Specificity dip in metrics for long bone reconstruction

1

Hi, I am currently working on building an SSM for deformed long bones (malunited ulna).

I am trying to tweak the optimization parameters for an optimal SSM. The compactness and generalization is good, but for every run a dip in specificity remains. Does anyone have a recommendation for what parameter to change?

  1. Number of particles: 2048
  2. Initial Relative Weighting: 0.02
  3. Relative Weighting: 13
  4. Starting Regularization: 800
  5. Ending Regularization: 10
  6. Iterations per Split: 3000
  7. Optimization Iterations: 2000
  8. Multiscale Mode: on, 64

image
image342×804 30.8 KB

2

The metrics actually look pretty good to me. Since lower specificity is better, the dip around 3-5 modes is the model’s best region and the rise after is to be expected as the higher modes will start to represent noise. The specificity axis only spans 2.1-2.45 range.

1 Like