What is direct volume rendering?
Direct volume rendering (DVR) draws a CT or MR volume without first extracting a mesh. For each pixel a ray crosses the volume, the data is sampled along the ray, each sample is mapped to a colour and an opacity by a transfer function, and the samples are composited front to back.
This keeps soft tissue, bone and vessels visible together, and lets you change the window, the colormap or a clipping plane and see the result immediately.
Why WebGPU and TSL
Ray marching is a heavy, parallel workload that fits the GPU. WebGPU gives the browser modern access to it, including 3D textures and compute shaders, and TSL lets the shader be written as portable node graphs on top of three.js instead of hand-written shader strings.
The renderer is written for WebGPU only: there is no WebGL fallback, so a browser with WebGPU is required.
How Lumivox renders a volume
The volume is uploaded as a 3D texture sized to the dataset. Each pixel casts a ray that is intersected with the volume box, stepped in millimetres so that anisotropic voxels are sampled correctly, and composited with emission-absorption until the opacity is saturated or the ray exits.
Early ray termination and a per-pixel jitter keep it fast and free of banding. Shading uses gradients computed on the fly from the data, with a headlight Blinn-Phong model.
Built to sit in a three.js scene
A Lumivox volume is a regular three.js mesh, placed with the dataset affine. Opaque geometry in the same scene can occlude the volume, and world-space clipping planes and a spherical region of interest cut the rays analytically.
GPU histogram and automatic window
A WebGPU compute shader builds the intensity histogram of the volume, processing very large series in chunks that respect the device limits, and suggests a window and level from the percentiles.
Not for diagnostic use. Lumivox is a visualization engine. It is not a certified medical device and must not be used for diagnosis, treatment decisions or any clinical purpose.