Dr. Mike

I agree with kopachini. A good arterial phase can be automatically segmented using the correct threshold level. The default of 150 is probably OK, but you can adjust this to meet the specific parameters if your scan. Check to see what the hounsfield density is within a major vessel, such as the aorta, on the arterial phase of the scan. Venous phase is more difficult because by the time the contrast is in the veins there is also usually organ enhancement and the organs will be included in the threshold segmentation. If you have a scan with the right kind of contrast enhancement however, it will work. If the scan isn't perfect, then you may have to fall back on manual segmentation using software like 3D slicer. Hope this helps.

With a sharp or bone reconstruction kernel you are still going to have unnatural edge enhancement. These algorithms are designed to provide maximum conspicuity of fractures, not accuracy of the bony surface. The soft tissue algorithm will provide a truer approximation of the real bone surface.

Andreas, thanks for posting this update in this thread and for your continuing work with 3D Slicer. Please continue to update us in this forum regarding new features to Slicer. Dr. Mike

How much bigger are we talking about? Are you off by 10% or more like 100%

sam17franklin, thanks for the post. Which conference did you attend? The reason I ask is that I attend and often present at most of them, so I am curious. RNSA? Mayo 3D Printing? Raw data from a CT scan is useless from a 3D printing standpoint. It is the raw data that is acquired from the x ray tube spinning around the patient in a helix. FROM the raw data axial, sagittal, and coronal image series can be generated by computing the attenuations at given rotation angles and converting that into a matrix that can represent a "slice" in the stack of images. From the raw data you can generate 5mm slices, or 2.5 mm slices, or 1.25 mm slices, or 0.75 mm slices, although with diminishing image quality the thinner you go. Raw data is used to generate the standard slices from a scan, and then it is typically deleted from the scanner to save storage space since it consumes a lot of it. I think what the radiologist meant was that if you have access to the raw data, you can reconstruct a scan with very thin slice intervals (1.25 mm of less) for 3D printing that may not have been generated as the default slice thickness for that type of study. I agree with the radiologist's advice on soft tissue reconstruction. Soft tissue is a kind of "soft kernel" reconstruction as described in my tutorial about choosing the right scan for 3D printing. Bone reconstructions are grainy as they are a type of "hard kernel" or "sharp kernel". They are designed to make it easy to see fractures, and thus have a lot of edge enhancement applied to make edges appear unnaturally noticeable. Look at the picture at the bottom left. Notice that right outside the bone is unusually dark? That darkness isn't really there. It is a artifact of the algorithm and is designed to make the bone-tissue interface have higher contrast. This makes it easier to see fractures but isn't good for 3D printing because the pixels are noisier, i.e. they look speckled like a photograph in low light. Any bright pixels confuse the software because it thinks there are thousands of bright, pixels-sized speck of bone everywhere, but really it is just noise. Each bright pixel can therefore be converted into a speck of bone in your model. Bone and lung reconstruction algorithms have this type of edge enhancement. Soft tissue reconstruction algorithms are best for 3D printing because they don't have the speckling and thus don't generate a "sea of sand" when converting to a model. Hope this helps.

embodi3D doesn't host DICOMs because they aren't anonymized by default and there is a risk of inadvertent sharing of patient information. The NRRD format is anonymized and thus safe to share online. If you are using 3D slicer and want to manually extract the bones, you can do so easily with NRRD, which kopachini has already sent you a link to the extensive femur scan library. 3D Slicer can process both DICOM and NRRD, with the only difference being time to load the scan (DICOM is slower). Of course, if you DON'T want to manually extract the bones you can send the NRRD file to the democratiz3D service on this site, and it will generate a 3D printable model for you in about 10 minutes. Just another option for you to consider. Hope this helps. Dr. Mike

I'd like to elaborate on this topic a bit, as I recently had another member inquire about this issue. The member was creating a model from a CT scan of the clavicles. As you can see, there are holes in the medial (midline) ends of both clavicles. What is causing this? Is it a problem with democratiz3D? How can it be fixed? The issue lies with the patient's anatomy and the quality of the original CT scan. In the human body there are areas where bones are naturally very thin. Sometimes, the bone surface (cortex) can be paper thin. Also, some patients who have conditions like osteoporosis may have very little calcium in their bones. Issues like this make it very hard for the CT scanner to detect the bone wall, as you can see from the image below which shows the area on the left clavicle that has a hole in the final model (red arrow). The problem isn't with democratiz3D, but with the quality of the CT scan or with the patient having thin bones (how dare they!). democratiz3D is actually creating the model exactly as it appears on the CT, its just that the CT has holes we don't want! So, what can be done? If you encounter this problem you have two options. 1) Manually fix the holes in the model with a mesh editor like Meshmixer, or 2) decrease the threshold value in democratiz3D and re-process the scan. Decreasing the threshold tells the system to capture more voxels in your model, potentially capturing more thin or osteoporotic bone. But, be careful. If you reduce the threshold too much (less than 100), you run the risk of starting to capture muscle, organs, and vessels in your bone model. If you are not sure what threshold to use, you can experiment by running your scan through democratiz3D using different thresholds. To save time, I suggest you do this on low or medium quality setting. When you find a threshold that works, you can generate your final model using a higher (and more time consuming) quality setting, like High or Ultra. If you are familiar with mesh editing software, that is probably the fastest way to correct this problem. Just delete the edge of the hole, fill it in with a new face, and run a quick smooth operation on the area. It's a 1 minute fix if you know the keyboard shortcuts. I hope this tip helps. Dr. Mike

