Application of a Natural Language Processing Framework for Data Extraction From Pathology Reports Across Multiple...

Area of Science:

• Computational biology
• Medical informatics
• Natural Language Processing

Background:

• Pathology reports contain valuable clinical and pathological data but are challenging to extract for research.
• An efficient natural language processing (NLP) system was developed to automate data extraction from semi-structured pathology reports.
• The system facilitates streamlined storage, retrieval, and analysis of clinical data in a centralized database.

Purpose of the Study:

• To develop and evaluate an automated system for extracting clinical data from pathology reports.
• To compare the performance of different deep learning models for this NLP task.
• To identify the optimal model for accurate and efficient data extraction.

Main Methods:

• Comparative analysis of deep learning architectures including LSTM, CNN, and transformer-based models (BERT, BioBERT, ClinicalBERT).
• Evaluation of model performance based on accuracy and efficiency in extracting variables from pathology reports.
• Selection of ClinicalBERT as the base model due to its proficiency in medical terminology and context.

Main Results:

• ClinicalBERT demonstrated superior performance in classifying variables across multiple cancer types.
• High F1 scores (≥0.99) were achieved for most variables in liver cancer.
• Variable performance was observed for other cancers, with some achieving perfect scores (F1=1.0) and others requiring further optimization (e.g., distant metastasis in stomach cancer).

Conclusions:

• NLP systems, specifically ClinicalBERT, are effective in automating the extraction of clinical data from pathology reports.
• This automated approach simplifies data processing and enhances the accuracy of extracted information.
• The developed system shows promise for improving cancer research through efficient data utilization.