Use of Natural Language Processing to Extract and Classify Papillary Thyroid Cancer Features From Surgical Pathology Reports
Authors' conclusion
ThyroPath shows promise in automating the extraction and risk recurrence classification of thyroid pathology reports at large scale. It offers a solution to laborious manual reviews and advancing virtual registries. However, it requires further validation before implementation.
Study context
- Population
- Surgical pathology reports from adult PTC patients, 2010–2019pathology reports
- Participants
- 1,410 reports
- Comparison
- ThyroPath NLP extraction and risk classification vs consensus-based ground truth
- Structured reportsn = 120
- Unstructured reportsn = 50
- Outcome
- Extraction (strict F1-score) and recurrence-risk classification accuracy of ThyroPath
Methodological detailsrule-based NLP pipeline, training and test sets, Training on 225 reports (150 structured, 75 unstructured); testing on 170 reports (120 structured, 50 unstructured)
- Rule-based NLP pipeline
- Training and test sets
- Training on 225 reports (150 structured, 75 unstructured); testing on 170 reports (120 structured, 50 unstructured)
Performance
As reported in the abstract
Extraction of 18 PTC pathology features, overall strict F1-score
| Test or model | Value | Strict F1-score |
|---|---|---|
| Structured reports | 93% | 93% |
| Unstructured reports | 90% | 90% |
Value
Structured reports93%
Unstructured reports90%
Strict F1-score
Structured reports93%
Unstructured reports90%
- Overall accuracy of recurrence-risk classification from ThyroPath-extracted information
- 93%
- Accuracy, high risk
- 76.9%
- Accuracy, intermediate risk
- 86.8%
- Accuracy, low and very low risk
- 100%
Further findings
Recurrence-risk classification from human-extracted pathology information
- Accuracy across all risk categories
- 100%
Original publication
Use of Natural Language Processing to Extract and Classify Papillary Thyroid Cancer Features From Surgical Pathology Reports. Endocr Pract. Published 26 August 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →