Automated Extraction of Imaging and Pathology Data From Diverse Prostate Cancer Electronic Records

Area of Science:

• Urology
• Medical Informatics
• Oncology

Background:

• Prostate cancer (PCa) research requires accurate data extraction from clinical reports.
• Manual data extraction from pathology and MRI reports is time-consuming and prone to errors.
• Standardized data extraction methods are needed to facilitate large-scale PCa research.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm for extracting clinically relevant data elements from prostate biopsy and MRI reports.
• To automate the extraction of key metrics such as Gleason score, PI-RADS score, and prostate dimensions.

Main Methods:

• A rule-based NLP algorithm was developed using hand-annotated biopsy and MRI reports from the VA Cancer Registry System and Corporate Data Warehouse.
• The algorithm was trained to extract Gleason score, positive/total cores, PI-RADS score, PSA density, prostate volume, and dimensions.
• Algorithm performance was validated on a separate set of 250 biopsy and 250 MRI reports.

Main Results:

• The NLP algorithm demonstrated high performance with F1 scores exceeding 88% for all extracted data elements.
• Specific F1 scores included Gleason (96.9), PI-RADS (93.7), PSA density (99.5), prostate volume (95.7), and prostate dimensions (93.2).
• Errors were primarily attributed to complex or ambiguous language in the reports.

Conclusions:

• The developed NLP algorithm reliably extracts essential data for prostate cancer research from clinical reports.
• Automated data extraction using this algorithm can significantly support downstream research and clinical applications in PCa.
• This tool offers a scalable solution for accessing critical information from large volumes of medical records.