Pulmonary vascular disease & thromboembolism
Leveraging Hybrid Natural Language Processing Techniques for Large-Scale Pulmonary Embolism Identification
Authors' conclusion
The hybrid NLP approach required less training data than pure ML models and demonstrated high performance across diverse health care settings. A hybrid NLP pipeline can efficiently and accurately identify PE cases from radiology reports
Study context
- Population
- Radiology reports, Brigham and Women's Hospital (development) and Mass General Brigham health care system (validation)radiology reports (validation)
- Participants
- 49,611 reports
- Comparison
- Hybrid natural language processing (NLP) pipeline vs machine learning (ML) model alone for pulmonary embolism (PE) identification
- Outcome
- Accuracy, sensitivity, specificity, PPV and NPV for identification of PE cases
Methodological detailsdevelopment and validation
- Development and validation
Performance
As reported in the abstract
PE identification from radiology reports
| Strategy | AUC (BWH testing data set) | Sensitivity (Mass General Brigham (MGB) data set) | Specificity (MGB data set) | PPV (MGB data set) | NPV (MGB data set) | Accuracy (BWH testing data set) | Accuracy (MGB data set) |
|---|---|---|---|---|---|---|---|
| ML model | 0.90 | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | 91% | — not stated in the abstract |
| Hybrid NLP | — not stated in the abstract | 96.4% | 93.2% | 93.0% | 96.5% | — not stated in the abstract | 94.8% |
AUC (BWH testing data set)
ML model0.90
Hybrid NLP— not stated in the abstract
Sensitivity (Mass General Brigham (MGB) data set)
ML model— not stated in the abstract
Hybrid NLP96.4%
Specificity (MGB data set)
ML model— not stated in the abstract
Hybrid NLP93.2%
PPV (MGB data set)
ML model— not stated in the abstract
Hybrid NLP93.0%
NPV (MGB data set)
ML model— not stated in the abstract
Hybrid NLP96.5%
Accuracy (BWH testing data set)
ML model91%
Hybrid NLP— not stated in the abstract
Accuracy (MGB data set)
ML model— not stated in the abstract
Hybrid NLP94.8%
— not stated in the abstract
- ML model accuracy on the larger MGB data set
- 85%
Original publication
Leveraging Hybrid Natural Language Processing Techniques for Large-Scale Pulmonary Embolism Identification. JACC Adv. Published 21 May 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →