Unblock step 13 and write the cropped cine frames

[avhekkink]

All 20 steps of the pipeline now run successfully end to end on V-0004

Step 13 was blocked with an ITK error and produced nothing, so step 14 had no input. The crop itself was also never finished. The container's cropImages hardcodes the crop box to one case, so this fix crops on the host instead, using the codebase's own utilities and a padded bounding box of the LA.

Changes

• CropImage repairs the NIfTI header before it loads the segmentation. Why? The cemrg/ccta container rebuilds the header from the image affine and leaves qform_code at 0, which ITK refuses. SegmentExtract already calls CemrgNiftiUtils::RepairNiftiQform on the output of the same container; CropImage did not. This is the failure the step reported.

• CropImage writes the cropped frames. The crop loop and the afmotion cropImages call were commented out, so the step measured a bounding box, printed it to stdout and stopped. It now builds a mitk::Cuboid from the padded box and writes nifti/dcm-crop-<N>.nii through CemrgCommonUtils::CropImage, which MmcwView::CropinIMGS already uses to crop a cine series against one cube. Step 14 needs those files: the container writes <project>/nifti/dcm-crop- .nii into tracking/imgTimes.lst, and the host MIRTK register reads that list.

• afmotion cropImages is not used, and should not be as-is. uac_pipeline.py hardcodes the crop box to a specific case from another dataset and ignores the box the plugin measures. Its CropNifti_v3.py also writes the affine shift with a guessed sign for each axis, correct only for one orientation. mitk::BoundingObjectCutter calls IndexToWorld on the region start instead, which holds for any direction matrix.

• A HasLabel test guards the bounding box.
  itk::LabelStatisticsImageFilter::GetBoundingBox returns a default-constructed std::vector when the image does not hold the label, so the loop that follows read six elements past the end of an empty vector. This never fired only because the segmentation always found an atrium.

• The padding measures the box, not the position of the box. Each index was scaled by 30% of its own value, so the padding depended on where the atrium sat in the volume rather than how large it was. The padding is now a proportion of the length of each axis, held inside the image.

• CropImage reports its failures. It names a missing nifti/dcm-10.nii, replaces the working copy rather than silently keeping the old one, stops when the segmentation writes nothing, and names the frame it cannot read or crop. The box now reaches afib_log<date>.log through MITK_INFO; std::cout never did.

Impacted areas

Area	Files
Atrial Strain Motion plugin	AtrialStrainMotionView.cpp

Nothing shared changes. CemrgCommandLine, CemrgCommonUtils and CemrgNiftiUtils are used as they are, so no other plugin is affected.

The cropped frames land in nifti/, beside the frames the user supplies, because the container hardcodes that path.

Testing

Run in the GUI on V-0004, 21 frames of 512 x 512 x 269:

• Step 13. No ITK dialogue. The log carries the repair, the atrium box 7 306 98 255 73 249 and the padded box 0 395 51 302 21 268. It wrote 21 frames in 113 seconds and left no scratch file.
• Steps 14 to 20. All complete. MIRTK wrote tracking/tsffd.dof at 9,039,844 bytes. Every per-frame output holds different content, checked by md5sum, so no step repeated a result or wrote an empty one.

This is the first end-to-end run of all twenty steps.

Not tested:

• Any dataset other than V-0004. Both demo datasets carry 21 frames of the same geometry.
• A frame that is not signed short, or one on a different grid from dcm-10.nii.
• Windows. The step runs Docker and writes files; neither is platform specific, but nobody has run it there.