Hi everyone - I have segmented menisci for which I want to do SSM using Shapeworks however, I am having trouble getting good results post optimization where the images appear jagged/spiky. The results post grooming at least visually look fine. I have tried multiple iterations of grooming and optimization parameters and none of them really solve this issue. Attaching below the latest I tried including the original segmentations:
I have tried all different kinds of methods with the optimization, lower particle counts from 64 onwards, adding/removing procrustes, increasing iterations, converting to meshes etc. One thing that improves things but not good enough is increasing relative weighting but does not get desired results. So I am confused about next steps and unsure if it is a grooming problem or an optimization problem and how to proceed. Your help will be much appreciated!
P.S. this is also just a test run with 6 later menisci. I have more data I can use in case that solves the problem.
When you run this in Studio, do the particles spread out well in the initial splits (lower particle counts)? If not, I would try reducing the initial relative weighting (e.g 0.05). Also, increasing the optimization iterations will not generally help if the particles are not already in correspondence. You can see where before optimization by reducing the optimization iterations to 0.
They do spread out well in the initial splits (I think), and I think correspondence isn’t bad across the menisci although it isn’t perfect either. Attaching an image of the particles at 32 particle count below with the remaining settings being the same during optimization. And also attaching two more below with everything else being the same at 32 particles and optimization iterations at 0 OR initial relative weighting at 0.05.
Increased particles to 64, 128 and 256 while keeping everything else the same. Unsure what I am looking for so would be grateful for your input! But the optimization results are not improving as can be seen:
One thing I did notice was that when I increased the relative weighting to a 100, the optimized shapes didn’t have these spikes/jagged edges but they seemed very similar to each other visually, so unsure if that is a valid result.
The other thing is that the jaggedness decreases (doesn’t go away though) if I increase the optimization iterations but also unsure if that is a valid approach given that the 0 optimization iterations are not giving good results.
I think the next thing I would try is turning on “normals” in optimization and use a normals weight of perhaps 5. This may help the optimization in the later splits.
Thank you for your response, I was able to generate a decent model based on these recommendations.
I am now trying to compare two groups via the group particle p-values option and when I export the mesh or its scalars, I see 8700+ values in the csv whereas I used 512 particles for my SSM. Would you be able to explain what these mean? And how do I get p-values for the particles I am seeing visually in Studio?
In addition to the “Export Current Mesh Scalars” option, there is another export option, “Export Current Particle Scalars”. This will have the 512 values you are looking for. If you are currently displaying the group p-values, then these values will be the p-values.
You can identify any particle by hovering over it with the mouse and pressing the “1” key, this will highlight it in white and color the rest of the colors by the distance on the mean shape. The particle ID will be shown in the status bar.
