Ultrasound viewer

Ultrasound read in the browser: hand-held studies from a public research archive, with the probe and acquisition shown alongside.

Starting

Questions

A demonstration of browser rendering, not a reporting workstation. Nothing here is for diagnostic use.

What is this built from?

It is built on Cornerstone3D, the MIT-licensed rendering engine that also sits underneath the OHIF Viewer. OHIF itself is AGPL-3.0, so none of its code is used here.

Every image above is real patient data, pulled slice by slice from a public research archive over DICOMweb and decoded in a web worker on your own machine. On CT the pixel values are true Hounsfield units, so the window presets mean what they mean on a workstation.

What is Cornerstone3D?

An open-source JavaScript library for displaying medical images in a browser. It decodes DICOM pixel data, applies the rescale slope and intercept so CT values come out in Hounsfield units, draws through WebGL, and keeps the geometry of a series so windowing, zoom, stack scrolling and volume reconstruction behave the way they do on a workstation. It also carries the tool layer that measurement, annotation and segmentation display are built on.

It is the third generation of the Cornerstone project, licensed MIT, and developed in the open by contributors who also work on OHIF. The permissive licence is the reason it can be used on a personal site like this one.

What is the OHIF Viewer?

A complete web DICOM viewer from the Open Health Imaging Foundation: study list, hanging protocols, measurement panels, segmentation display, extensions. Research platforms deploy it as the front end for their archives, including the NCI Imaging Data Commons, whose public viewer is OHIF.

OHIF is licensed AGPL-3.0, which requires anyone who deploys a derivative over a network to publish the source of that derivative. This page avoids the question by using only Cornerstone3D, the MIT-licensed engine below OHIF, and writing its own interface on top.

What is DICOMweb?

The HTTP side of the DICOM standard. Classic DICOM moves images over DIMSE, a TCP protocol a browser cannot speak. DICOMweb exposes the same archive over ordinary web requests: QIDO-RS to search studies, series and instances, WADO-RS to retrieve metadata and pixel frames, STOW-RS to store new objects.

This viewer uses QIDO-RS to find a series and WADO-RS to pull its frames, one request per frame, straight from the archive to your browser. It never calls STOW-RS, so nothing is written back, and no server of mine sits in the middle.

What dataset is being used?

The NCI Imaging Data Commons, which hosts collections from The Cancer Imaging Archive. The images are de-identified and public, and the endpoint is a viewing-only proxy, so the page renders pixels and never offers a download. Collections in the index include NLST lung screening, LIDC-IDRI, several TCGA cohorts, UPENN-GBM, Duke breast MRI, ISPY, ProstateX, MIDRC and CBIS-DDSM. The page opens on a lung screening chest CT of 277 slices.

The finding you pick names the entry criterion of the research collection a case came from. Every patient in a glioblastoma collection has glioblastoma, so the label is true of the case, but it is not a report written from the pixels on screen.

What is not here yet?

Measurements, multiplanar reconstruction, a study browser and local file reading are the next pieces. Segmentation editing, structured reports and hanging protocols are out of scope: this is a demonstration, not a reporting workstation, and nothing here is for diagnostic use.