4 Medical Image Viewers
You have just learned what medical images are (Chapter 2) and how they move through a hospital (Chapter 3). The obvious next question is disarmingly practical: how do I actually open one? Double-clicking a .dcm file rarely ends well, a 4 GB whole-slide image will defeat every ordinary photo app, and an MRI study is a folder of thousands of files that only make sense as a set.
This chapter is a field guide to the viewers that solve that problem. For each one: what it is, what formats it opens, how to install it, what the license allows, and what it costs. Both readers need this chapter — the clinician exploring research tools beyond the hospital PACS, and the engineer who has a folder of DICOM files and has never seen a window/level control. And for the AI agent: the machine-readable version of this chapter lives in data/viewers.csv.
Most viewers in this chapter are not cleared for primary diagnosis. Software used to make clinical decisions about real patients must be a regulated medical device (FDA-cleared in the US, CE-marked under MDR in the EU) — a bar only a few entries here meet, and the ones that do say so explicitly. Everything else is for research, education, and software development.
4.1 Three kinds of “viewer”
The word covers three different species, and knowing which one you are holding avoids most confusion:
- Clinical workstations — the FDA/CE-cleared software radiologists actually read from (enterprise PACS clients, OsiriX MD, MIM). Regulated, validated, and priced accordingly.
- Research viewers — free desktop applications built by and for the scientific community (3D Slicer, ITK-SNAP, napari, ImageJ/Fiji, QuPath). This is where new AI methods meet images first.
- Viewer components — libraries and web frameworks (OHIF, Cornerstone3D, NiiVue) that developers embed to build the other two. If you are building an imaging product, you will likely start here rather than write a renderer from scratch.
A useful mental model: the hospital PACS viewer is a sealed appliance; research viewers are workshops; viewer components are the parts bin.
4.2 The research workhorses
4.2.1 3D Slicer
What it is. The closest thing research imaging has to a universal workstation: visualization, segmentation, registration, and analysis for 3D and 4D data, developed continuously since the late 1990s.
Formats. DICOM (full study import with its own database), NRRD, NIfTI, MetaImage, VTK models, STL meshes, and many more via plugins.
Installation. Point-and-click installers for Windows, macOS, and Linux from slicer.org; a Preview release tracks the cutting edge.
License & cost. Free, open source under a permissive BSD-style license — explicitly usable in commercial products. Not cleared for clinical use.
Best for. Anything 3D: scrolling a CT in three planes, volume rendering, drawing or editing segmentations, converting between formats. Its Sample Data module makes it the fastest way to get legal example studies onto your screen.
AI ecosystem. The Extension Manager is the killer feature: MONAI Label connects Slicer to an AI segmentation server for active-learning annotation; TotalSegmentator and MONAI Auto3DSeg run full-body CT/MR segmentation with one click; a built-in Python console scripts everything. When later chapters say “run an automatic segmentation and inspect it,” this is the default place to do it.
4.2.2 ITK-SNAP
What it is. A deliberately small tool that does one thing exceptionally well: labeling anatomical structures in 3D images. Where Slicer is a workshop, ITK-SNAP is a scalpel.
Formats. DICOM series, NIfTI, NRRD, MetaImage, Analyze, MGH.
Installation. Installers for all three major OSes from itksnap.org.
License & cost. Free, open source (GPL).
Best for. Producing ground-truth segmentations. Its semi-automatic active contour (“snake”) mode grows a 3D contour from seed points — often the fastest route to a clean label without any ML at all, and still a respectable baseline to compare learned methods against.
4.2.3 napari
What it is. A Python-native n-dimensional viewer that lives inside your analysis environment rather than beside it: viewer.add_image(array) and your NumPy tensor is on screen, with layers, annotations, and GPU-accelerated rendering.
Formats. TIFF/OME-Zarr and anything imageio reads natively; DICOM, NIfTI, and NRRD via plugins (napari-itk-io, napari-medical-image-formats) or by loading through SimpleITK/nibabel yourself.
Installation. pip install "napari[all]" or conda — it is a Python package, not an application, though bundled installers exist.
License & cost. Free, open source (BSD-3).
Best for. The ML engineer’s inner loop: visually checking tensors mid-pipeline, overlaying model predictions on inputs, annotating training data with plugins. Leans microscopy, but nothing stops you from feeding it a CT volume.
4.2.4 ImageJ / Fiji
What it is. The elder statesman of scientific imaging (born 1997 at NIH) and still ubiquitous in biology labs. Fiji (“Fiji Is Just ImageJ”) is the batteries-included distribution with hundreds of plugins and an auto-updater — for practical purposes, install Fiji.
Formats. TIFF at heart, plus roughly 160 scientific formats through the bundled Bio-Formats library — including OME-TIFF and basic DICOM (single images fine; full multi-series studies are not its strength).
Installation. Download, unzip, run — fully portable, no admin rights needed (fiji.sc).
License & cost. Free. Original ImageJ is public domain; ImageJ2 core is BSD-2; Fiji as a distribution is GPL.