Dr. Dave, Thanks for the detailed reply. 50 printers!?! I'm shocked because I thought I knew almost everybody that is doing high volume medical 3D printing. I've known the Mayo Clinic radiology people for years, and I don't think even they are doing the kind of volume that you are describing. I'd love to learn more about your operation. Can I message you? Thanks again for your post and being a contributor to the medical 3D printing community. Dr. Mike

Dr. Dave, this is very helpful. In Florida humidity must be a big problem. Do you have any recommendations for the plastic container? Did you make it yourself or purchase it? Is it anything like the PolyBox on Amazon? https://www.amazon.com/Polymaker-PolyBox-Filament-Filaments-Printing/dp/B075DBPY6F/ref=sr_1_3?ie=UTF8&qid=1526198281&sr=8-3&keywords=printdry

If the model looks very large, the unit is probably in mm, but I am unfamiliar with Drishti. If you have access to the DICOM data, and want to try other software, I would anonymize it to NRRD format and upload it to democratiz3D, the automatic scan to STL model conversion service on this website. It should take you about 10 minutes. democratiz3D STL files use millimeter as the unit of measurement. Hope this helps. Dr. Mike

I've recently discovered that a lot of trouble I've been having with my Ultimaker 3E when printing with PVA support that has been due to the PVA absorbing moisture from the air. I live in the Pacific Northwest and in the wintertime the humidity is very high most of the time. I stored it in an airtight bag, but apparently that wasn't good enough. The PVA has been popping and crackling during printing, presumably due to vaporizing water in the hot extrusion nozzle. Printing results have been very inconsistent, but got immediately better when I swapped in a new roll of PVA. I tried drying the old PVA roll in a food drier at 125 F for 12 hours and when I tried printing with that roll I obtained much better results, but it still wasn't as good as the fresh PVA. There was still some popping and crackling and the extruded support looked "hairy," i.e. not quite that clean. Has anybody had success with rejuvenating PVA that has been exposed to atmospheric moisture? Can you share any tips you have? Thanks in advance.

Ultimaker has just released its new S5, which has a larger build volume as well as automatic material detection (finally!) so the print will pause when you run out of material. It looks very intriguing, and I cannot wait to try it out. It's going to set you back about $6k, so it isn't cheap, but it is still within the realm of what I would call "pro-sumer" level 3D printer.

I received this inquiry from a member. I am going to post the response here so that it can help others with the same question: QUESTION: "I am printing out a spine model.... Why are there so many defects in the rendering? I can't print this out on a 3d printer, half of the vertebrae are hollow. I get these from a 3d CT and on a computer monitor, the vertebrae are whole. Just take a look at the thumbnails and you'll know what I'm talking about. I don't have the expertise or time to fill all of the defects. Is there a paid service somewhere that could do this for me? I'm just surprised the STL file wouldn't look like the 3d CT since they use the same dicom imagery?" ANSWER: If you are creating bony models and are finding that the bones have holes or other large defects in them (see above), this is probably an issue with the Threshold value used during the conversion. Threshold is the number of Hounsfield units to use to create the surface of the model. Anything above the threshold value is considered bone and is included. Anything below is not considered bone and is excluded. Normal cortical bone is very dense, greater than 300 Hounsfield units, so the default threshold of 150 is more than enough to catch it. The inside of the bone (medullary, or marrow cavity) is filled with fatty bone marrow and is a much lower Hounsfield value. If the patient has osteoporosis or very thin cortical bones they may not register as bone if the default threshold of 150 is used. You can decrease this to a lower threshold value (maybe 100 or so) and you will be more likely to capture this thin, deossified bone. If you go too low though (60 or so) you will start to capture non-bony structures like muscle. Another thing that may help get the highest quality models is using premium operations such as Very Detailed Bone and Ultra quality level. These operations are time-consuming however. To save on time, you can run your scan through democratiz3D using free operations such as Detailed Bone and medium or high quality until you find the threshold you like. Once you find the threshold value you like, you can run you scan through a final time using the highest quality (and slowest) operation settings, such as Very Detailed Bone and Ultra quality. Hope this helps! Dr. Mike