From DICOM to Interactive 3D: A Practical Guide for Researchers

· 7 min read · AcademicAR Team

CT slices turning into a 3D skull shown in AR on a phone

Most medical 3D work starts in the same place: a stack of DICOM images from a CT or MRI scanner. Turning that stack into something a reader can rotate, cut open, and place on the desk in augmented reality used to take a segmentation workstation, a mesh editor, and a web developer. Here is how to do it in one upload — and what to think about before you press the button.

From a DICOM series to an AR-ready 3D model in five steps From an anonymized DICOM series to a GLB + USDZ model. The raw scan is deleted after conversion.

What a DICOM series actually is

A scan is not one file. It is a series of 2D slices, each a separate DICOM file carrying pixel data plus metadata: slice position, pixel spacing, orientation, and — importantly — patient identifiers. To rebuild the volume correctly, the converter needs every slice of the series together, so you upload them as a single ZIP.

AcademicAR reads uncompressed series as well as the common compressed transfer syntaxes (JPEG Lossless, JPEG-LS, JPEG 2000, and RLE), so exports straight from a PACS or a research archive usually work as they are.

Step 1: Anonymize before you export

The rule is simple: remove patient identifiers before the data leaves your institution. Use your hospital’s de-identification workflow or a tool such as the anonymizer in your DICOM viewer, and follow your ethics approval. AcademicAR asks you to confirm anonymization, rights, and ethics responsibility on every upload, and medical uploads carry an extra confirmation on top of that.

Two further safeguards are built in:

  • The raw scan is deleted after conversion — whether the conversion succeeds or fails. Only the resulting surface mesh is kept.
  • Medical uploads are never archived or mirrored as source files, and patient identifiers are never written to logs.

A CT slice stack and a threshold selecting bone

Step 2: Choose what to extract

A volume becomes a 3D model by deciding which voxels belong to a structure. For CT you can pick one or more threshold presets:

Preset What it keeps Works for
Bone Dense tissue, 250 HU and above CT
Skin The outer body surface, -300 HU and above CT
Contrast vessels Contrast-filled vessels, 150 HU and above Contrast CT
Auto threshold An automatic (Otsu) split into two intensity classes CT and MR
Custom range Your own minimum (and optional maximum) in HU CT

Each preset you select becomes its own layer in the viewer, with its own colour. Why the presets sit where they do is explained in Hounsfield units explained. If the presets aren’t precise enough — for example, you need individual organs — segment the scan first and upload the segmentation instead (see our guide to sharing segmentations).

Step 3: Let the worker do the heavy lifting

Conversion runs in a background worker, not in your browser. It rebuilds the volume, extracts a surface for each layer, smooths and decimates it to a size that renders well on a phone, and produces a web-optimized GLB plus a USDZ for iPhone and iPad AR. You can watch the progress on the model page.

Step 4: Explore and share

The finished model opens in the viewer with a Layers panel (show, hide, fade, or recolour bone, skin, and vessels independently), a section plane to cut through it, and a 2D slice view that shows the cut in the familiar axial, coronal, and sagittal conventions. Share it with a stable link or QR code in your paper, poster, or lecture slides.

A layered skull model with a section plane in the viewer

A note on scope

These models are for research communication and education. They are surface reconstructions, not diagnostic images, and they should not be used for clinical decisions. If you need a physical model instead, see DICOM to STL for 3D printing.

Frequently asked questions

Do I need to upload every DICOM file in the series?

Yes. The volume can only be rebuilt from all of its slices, so put the whole series in a single ZIP and upload that one file.

Which compressed DICOM files are supported?

Uncompressed series and the common compressed transfer syntaxes: JPEG Lossless, JPEG-LS, JPEG 2000, and RLE. Exports straight from a PACS or a research archive usually work as they are.

Is the original scan kept after conversion?

No. The raw scan is deleted after conversion, whether it succeeds or fails, and only the surface mesh is kept. Medical source files are never archived or mirrored, and patient identifiers are never written to logs.

Create a free account and turn your next anonymized scan into an interactive figure.