Hybrid ReGex and Natural Language Inference Model as a Zero-Shot Classifier for Extracting Data From Medical Reports.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Clinical data extraction from unstructured reports is challenging.
• Existing methods have limitations in accuracy and efficiency.
• Prostate cancer research requires precise symptom data.

Purpose of the Study:

• To develop and evaluate a hybrid AI and regular expression (ReGex) pipeline for medical data extraction.
• To improve the accuracy and efficiency of extracting key clinical information from prostate cancer reports.
• To address limitations of individual data extraction techniques.

Main Methods:

• A hybrid pipeline combining ReGex for initial extraction and Natural Language Inference for classification was developed.
• The pipeline was applied to 1,000 prostate cancer patient reports.
• Performance was assessed using precision, recall, accuracy, F1-score, and Cohen's kappa.

Main Results:

• The pipeline achieved high precision (0.778-0.954) and recall (0.920-1.00).
• F1-scores demonstrated balanced accuracy across identified symptoms.
• Cohen's kappa values (0.871-0.951) indicated strong agreement with physician-labeled data.

Conclusions:

• The developed pipeline is efficient, fast, and computationally lightweight.
• It demonstrates high accuracy in extracting medical data from clinical reports.
• This tool offers a practical solution for clinical research and healthcare applications.