CT and MRI scans are used extensively to diagnose patient health. From the MRI of a shoulder to look for ligament damage (which I have had done after a few skiing accidents) to the CT of a hip or neck to examine bone alignment or bone health. The scans give crucial information about the insides of the body without requiring exploratory surgery.
Unfortunately, when it comes to re-creating anatomical structures many of these scans will have good in plane resolution but horrible out of plane resolution (slice spacing). For example the best MRI of my shoulder has a high quality 0.3125mm/pixel in plane resolution with a slice spacing of 4mm!
From the perspective of the physician this is fine, slices at different locations allow for a detailed look at the anatomy at different areas. From the perspective of creating a 3D model this will result in very chunky shapes when segmented and surfaced. Smoothing can only go so far.
So if the machines can output sub-millimeter resolution why not give us that data?
The answer is time. For a CT scan to give nice out of plane sub mm voxel sizes the scan will take longer which means longer exposure to radiation which is to be avoided. For an MRI there isn’t danger from ionizing radiation but a MRI scan takes longer and since MRI’s in particular are costly and time is money the scans will be done with larger slice spacing. The other issue with long MRI scans is that if a patient moves during the scan the data will for the most part be useless. This isn’t a big problem for a 2 minute scan but for a longer scan like 20 minutes can you really hold still for that amount of time? Fixturing a patient can help but can only go so far.
A recent client was looking to use scan data of heads to create accurate skull bone flaps. The patients already had MRI’s and CT scans to work with so why not try to use them. Well the best CT scan had 0.5mm in plane and 2.5mm out of plane voxel dimensions.
Segmenting the bone from the CT scan worked fairly well yet the low out of plane resolution gave significant artifacts at the superior end of the head despite significant smoothing.
In an ideal world we would just get a nice CT of each patient with sub millimeter voxel sizing (slice spacing) but that adds cost and in this case modification to the researchers IRB application. The best MRI had 0.98mm in plane and 1mm out of plane voxel size.
Segmenting bone from MRI data is less than ideal but can give decent results.
With more time spent segmenting and adjusting the smoothing parameters the holes could be filled but in general there is little contrast between bone and soft tissue in MRI. Since bone (inner bone surface in particular) was of interest the CT scan was really the best option.
So after a few meetings my client requested looking at ways of improving the results while working with the preliminary data that we had. Could results be improved by combining data sets together in hopes of ‘filling in’ the data. Adding scan data together is not trivial it requires registering the scans together then adding up the intensities at each voxel. 3DSlicer was used to accomplish this.
After these extra steps it is debatable whether the results were significantly improved. My client was able to go back to the radiologist and look for possible ways to get better data. As it turns out the CT scan was done at higher quality but saved in a lower quality. Getting another dataset saved at greater density resulted in an in plane resolution of 0.39mm and an out of plane resolution of 0.6mm. This gave four times the out of plane resolution and 25% more in plane resolution; the data was now plenty adequate for a good 3D model.
Why the scan data was saved in a resolution lower than the actual scan is likely because most doctors are used to traditional methods of looking at data as a montage of images. While this may seem odd to take a 3D volume of data and not look at it in 3D, it is indeed simpler and faster to see it in a ‘flat’ way. After working with scan data from an engineers perspective for many years it is easy to forget that doctors don’t have the desire, time, experience, or software to see the scan data in all its 3D glory.