CT Scan to 3D Model Workflow using FIJI, MeshLab, and Maya
I’ve been trying out a new workflow for creating 3D models from CT scans. I reconstructed CT scans for many years back when I did whale research at Boston University. I used the program Amira. The process involved importing a stack of CT (or MRI) images, setting the threshold levels for different materials like bone and soft tissue, and then slowly segmenting out the structures of interest. The program could then build a color-coded 3D mesh from the segmented scans. But, aside from not having access to Amira anymore, I wanted to find a faster method of generating 3D models from CT scans. I was able to achieve this using open-source software.
The CT scan I used was that of the Panamanian fossil dolphin jaw bone provided for free by Smithsonian X 3D. The website provides scan data as well as full-resolution models. Since I am only interested in developing a workflow for now, I used only the CT scans as an example data set.
I used the software FIJI to view the scans. FIJI, like its precursor ImageJ, is an open source scientific image processing program. I have used ImageJ in the past, and FIJI is pretty much the same thing with many more features. I didn’t have to do any processing with the scans, so I directly used the 3D Viewer plugin after importing the scans. From there you can save the surface reconstruction as at STL file. The caveat with this method is that, in order to get a full model using all the of the CT slices, you will need a lot of computing power. That, I do not have. So I had to take the scan in parts. I also didn’t need the full resolution, so I ended up using every 4th slice. I split the scans into 5 segments, resulting in 5 STL files which could be pieced together later.
The next step was to convert the STL files into a file format that can be read by Maya. Maya cannot import STL files, so the models needed to be converted to OBJ. For this, I used another open source software called MeshLab. I only used it to import the STL and save as an OBJ, but now that I have it, I will have to experiment some other time with re-meshing models.
I then retopologized the model in Maya working off of a polygon cube primitive, resulting in the first image above. Overall, this was a fairly simple process getting from imaging data to a clean, low-res 3D model.