PYCAD SSR Engine

Viewer-agnosticCPU and GPU renderingBuilt in Rust

Huge medical images, in any browser

A server-side rendering engine for medical imaging. It renders multi-gigabyte CT, MR and tomosynthesis studies on a server and streams only what each screen shows, so the web viewer you already use opens them in seconds, on any laptop and any connection.

Secondsinstead of minutesMPR and 3D on multi-gigabyte CTs: 2.3 s instead of 188 s on a 5 GB study
Megabytesinstead of gigabytesSent to the browser: 9 MB instead of 5.1 GB for that same study
Under 1 GBinstead of up to 11 GBOf browser memory for MPR on a 5 GB CT, so laptops keep up
About 20 readersper 8-vCPU serverScrolling smoothly under a heavy synthetic load, with one GPU covering 3D

What it enables

Big data stays on the server. The viewer stays fast.

Open any study size

Photon-counting CT, thin-slice protocols and breast tomosynthesis open as quickly as a small study, because the browser never downloads the whole series.

Read from anywhere

A home connection or an ordinary laptop is enough: the viewer receives megabytes of images instead of gigabytes of raw data.

Serve many readers per server

One engine process serves every user and shares loaded studies between sessions, so capacity grows with CPU cores, not with copies of the data.

Keep the full dataset off the device

The complete volume stays on your server. Each browser only receives the images it displays.

Keep your viewer and your PACS

The engine plugs into an existing web viewer through a connector and reads from your PACS or storage. Your interface, tools and workflow don’t change.

Add 3D without heavy clients

Volume rendering runs on the server’s GPU and streams to the browser, so 3D works on devices that couldn’t render it themselves.

How it works

One engine between your data and your viewer

  1. Reads your data once

    The engine pulls a series from your PACS over DICOMweb, or from your storage, into a bounded cache. It decodes compressed and multi-frame formats such as JPEG 2000 and HTJ2K itself, and never writes back.

  2. Renders what each view needs

    Slices, MPR and oblique planes, thick slabs and 3D volume rendering, on all CPU cores and a GPU. Only the newest request per view is rendered, so fast scrolling never queues stale frames.

  3. Streams to the viewer

    Light previews while the user scrolls or rotates, then the full-resolution image when they stop. Pixel values are preserved, so measurements and HU statistics stay exact.

  4. Connects through a small connector

    Each viewer gets a connector that routes the heavy views to the engine and leaves everything else as it was, including the viewer’s own tools and reports.

Works with your viewer

Built to connect to many viewers

The engine doesn’t come with its own viewer. It sits behind the one your users already know.

Available now

OHIF 3 and OHIF 2

Connectors for stock OHIF 3.13 and the last OHIF 2 release, added through OHIF’s own extension mechanism, with no fork and no rebuild. Tested for MPR, 3D, measurements, segmentation and reports.

See the OHIF integration and results

Through a connector

Custom and in-house viewers

The engine has a documented interface, and the OHIF connectors follow a template meant for new viewers. For an in-house or commercial web viewer, we write the connector and keep your interface and tools as they are.

Ask about your viewer

Proof

Measured, not estimated

In the OHIF viewer · public data

MPR on screen, 5 GB CT

OHIF + PYCAD engine
2.3 s
Stock OHIF
188 s

3D on screen, 2.5 GB CT

OHIF + PYCAD engine
5.8 s
Stock OHIF
83 s

Peak browser memory, MPR on a 5 GB CT (OHIF 2)

OHIF + PYCAD engine
0.9 GB
Stock OHIF
10.9 GB

Slices shown scrolling tomosynthesis (OHIF 2)

OHIF + PYCAD engine
97%
Stock OHIF
3%

Stock OHIF 3 and OHIF 2, same study with and without the engine, over the real internet. All OHIF results

Capabilities

What the engine renders

2DLarge stacks with smooth scrolling, MPR and oblique planes, thick slabs (MIP, MinIP, average), window/level kept in the browser
3DGPU volume rendering with OHIF’s 25 presets, MIP, MinIP and average blending, rotation streamed at interactive rates
DataDICOM over DICOMweb from your PACS, multi-frame and enhanced CT/MR, breast tomosynthesis, JPEG 2000 and HTJ2K decoded on the server
OverlaysSegmentations (DICOM SEG and labelmaps) on the engine’s planes, with the same voxel counts as the viewer’s own path
FidelityOriginal pixel values preserved: HU, ROI statistics and measurements match the viewer without the engine

Deployment and scaling

Sized by CPU cores, with GPUs only where 3D needs them

One processserves every user, with shared, bounded caches for volumes, decoded images and GPU textures
About 0.4 vCPUper actively scrolling reader in our heaviest synthetic test, about 20 readers per 8-vCPU server
1 GPUcovered the 3D of 100 simulated users; not every server needs one

Packaged as a container with GPU access. Capacity figures come from simulated readers working without pauses, a heavier load than real reading; your numbers depend on your studies and how many different ones are open at once.

Have a viewer that struggles with large studies?

Tell us which viewer you run, where your images live and how large your biggest series are. We’ll tell you what the engine can change.

Book a call

The PYCAD SSR Engine is not cleared as a medical device. Results come from tests with public or synthetic data and depend on the network, the server and the studies.

We build custom medical imaging platforms — advanced DICOM viewers, AI segmentation, and the clinical systems around them.

Copyright © 2026 PYCAD. All Rights Reserved.