Related Experiment Video
Updated: Apr 4, 2026
07:50
A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts
Published on: September 20, 2018
16.6K
Automated Extraction of Cancer Registry Data from Pathology Reports: Comparing LLM-Based and Ontology-Driven NLP
Thomas McPhaul1, Kory Kreimeyer1, Alexander Baras1
1The Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University School of Medicine, Baltimore, MD, USA.
Medrxiv : the Preprint Server for Health Sciences
|April 3, 2026
Summary
Automated extraction platforms can standardize cancer data. A large language model (LLM)-based system, Brim Analytics, demonstrated high accuracy and efficiency in processing pathology reports for cancer registries.
Area of Science:
- Oncology
- Medical Informatics
- Natural Language Processing
Background:
- Cancer data standardization is crucial for registry variables.
- Manual data abstraction from pathology reports is resource-intensive.
- Automated systems are needed to improve efficiency and accuracy.
Purpose of the Study:
- To evaluate Brim Analytics, a LLM-based system, against DeepPhe, an ontology-driven system.
- To assess the accuracy and processing time of automated extraction platforms for cancer data standardization.
- To compare performance across pancreatic and breast cancer pathology reports.
Main Methods:
- Two automated extraction platforms, Brim Analytics and DeepPhe, were evaluated.
- 330 pancreatic adenocarcinoma and 34 breast cancer pathology reports were used.
- Accuracy for seven registry variables and per-report processing times were assessed.
Main Results:
- Brim Analytics achieved high accuracy (mean 96.7% pancreatic, 93.7% breast) across registry variables.
- DeepPhe showed comparable performance for N stage but deficits in T stage.
- Brim Analytics demonstrated faster processing times (0.9s pancreatic, 4.6s breast) compared to DeepPhe (1.1s pancreatic, 3.5s breast).
Conclusions:
- LLM-based extraction, exemplified by Brim Analytics, achieves high accuracy in cancer data standardization.
- Automated platforms can significantly support and enhance cancer data workflows.
- Brim Analytics shows promise for accurate and efficient processing of pathology reports across different cancer types.
