VBoussot commited on
Commit
4ef6ef8
·
verified ·
1 Parent(s): 89834cc

Update PresetDatabase.json

Browse files
Files changed (1) hide show
  1. PresetDatabase.json +3 -4
PresetDatabase.json CHANGED
@@ -21,11 +21,10 @@
21
  "preprocess_function": "",
22
  "iterations": 1000,
23
  "short_description": "Rigid registration using mutual information and a multi-resolution pyramid."
24
-
25
  "description": "This preset performs rigid alignment using an Euler transform optimized with Adaptive Stochastic Gradient Descent. It uses a 4-level multi-resolution strategy and Mattes mutual information as similarity metric. Initial alignment based on image centers are enabled to ensure robust convergence for multimodal images."
26
  },
27
  {
28
- "display_name": "IMPACT BSpline M730",
29
  "parameter_maps": [
30
  "ParameterMap_SynthRad2023_MRI_pelvisV3.txt"
31
  ],
@@ -34,8 +33,8 @@
34
  ],
35
  "preprocess_function": "Preprocess:standardize_MRI",
36
  "iterations": 1900,
37
- "short_description": "Synthrad 2023.",
38
- "description": "blablabla1"
39
  }
40
  ]
41
  }
 
21
  "preprocess_function": "",
22
  "iterations": 1000,
23
  "short_description": "Rigid registration using mutual information and a multi-resolution pyramid."
 
24
  "description": "This preset performs rigid alignment using an Euler transform optimized with Adaptive Stochastic Gradient Descent. It uses a 4-level multi-resolution strategy and Mattes mutual information as similarity metric. Initial alignment based on image centers are enabled to ensure robust convergence for multimodal images."
25
  },
26
  {
27
+ "display_name": "IMPACT BSpline M730",
28
  "parameter_maps": [
29
  "ParameterMap_SynthRad2023_MRI_pelvisV3.txt"
30
  ],
 
33
  ],
34
  "preprocess_function": "Preprocess:standardize_MRI",
35
  "iterations": 1900,
36
+ "short_description": "IMPACT-based multimodal BSpline registration with deep semantic features (M730)",
37
+ "description": "A deformable BSpline registration using the IMPACT metric to align semantic features extracted from pretrained models (MIND + M730). Mutual information and bending energy are combined for additional robustness and regularization. The method uses 5 resolution levels."
38
  }
39
  ]
40
  }