[Insight-users] multi resolution registration : problem in pyramid

[DE CRAENE Mathieu]

Hi all !

I'm  encountering a problem with the MultiResolutionImageRegistrationMethod. At the last registration level, I get (on certain images) a seg fault. Trying to debug, I have displayed the fixed image region and called the PrintSelf of the fixed Image at each level. Here is what I get : 

1. BEFORE REGISTRATION

Registration : Fixed Image Region : ImageRegion (0xbffff928)
  Dimension: 3
  Index: [50, 80, 0]
  Size: [150, 96, 19]

Registration : Fixed Image : Image (0x816a9b8)
  Reference Count: 3
  Modified Time: 237
  Debug: Off
  Observers: 
    none
  Source: (0x815a7d8) 
  Source output index: 0
  Release Data: Off
  Data Released: False
  Global Release Data: Off
  PipelineMTime: 234
  UpdateTime: 239
  LastRequestedRegionWasOutsideOfTheBufferedRegion: 1
  LargestPossibleRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [256, 256, 19]
  BufferedRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [256, 256, 19]
  RequestedRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [256, 256, 19]
  Origin: [-109.57, -219.141, -45]
  Spacing: [0.859375, 1.71875, 5]
  PixelContainer: ImportImageContainer (0x815a848)
  Reference Count: 2
  Modified Time: 238
  Debug: Off
  Observers: 
    none
  Pointer: 0x815a874
  Container manages memory: true
  Size: 1245184
  Capacity: 1245184

2. FIRST REGISTRATION LEVEL

FIXED IMAGE REGION : ImageRegion (0xbffff430)
  Dimension: 3
  Index: [25, 40, 0]
  Size: [75, 48, 9]


Image (0x81497d0)
  Reference Count: 3
  Modified Time: 292
  Debug: Off
  Observers: 
    none
  Source: (0x8148f90) 
  Source output index: 0
  Release Data: Off
  Data Released: False
  Global Release Data: Off
  PipelineMTime: 252
  UpdateTime: 443
  LastRequestedRegionWasOutsideOfTheBufferedRegion: 0
  LargestPossibleRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [128, 128, 9]
  BufferedRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [128, 128, 9]
  RequestedRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [128, 128, 9]
  Origin: [-109.57, -219.141, -45]
  Spacing: [1.71875, 3.4375, 10]
  PixelContainer: ImportImageContainer (0x81498b8)
  Reference Count: 2
  Modified Time: 293
  Debug: Off
  Observers: 
    none
  Pointer: 0x81498e4
  Container manages memory: true
  Size: 147456
  Capacity: 147456

3. SECOND (FULL RESOLUTION) REGISTRATION LEVEL

FIXED IMAGE REGION : ImageRegion (0xbffff430)
  Dimension: 3
  Index: [50, 80, 0]
  Size: [150, 96, 19]


Image (0x8149938)
  Reference Count: 3
  Modified Time: 370
  Debug: Off
  Observers: 
    none
  Source: (0x8148f90) 
  Source output index: 1
  Release Data: Off
  Data Released: False
  Global Release Data: Off
  PipelineMTime: 252
  UpdateTime: 444
  LastRequestedRegionWasOutsideOfTheBufferedRegion: 0
  LargestPossibleRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [256, 256, 19]
  BufferedRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [256, 256, 18] >>>>>>>>>>>>>>>> ???????????????
  RequestedRegion: 
    Dimension: 3
    Index: [0, 0, 0]
    Size: [256, 256, 18] >>>>>>>>>>>>>>>> ???????????????
  Origin: [-109.57, -219.141, -45]
  Spacing: [0.859375, 1.71875, 5]
  PixelContainer: ImportImageContainer (0x8149a20)
  Reference Count: 2
  Modified Time: 371
  Debug: Off
  Observers: 
    none
  Pointer: 0x8149a4c
  Container manages memory: true
  Size: 1179648
  Capacity: 1179648

It seems that at the last (full resolution) registration level, the buffered region differed from the fixed image region. I'm using the MattesMutualInformationMetric... So it seems that when the metric computes the maximum and the minimum of the region (to compute the size of bins), the metric tries to reach voxels outside the bufferd region... which could explain the segmentation fault.

I don't understand why this problem appears only at the full registration level and why the full resolution image in the pyramid is not identical to the (fixed) image I passed to the MultiResolutionImageMethod.

Any suggestion or comments are welcome !!

Thanks for your help,
Math.
----------------------------
Mathieu De Craene
Brigham and Womens Hospital
Surgical Planning Laboratory	
L1 - room 050
75 Francis Street
Boston MA 02115
decraene@bwh.harvard.edu