Natural Language Processing Methods Automate Molecular Marker Extraction From Glioma Pathology Reports.

Area of Science:

• Computational pathology
• Bioinformatics
• Natural Language Processing (NLP)

Background:

• Accurate molecular marker status (Isocitrate Dehydrogenase - IDH, Alpha-thalassemia/mental retardation syndrome X-linked - ATRX) is critical for glioma classification and treatment.
• Manual extraction of these markers from pathology reports presents a significant bottleneck for research.
• Evaluating the performance of NLP approaches with varying computational complexity is essential for optimizing research workflows.

Purpose of the Study:

• To compare the effectiveness of three Natural Language Processing (NLP) approaches—Regular Expressions (RegEx), Term Frequency-Inverse Document Frequency (TF-IDF), and Bidirectional Encoder Representations from Transformers (BERT)—for extracting IDH and ATRX molecular markers from glioma pathology reports.
• To determine if more computationally intensive NLP methods offer significant performance advantages over simpler methods in computational pathology research.

Main Methods:

• Analysis of pathology reports from 404 patients (Institution A) and 197 patients (Institution B) for external validation.
• Application of identical preprocessing steps, including text normalization and terminology standardization, to all evaluated NLP approaches.
• Performance evaluation using standard classification metrics (accuracy, AUC) and memory usage benchmarks on both internal and external datasets.

Main Results:

• Simpler NLP approaches, RegEx and TF-IDF, outperformed complex BERT-based models in accuracy and AUC for both IDH and ATRX marker extraction on external validation data.
• RegEx achieved near-perfect accuracy (99-100%) and TF-IDF maintained high accuracy (94.2-98.0%) for both markers.
• BERT-based approaches required substantially more memory (1825-1953 MB) compared to RegEx (0.82-5.52 MB) and TF-IDF (17.27-34.89 MB).

Conclusions:

• Simple NLP approaches, particularly RegEx, provide a highly accurate and computationally efficient solution for automating molecular marker extraction from pathology reports.
• The findings suggest that simpler NLP methods are sufficient for many computational pathology research tasks, enabling larger sample sizes and multi-institutional analyses.
• Future research should focus on validating these findings across larger datasets and integrating NLP tools for broader application in biomarker research.