.jpg){kind=link}
- Published
- Updated
Optical character recognition makes scanned documents searchable but can misread letters, symbols and numbers. The image may look correct while its text layer contains a different date, amount or statement that later affects a review.
Preserve the meaning of data
Define the role of OCR output. Text used only to locate a document differs from text used to populate another system. Establish source-image verification for critical fields. Failure to find a term does not demonstrate that the information is absent from the document. Include poor scans and complex tables when selecting representative checks.
Worked example
For example, OCR misreads a date in a training record, and an automated list shows overdue training. An authorised reviewer checks the image and preserves the source and rationale for the corrected text. The team investigates whether similar document types are affected. Search text is not silently promoted to an authoritative observation record.
Review checks
- Compare critical numbers, dates and negative statements with the image.
- Include tables, handwriting and low-contrast examples.
- Avoid workflows that equate a failed search with an absent document.
Acceptance and boundaries
Acceptance means OCR text is suitable for its defined purpose and the source remains accessible when meaning is uncertain. Manual verification of every word may not be necessary for every use; choose proportionate checks. An OCR confidence score alone does not establish clinical correctness. Manage text corrections without concealing the content of the original scanned image.
Frequently asked questions
- Is OCR text the original record?
- Its role depends on the record arrangements; automatically recognised text does not become the original observation by default.
- Is a high confidence score enough?
- Errors affecting important meaning can remain, so verification should match the intended use.