Preserving Metadata When Converting NIfTI to DICOM

NIfTI to DICOM Workflow: From Research Images to Clinical Formats

Converting neuroimaging data from NIfTI (Neuroimaging Informatics Technology Initiative) format to DICOM (Digital Imaging and Communications in Medicine) is a common requirement when moving datasets from research environments into clinical systems, PACS, or when preparing data for clinical review. This article outlines a clear, practical workflow covering preparation, tool selection, conversion steps, metadata handling, validation, and common pitfalls.

1. Overview: why convert?

• Compatibility: Clinical systems and PACS typically require DICOM for storage, viewing, and integration with hospital workflows.
• Metadata needs: DICOM encodes rich patient/study/series metadata required for clinical use.
• Regulatory/archival reasons: DICOM supports provenance, timestamps, and standard identifiers useful for records and audits.

2. Preparation

• Confirm intent: Decide whether converted files will be used for clinical review, archiving, or interoperability with clinical tools—this affects required metadata fidelity and patient identifiers.
• Gather source files: Ensure you have the NIfTI (.nii or .nii.gz) files and any associated JSON sidecars (e.g., BIDS format) or provenance files that contain acquisition parameters.
• Assemble metadata: Collect patient/study-level metadata: patient name/ID, DOB, sex, study date/time, referring physician, modality, and series descriptions. For MRI/CT, collect acquisition-specific fields: TR/TE, slice thickness, pixel spacing, orientation, and echo number if applicable.

3. Tool selection

Common tools and libraries:

• dcm2niix (reverse conversion not primary but helpful for metadata extraction)
• nibabel (Python) — read/write NIfTI and access header fields
• pydicom (Python) — create and modify DICOM objects
• dicom-nifti converters: HeuDiConv (primarily DICOM → BIDS), custom scripts using nibabel+pydicom, and specialized tools like “nifti2dicom” utilities or MRI vendor converters Choose: For reproducible, scriptable workflows, use Python with nibabel + pydicom. For simple GUI-based needs, look for dedicated converters that preserve metadata.

4. Mapping metadata: NIfTI header → DICOM tags

• Extract spatial info: affine matrix in NIfTI provides voxel-to-world mapping; convert to DICOM Image Position (Patient) and Image Orientation (Patient).
• Pixel data: NIfTI stores voxel intensities; map to DICOM Pixel Data, Bits Allocated, Bits Stored, High Bit, and Photometric Interpretation.
• Scalars and scaling: Respect scl_slope and scl_inter if present to reproduce correct intensity values.
• Slice ordering: Ensure proper slice order and spacing (Image Position and Slice Thickness) so slices appear correctly in clinical viewers.
• Clinical tags: Populate PatientName, PatientID, StudyInstanceUID, SeriesInstanceUID, StudyDate/Time, Modality, Manufacturer, and SeriesDescription. Generate persistent UIDs (e.g., using pydicom.uid.generate_uid()).

5. Conversion process (scripted approach with Python)

1. Read NIfTI using nibabel: load image data, affine matrix, header fields, and any sidecar JSON.
2. Prepare DICOM dataset template: start from a minimal DICOM file or create new FileDataset with appropriate filemeta (Transfer Syntax UID, SOP Class UID).
3. For each slice (2D frame) in the NIfTI volume:
   • Compute Image Position (Patient) from affine.
   • Set Image Orientation (Patient) from affine row/column direction cosines.
   • Assign InstanceNumber and SliceLocation.
   • Convert pixel array to correct dtype and byte order; set BitsAllocated, BitsStored, HighBit.
   • Insert PixelData and any required per-frame tags.
   • Add shared series/study tags for all slices.
   • Save slice as individual DICOM file or assemble into multi-frame DICOM (Enhanced MR or Secondary Capture) depending on target system.
4. If converting to multi-frame enhanced DICOM, follow the Enhanced MR Image Module structure and relevant per-frame functional groups.

Code
 ffON2NH02oMAcqyoh2UU MQCbz04ET5EljRmK3YpQ CPXAhl7VTkj2dHDyAYAf” data-copycode=“true” role=“button” aria-label=“Copy Code”>Copy CodeCopied
# read NIfTI with nibabel img = nib.load(‘image.nii.gz’) data = img.get_fdata() affine = img.affine# for each slice: for i in range(data.shape[2]):

 
slice_pixels = data[:, :, i] ds = create_dicom_dataset_template() ds.ImagePositionPatient = compute_position(affine, i) ds.ImageOrientationPatient = compute_orientation(affine) ds.PixelData = slice_pixels.tobytes() ds.save_as(f'slice_{i:03d}.dcm') 


(Implement proper UID generation, header population, datatype conversions, and file_meta.)
6. Metadata preservation and anonymization

If transferring research data into clinical systems, ensure patient identifiers are accurate and appropriate. For de-identified research data, ensure you do not accidentally populate PHI fields if the target system expects real patient data.
When anonymizing, remove or replace PatientName, PatientID, AccessionNumber, and any private tags. Keep Study/Series UIDs consistent if linking across datasets is needed.

7. Validation and QA

Open converted DICOMs in multiple viewers (e.g., OsiriX, RadiAnt, Horos, 3D Slicer with DICOM plugin) to check orientation, spacing, and intensity.
Verify DICOM tags: StudyInstanceUID and SeriesInstanceUID are unique and consistent; check Modality and SOP Class.
Compare distances and angles between voxels using slice positions; visually inspect alignment with any original coordinate references.

8. Common pitfalls

Wrong orientation/left–right flips from incorrect affine-to-DICOM mapping.
Incorrect pixel value scaling if scl_slope/scl_inter ignored.
Mismatched data types leading to clipping or loss of dynamic range.
Missing or incorrect UIDs causing PACS rejection.
Multi-frame vs single-slice expectations in target viewers — some require Enhanced DICOM format.

9. Example use cases

Preparing research MRI scans for multidisciplinary clinical review.
Archiving study datasets into institutional PACS for long-term retention.
Creating DICOMs for third-party analysis tools that accept only DICOM input.

10. Resources and further reading

nibabel documentation — reading NIfTI headers and affines.
pydicom documentation — building and writing DICOM files.
DICOM standard (PS3) — modules for Enhanced MR Image, Image Pixel, and SOP Classes.
3D Slicer and other viewers for validation.

Converting NIfTI to DICOM requires careful handling of spatial metadata, pixel scaling, and DICOM tag population. With a scripted approach using nibabel and pydicom you can build reproducible pipelines that produce clinically compatible DICOMs while preserving provenance and image fidelity.