Accurate pattern-based extraction of complex Gleason score expressions from pathology reports

Area of Science:

• Urology
• Computational Pathology
• Bioinformatics

Background:

• Prostate cancer grading relies on the Gleason score, a critical prognostic factor.
• Automated extraction of Gleason scores from pathology reports is challenging due to complex score expressions.
• Existing methods often fail to capture nuanced Gleason score patterns.

Purpose of the Study:

• To develop and validate a novel regular expression-based program for extracting both simple and complex Gleason score expressions from prostate cancer pathology reports.
• To improve the accuracy and efficiency of Gleason score data retrieval from clinical texts.

Main Methods:

• Developed a regular expression-based program to identify Gleason scores in pathology reports.
• Evaluated the program on manually processed Finnish prostate cancer cases from 2016-2017.
• Measured performance using recall, precision, and F1-score, estimating complex expression proportion.

Main Results:

• The program achieved 92% recall and 98% precision for all Gleason score expressions.
• For addition-only expressions (e.g., 3+4), recall was 65% with 100% precision.
• An estimated 35% of Gleason score expressions were complex.

Conclusions:

• Automated extraction of complex Gleason score expressions from pathology reports is feasible with high accuracy.
• The developed program demonstrates high recall and precision, outperforming previous methods.
• Recommends adapting this validated program for automated Gleason score extraction in clinical practice.