Using image sequence of medical scans as volume data in cycles

Question

I know it's possible to do this in BI, but I need your help to do this in cycles.
For now, I have a setup to use an image sequence as volume but it only shows 1 slice/frame. I'm using generated texture coordinates for easy mapping (0-1 on each axis). The images are mapped on XY and I would like to use normalized frame# as Z coordinate to map the sequence as a volume.

Images sequence can be found here (I'm using 16-bit TIF files of MRBrain in this exemple).

Any solution is welcome but try to keep it simple and user friendly :).

Final Edit:
I had a hard time choosing between Lemon's and PGmath's answers.
Lemon's is more user friendly and flexible, and uses the image sequence without any change inside blender.
PGmath's uses a 3rd party converter to make the sequence 1 single image, but allows GPU rendering.
Well, ImageMagick's command line is : "convert +append *.tif anim.tif" from the folder where the sequence is unzipped. Not so hard. That's why I accepted PGmath's answer.

Now, i played with it, and here's the result.

The "volume" object contains the volume shader (playing with the color ramp gives funny things) and the boolean modifiers (turn them off to see the complete volume, but reduce the final density multiplier for better view).
The "Slice Control" arrow empty controls the slice position/rotation.
The "Texture Mapping" object controls...the mapping, if it need some tunning (position, rotation, scale, and then tiling if necessary?).

Answer 1

If you don't want to use OSL as in lemon's answer I have created a method to do this using only Cycles nodes. (Remember OSL isn't supported on GPU.) The downside to my method is that it requires the image sequence be entirely contained in one image, strip-style (like below).

I found an open source tool called spritetool* that seems to work pretty well for combining large amounts of images into one strip.

It seems a little clunky having such a big image, especially considering the resolution of the files you are dealing with. But it's the best I can come up with right now since the Image Texture node is not very flexible with image sequences.

---

Here is the result:

And the nodes:

It works by chopping up the image and using pixelated z-coordinates to offset the x-coordinates in a stepped manner. The green value node is where you set the number of subimages in the whole image.

Here's the .blend so you can play around with it without having to manually recreate my node tangle. (The voxel scan is a modified version of a CC-0 one I found online somewhere.)

*_{I downloaded and used it and didn't get any viruses, but don't sue me if it destroys your computer. :)}

Answer 2

You can use an osl shader to do that.

The code in the capture above creates an osl node. Here are its inputs and outputs :

The input slots

• Vector : the mapping position (in this config, this must be from object texture coordinates)
• Directory : the folder where the image files can be found
• Prefix : the image prefix (image are supposed to be in the format "prefix+counter+extension)
• Extension : file extension including the dot
• framePositions : the amount of position to form the file (ie to have 3 position like in "004", set it to 3)
• zMax : the object elevation (I don't know if there is a way to find the object dimension and center in osl..). So for a default cube, set it to 2.
• fromFrame : the first image number
• toFrame : the last image number

The output slots

• colorData : image data as a color output
• nonColorData : image data as a non color output (here the image is encoded as non color in the tif files)

The code (see comments)

The principle is to use the Z (or here pos2) value to find the good image making the file name with it.

#include "stdosl.h"

shader volumeTexture(
    point Vector = P,
    string directory="//",
    string prefix="",
    int framePositions=3,
    string extension=".tif",
    float zMax = 2.0,
    int fromFrame = 1,
    int toFrame = 100,
    output color colorData = 1,
    output float nonColorData = 1,
)
{
    // Recenter the position around the object center to -1, 1 to 0, 1
    vector pos = 0.5 * (Vector + vector(1.0,1.0,1.0));

    // Calculate the image number
    int imageNum = (int)((1.0 + toFrame - fromFrame) * (pos[2] / (zMax * 0.5)));

    // Make a format string from the given length (ie "%03d")
    string frameFormat = concat( "%0", format( "%d", framePositions ), "d" );
    // Format the current frame/image number (ie "003")
    string imageFrame = format( frameFormat, toFrame - imageNum );
    // Make the full file name
    string fileName = concat( directory, prefix, imageFrame, extension );

    // Get the color from the texture
    colorData = texture( fileName, pos[0], pos[1] );
    nonColorData = texture( fileName, pos[0], pos[1] );
}

Here is the blend :

Don't forget to set your rendering to CPU and to check 'open shading language'.

Answer 3

UDIM

Since UDIM implementation, you don't have to stitch image sequence into Sprite Sheet, to be limited by CPU calculation with OSL or wait for Voxel Data texture implementation (that was a feature lost in 2.8).

You can use image sequence natively now.
The only thing you have to do is to keep naming convention of your files (.jpg, .png) for UDIM's kind of files ... numbering has to start with 1001, like ... filename_1001.png

Open a new texture in Image Texture node. By selecting the first image, Blender recognize its UDIM format and automatically sets what needed ... grid 10 x ?rows ...

For a Cube object add a material with this node group (UDIM to Volume), it offset images position in texture space.

By default the volume texture is very soft, you need to Multiply color like 100 (it depends on your source data) ...

For even more solid effect I used Multiply 1000 (left image) and under Render Properties > Volumes > Step Rate Render/Viewport lowered step at 0.1 ... even hairs are visible now :)

You can place between Image Texture node and Multiply (Math) node black&white Color Ramp node or Map Range node or more complicated math to separate a specific of grey to bring visibility of specific parts, like muscles veins, bones ...

Blend file includes also node group controlling appearance (that I'm not proud of, but works kinda) and Cut Plane node group that is controlled by Plane object. CT Scan data are from https://graphics.stanford.edu/data/voldata/ and node group "UDIM to Volume" by koko_ze.