false,
      "radlex": {
        "id": "RID4566",
        "term": "hepatic steatosis",
        "confidence": 0.97
      },
      "snomed": {
        "id": "197321007",
        "term": "Steatosis of liver",
        "confidence": 0.95
      }
    },
    {
      "text": "right kidney",
      "type": "ANATOMY",
      "radlex": {
        "id": "RID29662",
        "term": "right kidney"
      },
      "snomed": {
        "id": "9846003",
        "term": "Right kidney structure"
      }
    }
  ]
}

Search by Concept, Not by Keyword

Because radiologists describe the same finding in dozens of ways. Aliri searches by ontology concept — so you never miss a match.

Synonym Resolution

Search for "hepatic steatosis" and also find "fatty liver", "steatohepatitis", and "fatty infiltration of liver".

RadLex: RID4566 → 4 synonyms matched

Hierarchical Navigation

Search for "lung nodule" and optionally include all child concepts — ground-glass nodule, calcified nodule, subsolid nodule.

SNOMED hierarchy: 3 levels deep

Complex Queries

Combine anatomy + finding + temporal filters: "all pulmonary embolism findings in the last 90 days, excluding negated mentions."

Supports AND / OR / NOT logic

Aliri NLP structures the data.
Aliri+ lets you talk to it.

Once Aliri NLP has structured your reports, Aliri+ lets you talk to the result. Ask questions in plain English across clinical findings and operational BI data — no SQL, no dashboards, no waiting — and get cited, grounded answers in seconds.

Explore Aliri+
Ask Aliri+

“Show me CTs with a 5–10mm pulmonary nodule in the last 6 months — and the radiologists whose follow-up rate is below average.”

RadLexSNOMED+ BI

See the platform in action

Schedule a technical demo and see how Aliri processes your radiology reports.

Request a Demo