Skip to content

Convention: DICOM Attribute lists

Teri M Sippel Schmidt edited this page Aug 21, 2026 · 3 revisions

Why an essential DICOM metadata subset is needed

Reproducibility of medical imaging AI depends on acquisition parameters, which are recorded in DICOM metadata. The Medical Imaging CDM (MI-CDM), an extension of the OMOP CDM, brings these parameters into a standardized, multi-institutional framework. Loading the full raw tag set is impractical, however, because hundreds of attributes attach to every image and many are redundant or modality-specific, imposing a storage and curation burden that is currently a principal barrier to MI-CDM adoption. A necessary and sufficient subset of acquisition metadata is therefore required.

Have a question? Post questions or issues to forums.ohdsi.org with the tag #imaging and we will get right back to you!

How the subset was selected

Expert review is the ideal way to decide which attributes matter, but the number of tags is too large to screen manually. We therefore derived a data-driven candidate list to seed and focus that review. Acquisition parameters were mined from open-access imaging AI literature published between July 2024 and June 2025 on PubMed Central (1,010 CT, 1,089 MR, 56 MG, and 727 X-ray publications), extracted with a few-shot LLM, and mapped to standard DICOM attributes through the pipeline shown below. Tags reported by more than 5% of papers per modality were retained as candidates, and this frequency-based selection was validated against model-performance variation through AI model replication, where the retained tags proved markedly enriched for performance-relevant attributes. The resulting modality-specific lists are data-driven candidates pending expert review.

figure_schema_literature_260511

Figure. From PubMed Central search to DICOM-matched protocol elements. After LLM-based abstract screening and few-shot protocol extraction, each element passes through a four-stage transformation (rule-based classifier, top-5 hybrid retriever, LLM adjudicator, and LLM fallback) against a DICOM standard knowledge base built from PS3.6 (tag registry, 5,190 elements), PS3.3 (modality relevance, 41,336 combinations), and vector embeddings (44,491 documents).

Each tag is listed with its custom OMOP concept ID under the DICOM vocabulary (vocabulary_id = DICOM), for direct use in MI-CDM.


CT

DICOM Tag Attribute Name Concept ID (DICOM)
(0018,0060) KVP 2128000818
(0018,1151) X-Ray Tube Current 2128000933
(0018,0010) Contrast/Bolus Agent 2128000788
(0018,9344) Contrast/Bolus Agent Phase 2128001396
(0018,9305) Revolution Time 2128001360
(0018,1046) Contrast Flow Rate 2128000871
(0018,1210) Convolution Kernel 2128000983
(0018,9315) Reconstruction Algorithm 2128001370
(0018,9345) CTDIvol 2128001397
(0018,1153) Exposure in μAs 2128000935
(0018,0050) Slice Thickness 2128000817
(0028,0010) Rows 2128002092
(0028,0030) Pixel Spacing 2128002096
(0018,9307) Total Collimation Width 2128001362
(0018,1100) Reconstruction Diameter 2128000909
(0028,0011) Columns 2128002093
(0018,9309) Table Speed 2128001364
(0072,0512) Reformatting Thickness 2128003565
(0018,0088) Spacing Between Slices 2128000833
(0018,9311) Spiral Pitch Factor 2128001366
(0028,1050) Window Center 2128002173
(0028,1051) Window Width 2128002174
(0008,1090) Manufacturer's Model Name 2128000177
(0008,0070) Manufacturer 2128000056
(0018,0015) Body Part Examined 2128000792
(0018,5100) Patient Position 2128001097

MR

DICOM Tag Attribute Name Concept ID (DICOM)
(0018,0087) Magnetic Field Strength 2128000832
(0018,0080) Repetition Time 2128000825
(0018,0081) Echo Time 2128000826
(0018,0010) Contrast/Bolus Agent 2128000788
(0018,0082) Inversion Time 2128000827
(0018,9117) MR Diffusion Sequence 2128001284
(0018,9025) Spectrally Selected Suppression 2128001213
(0018,1050) Spatial Resolution 2128000875
(0018,9073) Acquisition Duration 2128001251
(0018,9046) Multi-Coil Configuration 2128001230
(0018,0020) Scanning Sequence 2128000793
(0018,0083) Number of Averages 2128000828
(0008,0008) Image Type 2128000013
(0018,0021) Sequence Variant 2128000794
(0018,0050) Slice Thickness 2128000817
(0018,1100) Reconstruction Diameter 2128000909
(0028,0030) Pixel Spacing 2128002096
(0018,1314) Flip Angle 2128000999
(0028,0010) Rows 2128002092
(0028,0011) Columns 2128002093
(0018,0088) Spacing Between Slices 2128000833
(0018,1310) Acquisition Matrix 2128000997
(0054,0081) Number of Slices 2128003026
(0018,0023) MR Acquisition Type 2128000796
(0018,0095) Pixel Bandwidth 2128000839
(0018,1046) Contrast Flow Rate 2128000871
(0008,1090) Manufacturer's Model Name 2128000177
(0008,0070) Manufacturer 2128000056
(0018,9087) Diffusion b-value 2128001264
(0018,1250) Receive Coil Name 2128000988
(0018,0015) Body Part Examined 2128000792
(0020,0037) Image Orientation (Patient) 2128001703

Mammography (MG)

DICOM Tag Attribute Name Concept ID (DICOM)
(0018,7060) Exposure Control Mode 2128001189
(0018,0060) KVP 2128000818
(0028,0010) Rows 2128002092
(0028,0011) Columns 2128002093
(0028,0030) Pixel Spacing 2128002096
(0028,0101) Bits Stored 2128002123
(0008,0070) Manufacturer 2128000056
(0008,1090) Manufacturer's Model Name 2128000177
(0018,7004) Detector Type 2128001152
(0018,5101) View Position 2128001098
(0020,0062) Image Laterality 2128001707
(0018,0015) Body Part Examined 2128000792

X-ray (Chest Radiography)

DICOM Tag Attribute Name Concept ID (DICOM)
(0018,0015) Body Part Examined 2128000792
(0018,5101) View Position 2128001098
(0018,5100) Patient Position 2128001097
(0008,0060) Modality 2128000050
(0008,1090) Manufacturer's Model Name 2128000177
(0028,0010) Rows 2128002092
(0020,0062) Image Laterality 2128001707
(0018,1000) Device Serial Number 2128000840
(0028,0030) Pixel Spacing 2128002096
(0018,1110) Distance Source to Detector 2128000910
(0018,1152) Exposure 2128000934
(0018,1151) X-Ray Tube Current 2128000933
(0018,1149) Field of View Dimension(s) 2128000931
(0008,0070) Manufacturer 2128000056
(0018,0060) KVP 2128000818
(0028,2110) Lossy Image Compression 2128002228
(0018,7060) Exposure Control Mode 2128001189
(0018,7004) Detector Type 2128001152
(0018,1164) Imager Pixel Spacing 2128000944
(0018,1150) Exposure Time 2128000932
(0028,0100) Bits Allocated 2128002122
(0028,0011) Columns 2128002093
(0028,0101) Bits Stored 2128002123
(0028,1050) Window Center 2128002173
(0028,1051) Window Width 2128002174
(0028,1052) Rescale Intercept 2128002175
(0028,1053) Rescale Slope 2128002176
(0028,0004) Photometric Interpretation 2128002086

Protocol & reference

The candidate selection protocol is available from the dicom-ai-survey GitHub repository: https://github.com/dr-you-group/dicom-ai-survey

Clone this wiki locally