菜单
Inspect DICOM (X-ray) header labels — verify presence/missingness of 21 clinically important header tags and compute a curation score (0-100). Use for LYDUS data quality assessment of DICOM imaging metadata before large-scale analysis.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/28sungmin/m4-add-skills --skill dcm-checker命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Detect and label de-identification–target personal information (names, IDs, addresses, dates/times/periods, other hospitals) in unstructured Korean clinical notes using regular-expression rules, wrapping each match as <LABEL>string</LABEL>. Sensitive-diagnosis rules (KCD/name) ship but are off by default. Use for LYDUS pseudonymization/de-identification of free-text medical records.
Check the label-curation quality of GE MUSE ECG exports — verify presence/missingness of 35 ECG header attributes (3 classes) and compute a curation score (0-100), optionally rendering the 12-lead waveform. Use for LYDUS data quality assessment of ECG-derived labels before large-scale analysis.
Detect demographic bias in LYDUS data quality metrics by computing all 16 QUIQ quality metrics (completeness, range/date/format/sequence validity, preciseness, fidelity, class/instance diversity, sentence/vocabulary diversity, logical accuracy, cross-sectional/time-series consistency, classification, note accuracy/fidelity) per Sex/Race/Age group and computing GDI (Group Disparity Index). Uses Claude CLI (no API key required).
正在显示 SKILL.md
基于 SOC 职业分类
| name | dcm-checker |
| description | Inspect DICOM (X-ray) header labels — verify presence/missingness of 21 clinically important header tags and compute a curation score (0-100). Use for LYDUS data quality assessment of DICOM imaging metadata before large-scale analysis. |
| tier | community |
| category | lydus |
| parameters | {"dcm_path":{"description":"Path to a single DICOM file (.dcm).","type":"string"},"threshold":{"description":"Curation score below which the DICOM file is flagged as risky for research.","default":90,"type":"integer"}} |
Inspects the header-label quality of DICOM (X-ray) imaging data: parses the DICOM header, checks which of 21 clinically important tags are present vs missing, and assigns a curation score (0-100). A low score means many missing labels that may cause trouble in large-scale analysis.
21 of ~250 DICOM headers, selected for clinical/technical X-ray use and confirmed by clinicians/radiologists:
| Group,Element | Name | Group,Element | Name | |
|---|---|---|---|---|
| 0008,002A | Acquisition DateTime | 0018,1147 | Field of View Shape | |
| 0008,0060 | Modality | 0018,1149 | Field of View Dimensions | |
| 0008,0070 | Manufacturer | 0018,1164 | Imager Pixel Spacing | |
| 0008,1030 | Study Description | 0018,5101 | View Position | |
| 0008,103E | Series Description | 0020,0060 | Laterality | |
| 0010,0020 | Patient ID | 0028,0004 | Photometric Interpretation | |
| 0010,0040 | Patient's Sex | 0028,0010 | Rows | |
| 0010,1010 | Patient's Age | 0028,0011 | Columns | |
| 0018,0015 | Body Part Examined | 0028,0030 | Pixel Spacing | |
| 0018,1000 | Device Serial Number | 0028,0106 | Smallest Image Pixel Value | |
| 0028,0107 | Largest Image Pixel Value |
curation_score = int( present / 21 * 100 ) # 0-100
A tag is present if it exists in the header and its value is non-empty; otherwise missing.
The abstract describes auto-filling missing labels that can be derived from other fields (e.g., Patient's Age from Birth Date + Study Date). This is not implemented in the original code (v1.0.6) — the shipped tool only checks presence/missingness — so this skill does not implement it either, staying faithful to the source. It is noted here only so the gap between the abstract and the code is explicit.
Not applicable. Operates on raw DICOM files, not a QUIQ table.
scripts/dcm_checker.py is a headless reimplementation of the original PyQt5 GUI, preserving the same 21-tag set and score formula, using pydicom for portability. (The original GUI is not vendored — see PROVENANCE.yaml for the source repo.)
from scripts.dcm_checker import get_curation_score, format_report
result = get_curation_score("/path/to/image.dcm")
print(format_report(result, threshold=90))
# result["curation_score"], result["missing"], result["present"]
Or as a CLI:
python scripts/dcm_checker.py --dcm_path /path/to/image.dcm --threshold 90
Faithful to the original — no auto-fill. The abstract describes auto-filling derivable labels, but the shipped program.py (v1.0.6) only checks presence/missingness. This skill keeps that behavior; dcm_checker.py is a headless refactor of the same logic. This slimmed skill ships only dcm_checker.py (the original PyQt5 GUI is not vendored).
Library choice. The original GUI reads DICOM via SimpleITK (metadata keys like 0010|1010). The headless version uses pydicom (tags like (0x0010,0x1010)) — more standard and testable. The 21-tag set and int(present/21*100) formula are identical, so scores match.
Presence only for range/type. The reference says "data type and data range will be considered," but v1.0.6 checks presence/missingness only. Range/type validation is not implemented.
Single file per run. v1.0.6 scores one DICOM file at a time. Multi-file batch analysis and integrated visualization are planned but not yet implemented.
Threshold. The reference flags files scoring below 90 (the default here).
Pixel data / image rendering. load_dicom reads headers only (stop_before_pixels=True) for speed. The original GUI also renders the image (SimpleITK + Qt); that display step is intentionally omitted from the headless checker.