Extracting laboratory test information from paper-based reports.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Data Digitization

Background:

• Many healthcare reports remain in paper format, hindering data accessibility.
• Digitizing paper lab reports is challenging due to varied layouts and data types.
• A scalable, lightweight method is needed to convert paper reports to structured data.

Purpose of the Study:

• To develop an end-to-end NLP pipeline for extracting information from paper-based laboratory reports.
• To convert diverse, non-standardized paper reports into a structured, digital format.
• To create a system with low computational resource requirements for clinical adoption.

Main Methods:

• An NLP pipeline combining Optical Character Recognition (OCR) and Information Extraction (IE) was developed.
• The OCR module identifies and locates text in scanned reports.
• The IE module extracts entities like test item name, results, units, and reference ranges using sub-modules including NER with CRF.

Main Results:

• The OCR module achieved an average accuracy of 0.93 for text detection and recognition.
• The IE module demonstrated an overall F1 score of 0.86 in extracting key laboratory test information.
• The pipeline processed each report in an average of 0.78 seconds on a single CPU.

Conclusions:

• A practical, lightweight pipeline was successfully developed for digitizing paper laboratory reports.
• The system effectively handles diverse report layouts and data types.
• High performance on a real-world dataset validates the pipeline's feasibility for clinical environments.