deidentifier/CITATION.cff at main · KatherLab/deidentifier · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
cff-version: 1.2.0
title: "Medical Document Anonymizer (deidentifier)"
message: >-
  If you use this software in your research, please cite it using the metadata
  from this file. This is a follow-up implementation and not the code evaluated
  in the LLM-Anonymizer paper (Wiest et al., NEJM AI 2025, listed under
  "references"); please cite that paper as well when referring to the
  underlying approach.
type: software

authors:
  - family-names: "Wolf"
    given-names: "Fabian"
    affiliation: >-
      Else Kroener Fresenius Center for Digital Health, Faculty of Medicine
      and University Hospital Carl Gustav Carus, TUD Dresden University of
      Technology, Dresden, Germany
    email: "fabian.wolf2@tu-dresden.de"

  - name: "KatherLab"
    website: "https://github.com/KatherLab"

repository-code: "https://github.com/KatherLab/deidentifier"
url: "https://github.com/KatherLab/deidentifier"

abstract: >-
  A locally deployable web application that anonymizes German clinical
  documents. Rule-based recognizers and a prompted large language model behind
  any OpenAI-compatible endpoint propose character spans; deterministic code
  applies every replacement on the immutable source text, and an independent
  leakage-validation pass re-scans the output. Documents are processed in
  memory only and nothing is persisted. The repository also contains a
  standalone evaluation harness that scores the pipeline against annotated
  ground truth, reporting document-level leakage alongside character- and
  span-level metrics.

keywords:
  - "de-identification"
  - "anonymization"
  - "large language models"
  - "named entity recognition"
  - "clinical text"
  - "German clinical documents"
  - "medical natural language processing"
  - "privacy-preserving artificial intelligence"

license: "AGPL-3.0-or-later"
version: "0.4.0"
date-released: "2026-08-24"

references:
  - type: article
    title: >-
      Deidentifying Medical Documents with Local, Privacy-Preserving Large
      Language Models: The LLM-Anonymizer
    authors:
      - family-names: "Wiest"
        given-names: "Isabella C."
      - family-names: "Leßmann"
        given-names: "Marie-Elisabeth"
      - family-names: "Wolf"
        given-names: "Fabian"
      - family-names: "Ferber"
        given-names: "Dyke"
      - family-names: "Van Treeck"
        given-names: "Marko"
      - family-names: "Zhu"
        given-names: "Jiefu"
      - family-names: "Ebert"
        given-names: "Matthias P."
      - family-names: "Westphalen"
        given-names: "Christoph Benedikt"
      - family-names: "Wermke"
        given-names: "Martin"
      - family-names: "Kather"
        given-names: "Jakob Nikolas"
    journal: "NEJM AI"
    year: 2025
    volume: 2
    issue: 4
    number: "AIdbp2400537"
    doi: "10.1056/AIdbp2400537"
    url: "https://ai.nejm.org/doi/full/10.1056/AIdbp2400537"
    notes: >-
      The predecessor work this software builds on. It established the
      approach — local, privacy-preserving LLMs proposing identifiers in
      clinical documents — and reported the accuracy figures for that
      implementation. This repository is an independent reimplementation with a
      different architecture, and its behavior and accuracy are not those
      measured in the paper.