radiology-ner-model

Radiology Report NER Model

Named Entity Recognition model for extracting structured information from chest X-ray radiology reports.

Model Details

• Architecture: spaCy 3.8 with HashEmbedCNN
• Training Data: Indiana University Chest X-Ray Reports (2,674 reports)
• Entity Types: ANATOMY, OBSERVATION
• F-Score: 99.94%
• Precision: 99.93%
• Recall: 99.95%

Usage

import spacy

# Load the model

nlp = spacy.load("MakPr016/radiology-ner-model")

# Process text
text = "The cardiac silhouette is within normal limits. No pleural effusion."
doc = nlp(text)

# Extract entities

for ent in doc.ents:
print(f"{ent.text} - {ent.label_}")

Entity Labels

• ANATOMY: Body parts and anatomical structures (e.g., lung, cardiac silhouette, diaphragm)
• OBSERVATION: Medical findings and observations (e.g., consolidation, effusion, normal)

Training Details

• Training Steps: 11,500
• Best Model: Epoch 62
• Dataset: Indiana University Chest X-Ray Reports
• Validation: 573 test reports

Limitations

• Optimized for chest X-ray reports
• May not perform well on other body parts (knee, spine, etc.)
• Trained on English language reports only