Development of Optical Character Recognition-Based Quality Control Process of Paper-Based Consent Forms

Area of Science:

• Biobanking and Data Management
• Artificial Intelligence in Healthcare
• Digital Transformation in Research

Background:

• Efficient consent management is crucial for ethical and legal biobanking.
• Traditional paper-based consent forms present challenges in legibility and scalability.
• Automating the reading and quality control of consent forms is needed.

Purpose of the Study:

• To automate the reading and quality control of paper-based consent forms.
• To develop a reliable optical character recognition (OCR) model for handwritten Korean consent documents.
• To enhance the digital transformation of traditional biobanking processes.

Main Methods:

• Optimized a proprietary optical character recognition (OCR) model for handwritten Korean characters.
• Generated and utilized synthetic standard and nonstandard consent documents for training.
• Evaluated the model's performance on 3,790 pages of consent forms in routine biobanking.

Main Results:

• The optimized OCR model achieved 88.94% accuracy on standard forms and 91.88% on nonstandard forms.
• Applied to routine biobanking, the model demonstrated 91.25% accuracy and an F1-score of 0.91.
• The model exhibited high performance and excellent generalization capabilities for consent data.

Conclusions:

• Developed a highly efficient and reliable AI-based OCR tool for paper-based consent management.
• This approach facilitates the digital transformation of traditional biobanking.
• The optimized OCR model enhances the processing of consent documents in biobanks.