Best for. 2D image processing and measurement at lab scale: intensity analysis, stacks, batch macros. For medical imaging specifically it is the utility knife you reach for on single images, not the tool for reading a CT study.
4.2.5 QuPath
What it is. The standard open tool for digital pathology. Ordinary viewers cannot open a 100,000 × 100,000-pixel whole-slide image at all; QuPath’s tile-streaming architecture makes browsing one feel like Google Maps.
Formats. Whole-slide formats via OpenSlide and Bio-Formats: Aperio SVS, Hamamatsu NDPI, Leica SCN, MIRAX, OME-TIFF, and more.
Installation. Installers for Windows, macOS, Linux from qupath.github.io.
License & cost. Free, open source (GPLv3).
Best for. Everything in Chapter 19: viewing slides, annotating regions, cell detection and classification, and Groovy scripting for batch analysis across a slide archive.
4.3 Web viewers and building blocks
4.3.1 OHIF Viewer
What it is. The Open Health Imaging Foundation’s zero-footprint web viewer: nothing to install for the end user, just a URL. More than an application, it is a framework — the de-facto starting point for building custom imaging applications, powering research portals like the NCI’s Imaging Data Commons.
Formats. DICOM over DICOMweb (it expects an archive to talk to, e.g. Orthanc or a cloud PACS), including overlays like DICOM SEG, RTSTRUCT, and SR.
Installation. For users: none — visit a deployment (a public demo lives at viewer.ohif.org). For developers: Node/yarn or Docker.
License & cost. Free, open source under MIT — the most permissive license in this chapter, deliberately chosen so companies can build products on it.
AI ecosystem. Rendering comes from Cornerstone3D (the underlying MIT-licensed JavaScript engine), and a MONAI Label mode brings AI-assisted annotation into the browser.
4.3.2 Weasis
What it is. A free, cross-platform desktop DICOM viewer designed to be launched by hospital systems — the “view images” button in an EHR or PACS portal often opens Weasis.
Formats. DICOM and DICOMweb, with proper handling of clinical niceties (presentation states, key objects, multi-monitor hanging).
Installation. Installers for Windows, macOS, Linux (weasis.org); also web-launchable.
License & cost. Free, open source (EPL-2.0 / Apache-2.0).
Best for. A no-cost, no-license-anxiety DICOM viewer that behaves like clinical software, and the integration story when you need image viewing wired into another system.
4.4 Desktop DICOM viewers
4.4.1 Horos and OsiriX
One lineage, two licenses. OsiriX pioneered the Mac DICOM workstation; Horos is its free, LGPL, community-maintained fork (horosproject.org) — capable, Mac-only, and not cleared for diagnosis. OsiriX MD (osirix-viewer.com) is the commercial continuation: FDA-cleared as a Class II device and CE-marked, sold as a subscription (on the order of $70/month) or one-time license. The pair is the cleanest illustration in this chapter of what the regulatory line actually buys: nearly identical software, on opposite sides of it.
4.4.2 RadiAnt and MicroDicom
The lightweight Windows pair. RadiAnt (radiantviewer.com) is famously fast at opening large studies — commercial with a free trial, CE-marked in the EU but not FDA-cleared. MicroDicom (microdicom.com) is freeware for non-commercial use and the quickest answer to “a colleague sent me a CD of DICOM files.” Both are viewers in the narrow sense: browsing and measuring, not segmentation or analysis.
4.5 Neuroimaging specialists
Neuroscience grew its own viewer lineage around NIfTI rather than DICOM. FSLeyes, the viewer of the FSL suite, is the standard for overlaying statistical maps, atlases, and diffusion data on brain volumes (free for non-commercial use under FSL’s terms; installs with FSL or via conda). NiiVue is the modern web answer: a BSD-licensed WebGL library that renders volumes, meshes, and tractography in any browser or notebook — less an application than a component, and the easiest way to put a brain volume into a web page. Honorable mentions in the same family: MRIcroGL and Mango.
4.6 The clinical tier: MIM and the enterprise workstations
At the top of the market sit regulated enterprise platforms — MIM (mimsoftware.com) is a representative example, alongside vendor PACS clients from Sectra, GE, Philips, and Agfa. MIM specializes in radiation oncology and nuclear medicine: PET/CT fusion, SUV quantification, RT structure handling, dose review, with FDA-cleared modules throughout. There is no download link and no price list; licensing is quote-based and sold to departments, not individuals. You will not install these to learn from this book — they appear here so the landscape has its ceiling: this is what “cleared for diagnosis, integrated with the hospital, priced like infrastructure” looks like, and it is the tier an imaging AI product must ultimately meet.
