I cannot answer for Mimics, but here are some ideas acquired in generating .stl models and 3D printing using MR datasets with Vitrea software (Vital Images).

1. For the best results, when possible, choose 3D acquisition with small voxels, preferably isotropic or near isotropic. The resulting segmentation will be much better. It is a paradigm shift for many MR specialists but the final results are much improved for example in building oncologic 3D models.

2. The combination of sagittal, axial and coronal views will not be possible or beneficial without specific processing such as super-resolution (generating higher resolution volume reconstruction using several datasets). 

3. It is possible with an application like Vitrea to fuse 2 volumes with the same frame of reference but different acquisition techniques. For example, segmenting the anal fistula in a series with contrast and segmenting the anal sphincters in another series. See reference below.

The 2 .stl are combined in the fused image and, more importantly in the final 3D virtual model or 3D printed model. The following paper summarizes the technique: Preoperative mapping of fistula-in-ano: a new three-dimensional MRI-based modelling technique. Colorectal Dis. 2013 Nov;15(11). 

I have trouble to understand what exactly you mean by "create a 3D model from the combination of axial, sagittal and coronal views"?

Do you want to segment structures and create surface/volume models of these segmentations?

Pascal - Thank you. I will try and find out the characteristics of the scans that I am working with to see if there would be a way to improve quality in the future.

Axel - I want to create a 3d model of the intervertebral disc from MR images using mimics (for using in a finite element model). I am unsure if I should do the segmentation in only one view (such as sagittal), or if it is possible to do that in the sagittal view, then one in the transverse view, and one in the coronal view and then add it all together in another program since I can only see clearly one at a time (the reconstructed other views are very "pixelated" - not sure what is the correct term). Any suggestions are welcome.

I think I understand your problem now. Although I do not really know Mimics, the solution should be generally the same.

Each segmentation will give you a segmented volume. Now you must decide how to fuse them.

If you need a binary mask, then a "most-of-three" mode will probably the best solution. I.e., Only voxels that are segmented in at least 2 of the three volumes go into the final result. Alternatively you could also choose an "any/one-of-three" or "all-of-three" strategy.

If you can work with a "fuzzy" mask, then the average value could be stored (e.g. 0 for background, 0.33 for "segmented in only one view", 0.67 for "segmented in only two views", 1 for "segmented in all views"). This yould preserve the uncertainty of the segmentation in the border areas.

Thank you for the input. 

Some information that may be useful for anyone in a similar situation: the MR image thickness needs to be 0.5mm or as close as possible for the reconstructed views to be useful. With that, all three views become clear and you can work on it as it should be and adjust your mask on the three available views.  

Yes

yes