Medical AI lab · Note de-identification
De-identifying a clinical note: interactive demo
De-identifying a clinical note means finding and removing or replacing the details that could identify the patient: names, dates, ID numbers, contact details, addresses and rare personal details. This demo runs a simple rule-based detector over a fully fictional discharge note in English, Hebrew or mixed Hebrew-English and scores it against a hand-made annotation, so you can see which identifiers are caught, which are missed and what is removed by mistake.
These demos are illustrative. Every signal, image, patient and note is synthetic and fictional, generated and scored in your browser, and every number shown is computed from that synthetic data. They show general principles, not the performance of any real model, product or project, and none of them is a medical device.
Nothing is uploaded. No patient data is used.
Clinical data de-identification / free text
Names are easy. The rest re-identifies.
What this shows: simple rules catch titled names and ID numbers early; dates, places and rare details such as an unusual job or a unique event need more, and turning the rules up also removes doses, drug names and ordinary words.
Run the detector
Rules switched on
Fictional discharge note
13 annotated identifiersGreen: identifier caughtRed: identifier missedAmber: over-redacted
Synthetic data · fictional note · illustrative · not a medical device
Free-text notes, letters and reports before they reach analysts, vendors or language models. Measure missed identifiers and over-redaction on your own notes and languages; de-identification lowers disclosure risk but is not a guarantee of anonymity.
What this experiment models
One fictional note in three versions, with a ground-truth annotation of 13 identifiers: names, dates, an ID number and a phone number (all zeros), an address and towns, and two rare details. The detector is a list of regular expressions and word lists, switched on in five levels. An identifier counts as caught when at least half of its characters are redacted; a redaction that touches no identifier counts as over-redaction. Real systems combine rules with trained models and human review.
The principle
How it works
The note contains 13 annotated identifiers: four names, two dates, an ID number and a phone number (both all zeros), a street address, two towns, an occupation and a unique event. The detector is a stack of regular expressions and word lists in five levels: titled names and ID numbers; then phone numbers, numeric dates and a short list of first names; then month-name dates, addresses, towns and capitalised words; then any number and a longer name list; and finally occupations and every capitalised word.
Names and ID numbers are caught early. Dates written in words, addresses and towns need more rules. Hebrew has no capital letters, so a name without a title is missed until it is on a list, and the list then also removes ordinary words that double as names, such as the Hebrew word for age. The occupation is caught only at the strictest level, the unique event never is, and doses, lab values and drug names are over-redacted along the way.
Lists of identifier types, such as the HIPAA Safe Harbor list, are a starting point. Measure missed identifiers and over-redaction on your own notes and languages, and keep in mind that de-identification lowers disclosure risk but does not guarantee anonymity. See clinical data de-identification.
Medical AI lab
More demos
All six demos on one page: the medical AI labClinical data de-identification →