Reconstruction: File Descriptions

HARD RECONSTRUCTION

Reconstructions performed with a reference volume rather than a probabilistic atlas

17-0333_hard_manualLabel_merged.mgz

Ground truth manual segmentation that has been transformed into the space of the final reconstructed volume. The ground truth was outlined on the 2D source photos and then composed into a volume by applying the saved transformations from the volume reconstruction This version has some label classes merged into each other, both to match with the output of SAMSEG and to remove boundaries that were deemed unreasonable to expect SAMSEG to pick up. For example the VDC is merged into the white matter.

17-0333_hard_manualLabel.mgz

Ground truth manual segmentation that has been transformed into the space of the final reconstructed volume The ground truth was outlined on the 2D source photos and then composed into a volume by applying the saved transformations from the volume reconstruction

17-0333.hard.mask.mgz

Volume composed of 2D binary photo masks, transformed into the space of the final reconstructed volume.

17-0333.hard.mat

Matlab save file holding the parameters from the reconstruction optimisation process History files contain parameters and costs from iterations of the reconstruction. Useful for generating visualisations. ParamsOptim holds the final iteration parameters for Modes 1&2 of the optimisation

17-0333.hard.recon.mgz

The reconstructed volume of photos including all three colour channels

17-0333.hard.recon.blue.1mm.mgz

17-0333.hard.recon.green.1mm.mgz

17-0333.hard.recon.red.1mm.mgz

I can’t find these in my filesystem so I can’t check directly, but from what I remember one of the options for segmentation was to output the 3 colour channels and run samseg on all three channels at once. I’d assume these are the three channels split out and resampled to 1mm voxels (either in-plane or isotropic, you’d have to check). There were issues with this approach so mostly I used a greyscale representation instead and operated on one channel. Mostly I think the staining procedure meant that some labels that were grouped together in the generative model weren’t actually coloured the same. For example exterior grey matter ended up more brown than the interior grey-matter structures which were more red.

17-0333.hard.warped_ref.mgz

The 3D reference mask warped onto the space of the reconstructed photo-volume. This is part of the registration process and is only an affine transformation performed through modifying the header not resampling.

17-0333.hard.recon.grayscale.1mm.mgz

17-0333.hard.recon.grayscale.mgz

Reconstructed volumes converted to grayscale to be put through samseg

17-0333_hard_samseg_merged.mgz

Result of running samseg on the hard reconstructed photo-volume This version has some label classes merged into each other, both to match with the output of SAMSEG and to remove boundaries that were deemed unreasonable to expect SAMSEG to pick up. For example the VDC is merged into the white matter.

17-0333_MRI_overlay.mgz

The ex-vivo mri that has been warped to overlay on-top of the reconstructed volume. This was done by modifying the header with the affine transformation generated for the volume mask as part of the reconstruction algorithm.

17-0333_MRI_overlay_SamsegEroded.mgz

17-0333_MRI_overlay_SamsegMerged.mgz

17-0333_MRI_overlay_Samseg.mgz

17-0333_MRI_overlay_SamsegSuppressed.mgz

These are the results of running samseg on the ex-vivo MRI volumes and then modifying the header to overlay on the reconstructed volume. Merged has the same label merging as mentioned above Suppressed removes some labels that weren’t merged as they weren’t present in any of the photo segmentation (such as brainstem and cerebellum) Eroded Is just the boundary voxels of each label. Used for generating figures as the boundary view in freesurfer can be a bit dodgy.

SOFT RECONSTRUCTION

Reconstructions performed with a probabilistic atlas

17-0333_manualLabel.mgz

17-0333_soft_manualLabel_merged.mgz

17-0333_soft_manualLabel.mgz

Ground truth manual segmentation that has been transformed into the space of the final reconstructed volume The ground truth was outlined on the 2D source photos and then composed into a volume by applying the saved transformations from the volume reconstruction

17-0333_soft_history.mat

Matlab save file holding the parameters from the reconstruction optimisation process History files contain parameters and costs from iterations of the reconstruction. Useful for generating visualisations. ParamsOptim holds the final iteration parameters

17-0333_soft_mask.mgz

Volume composed of 2D binary photo masks, transformed into the space of the final reconstructed volume.

17-0333_soft.mgz

The reconstructed volume of photos including all three colour channels

17-0333_soft_regatlas.mgz

A rasterization of the probabilistic atlas (probability of brain vs not brain, used in the registration process) overlayed on the reconstruction

17-0333_soft_samseg_merged.mgz

Result of running samseg on the soft reconstructed photo-volume This version has some label classes merged into each other, both to match with the output of SAMSEG and to remove boundaries that were deemed unreasonable to expect SAMSEG to pick up. For example the VDC is merged into the white matter.