How Natural Language Processing Can Aid With Pulmonary Oncology Tumor Node Metastasis Staging From Free-Text...

Area of Science:

• Medical Informatics
• Computational Linguistics
• Radiology

Background:

• Radiology reports often contain unstructured free text, limiting data mining and clinical decision support.
• Natural language processing (NLP) offers a method to structure this text for improved data utilization.
• Accurate oncological staging, crucial for patient care, can benefit from structured report data.

Purpose of the Study:

• To implement and validate an N-stage classifier for pulmonary oncology.
• To integrate the N-stage classifier with an existing T-stage classifier, creating a combined TN-stage classifier.
• To utilize free-text chest computed tomography reports for TNM classification.

Main Methods:

• Information extraction using SpaCy, PyContextNLP, and regular expressions.
• Development of additional rules for accurate N-stage extraction.
• Validation of a combined TN-stage classifier on radiological reports.

Main Results:

• The TN-stage classifier achieved high accuracy scores: 0.84 for the training set (N=95) and 0.85 for the validation set (N=97).
• Performance is comparable to the established T-stage classifier, which reported scores between 0.87 and 0.92.
• Demonstrated successful extraction of both T- and N-stages from free-text reports.

Conclusions:

• NLP is a viable tool for classifying pulmonary oncology using free-text radiological reports.
• The developed TN-stage classifier accurately extracts staging information according to the TNM system.
• NLP enhances the value of radiology reports for clinical decision support in oncology.