How can I create a logical mask from 3D data points?

I have an array of data points (size ~2000x3) compiled of points from several segmented CT slices. The end problem I'm trying to solve is to create an .stl file from the data, but a key step I'm struggling with is an efficient way of creating a 3D logical mask from the data.

What I'm trying to accomplish is an efficient way of creating a mask while keeping data precision on the order of 10E-4. I've tried scaling all the data points up (so they become integers), but that leaves the matrices WAY to large to be functional.

For example, one typical data point will be (before and after scaling):

PO_1 =

   -1.6550    1.1250    0.0625

>> PO_1 = 10000.*PO_1 + 16551

PO_1 =

           1       27801       17176

These could be used as locations in the mask, but you can see how large the matrix will end up.

Are there any hints or steps that I seem to be missing?

Thanks

[Roger Stafford]

I see no reason why you should rescale your data or round them to integers just to obtain a logical mask. The two endeavors are totally unrelated in my opinion. Perhaps you could explain your reasoning in this. A logical entity asks a question which is either yes or no concerning data and it can be asked just as easily and accurately with data in one range as it can if you have rescaled it.

By the way, rescaling is ordinarily understood to be a multiplication process, not one of addition. By adding the large constant you have used, there is danger of increasing round-off errors. Rounding to integer will further degrade the data.

[Cedric]

As Roger, I don't understand why you need what seems to be 3D coordinates to become positive integer indices.

If you want to "flag" points that correspond to a region of space, you can work directly with the original coordinates. For example, using a fake dataset of 5 points:

 PO =
    0.6000   -1.0000    0.2000
    0.9000    0.8000   -1.2000
   -1.0000    0.2000   -1.1000
   -0.5000    1.3000    0.7000
    0.6000    0.8000   -1.9000

For flagging points in the cube x,y in [0.5,1] and z in ]-2,-1], you would just do

 >> selection = PO(:,1)>=.5 & PO(:,1)<=1 & ...
                PO(:,2)>=.5 & PO(:,2)<=1 & ...
                PO(:,3)>-2 & PO(:,3)<=-1
 selection =
     0
     1
     0
     0
     1

Then you could extract these points coordinates or any data with matching indices, using selection and logical indexing:

 >> PO(selection,:)
 ans =
    0.9000    0.8000   -1.2000
    0.6000    0.8000   -1.9000

[Steve]

Thank you Roger and Cedric for responding.

I think I may be slightly off in my understanding of a logical mask. What I care about in my 3D volume is just whether or not there is a point at a certain location - hence the mask.

To preserve all the information in, for example, 1000 data points, wouldn't it require a 3D mask of 1000 "1"'s and the rest "0"'s, spanning the entire volume?

Assuming this sub-problem gets solved, my next task is to create a closed surface from all the points through either

mask->isosurface->surf2stl.m [ http://www.mathworks.com/matlabcentral/fileexchange/4512 ]

or

mask->CONVERT_voxels_to_stl.m [ http://www.mathworks.com/matlabcentral/fileexchange/27733-converting-a-3d-logical-array-into-an-stl-surface-mesh/content/CONVERT_voxels_to_stl/CONVERT_voxels_to_stl_example.m ]

Thanks

Answer 1

[Sean de Wolski]

Adam A has a bunch of files for *.stl meshes and similar. I'd recommend giving his submissions a try.