4.7 The comparison table
The living version of this table is data/viewers.csv in the book’s repository — corrections welcome as pull requests.
| viewer | platforms | key formats | license | cost | cleared for diagnosis | best for |
|---|---|---|---|---|---|---|
| 3D Slicer | Win / macOS / Linux | DICOM, NRRD, NIfTI, MetaImage, VTK, STL | BSD-style (permissive) | Free | No | 3D research workhorse: segmentation, registration, AI extensions |
| ITK-SNAP | Win / macOS / Linux | DICOM, NIfTI, NRRD, Analyze, MGH | GPL | Free | No | Focused manual + semi-automatic 3D segmentation |
| napari | Win / macOS / Linux (Python) | TIFF/OME-Zarr native; DICOM, NIfTI, NRRD via plugins | BSD-3 | Free | No | Python-scriptable n-D viewing inside an ML workflow |
| ImageJ / Fiji | Win / macOS / Linux (Java) | TIFF + ~160 formats via Bio-Formats (OME-TIFF, basic DICOM) | Public domain / GPL (Fiji) | Free | No | 2D scientific image analysis and batch macros |
| QuPath | Win / macOS / Linux (Java) | Whole-slide: SVS, NDPI, OME-TIFF, and more (OpenSlide/Bio-Formats) | GPLv3 | Free | No | Gigapixel pathology slides: viewing + cell-level analysis |
| OHIF Viewer | Any browser (zero-footprint) | DICOM via DICOMweb (incl. SEG / RTSTRUCT / SR) | MIT | Free | No (framework) | Embedding a web viewer into your own application |
| Weasis | Win / macOS / Linux + web launch | DICOM (+ DICOMweb) | EPL-2.0 / Apache-2.0 | Free | No | Free desktop DICOM viewer with PACS/HIS integration |
| Horos | macOS only | DICOM | LGPL | Free | No | Free Mac DICOM viewer (OsiriX lineage) |
| OsiriX MD | macOS only | DICOM | Commercial | ~$70/mo or one-time license | Yes — FDA Class II / CE | Clinical-grade reading on a Mac |
| RadiAnt | Windows only | DICOM | Commercial | Paid license (free trial) | No (EU CE only) | Fast, lightweight Windows DICOM browsing |
| MicroDicom | Windows only | DICOM | Proprietary freeware | Free (non-commercial) | No | Quick free DICOM viewing on Windows |
| FSLeyes | Win / macOS / Linux (Python) | NIfTI, GIFTI, DICOM (basic) | Apache-style (FSL terms; free for non-commercial) | Free (academic) | No | Neuroimaging: overlays, atlases, fMRI/DTI display |
| NiiVue | Any browser (JS/WebGL library) | NIfTI, DICOM, meshes, tractography | BSD-2 | Free | No | Embedding neuro visualization into web pages and notebooks |
| MIM | Win / macOS (enterprise) | DICOM (incl. RT objects, PET/CT fusion) | Commercial | Quote-based | Yes — FDA-cleared modules | Enterprise radiation-oncology and nuclear-medicine workflows |
4.8 Choosing a viewer
- “I have a folder of DICOM files and just want to look” → Weasis or MicroDicom (free), RadiAnt (Windows, fast), Horos (Mac).
- “I need to scroll a CT/MRI properly, in three planes, with 3D” → 3D Slicer.
- “I need to create segmentation ground truth” → ITK-SNAP for focused structures; 3D Slicer (+ MONAI Label) when AI assistance should do the first pass.
- “I’m mid-Python-pipeline and want to see my tensors” → napari.
- “It’s a pathology slide” → QuPath.
- “It’s a brain study with overlays” → FSLeyes; NiiVue if it needs to live in a browser.
- “I’m building an application” → OHIF/Cornerstone3D for DICOM, NiiVue for neuro.
- “It’s for actual patient care” → only the cleared tier: your PACS, OsiriX MD, MIM — nothing else in this chapter.
4.9 The agentic outlook
Viewers are where humans see images; increasingly they are also where agents act on them. Three properties of this chapter’s tools matter for that future. First, scriptability is an agent interface: 3D Slicer’s Python console, napari’s plain-Python API, ImageJ macros, and QuPath’s Groovy engine mean an agent can drive the same tools a human does — load a study, run a segmentation, export a measurement — without any bespoke integration. Second, web components dissolve the boundary between viewer and application: an agent that drafts a finding can render its evidence through OHIF or NiiVue directly in its own interface, turning “the viewer” from a destination into a widget the agent composes. Third, the regulatory line does not move: an agent operating research viewers inherits their non-diagnostic status, and an agentic workflow that touches patient care needs the cleared tier — a theme Chapter 10 takes up properly.
4.10 Further reading
- Each project’s own documentation is the authoritative source for formats and licensing: slicer.org, itksnap.org, napari.org, fiji.sc, qupath.readthedocs.io, docs.ohif.org, weasis.org.
- Fedorov et al., “3D Slicer as an image computing platform for the Quantitative Imaging Network,” Magnetic Resonance Imaging (2012) — the canonical Slicer reference.
- Bankhead et al., “QuPath: Open source software for digital pathology image analysis,” Scientific Reports (2017).
- Ziegler et al., “Open Health Imaging Foundation Viewer: An extensible open-source framework,” JCO Clinical Cancer Informatics (2020).