Notes

Fast Volume Reconstruction from Motion Corrupted Stacks of 2D Slices

Section IV-A: estimate relative amount of motion per stack of images - this is needed for reconstruction estimate correct alignment between slices

• the code provides an automated method of finding the stack that all other stacks are aligned to (called the template stack)
• in the example dataset, it seems they use the first stack as the stack that all other stacks are aligned to
• in Figure 2, this is translated into 1) motion measurement of the stacks that produce a single template such that 2) image registration is done to align all other stacks to it
• In code: StackRegistrations()

Section IV-B: motion compensated transformation of scanned 2D slices into a 3D reconstruction volume (this is the final image that is returned)

• 3D rigid volumetric registration between all stacks and template stack to account for global transformations of region of interest
• this steps figures out a transform such that a pixel in a 2D slice is able to be mapped to a particular voxel in the 3D reconstruction volume
• it also updates the voxel value in the 3D volume based on some distribution of pixel values in the original 2D slices
• in Figure 2, this would be the PSF-based volume update and the Slice to reconstructed volume transformation T
• In code: CreateSlicesandTransformations(), CoeffInit(), GaussianReconstruction(),

Section IV-C: Slice outlier removal and bias correction

• this part removes outlier slices, outlier slices are slices that are highly corrupt and not useful in the reconstruction of the 3D volume
• in Figure 2 this is the Simulate Slices and update robust statistics, EM-MOdel and Bias field and scale factors estimation sections.
• In code: InitializeEMValues(), SimulateSlices(), InitializeRobustStatistics(), EStep()

Section IV-D: Super-resolution reconstruction?

• want to minimize error between the slice pixels
• in Figure 2 this is the super resolution volume update + edge preserving regularization
• in code: MStep(), Superresolution()

Section IV-E: Slice to Volume registration

• at this point the algorithm produces a candidate approximate reconstruction volume of interest
• if the resulting volume has not converged, it will rerun the steps above
• in code: SliceToVolumeRegistration()