cannot access tutorial on how to determine when to use nonrigid alignment in 3D reconstruction.

[yuling999666]

[yuling999666]

I have clicked "3D reconstruction with pairwise alignment", but 404 page not found. Thanks.

[YifanLu2000]

Hi @yuling999666, apologies, the current link is indeed invalid. You can refer to this page instead, and we will update it as soon as possible.

Regarding when to use non-rigid alignment in 3D reconstruction, my opinion is that if a slice is significantly different from both the previous and the next slices, it may indicate the need for non-rigid alignment to restore its original shape. You can use the st.pl.overlay_slices_2d function to plot overlapping slices for better judgment.

As for how to perform non-rigid alignment, we generally use two approaches. Suppose the aligned slices from previous pairwise rigid alignment are stored in aligned_slices, you can do following:

1. Directly concatenate surrounding slices as a reference.

spatial_key = 'align_spatial'
key_added = 'align_spatial_stage2'
neighbor_size = 1 # neighborhood size to concat
cur_idx = 1 # the slice that need to use nonrigid alignment
min_idx = max(0, cur_idx - neighbor_size)
max_idx = min(cur_idx + neighbor_size + 1, len(slices))
reference_neighbor_slices = ad.concat([aligned_slices[i] for i in range(min_idx,max_idx) if i != cur_idx])
transformed_slice = aligned_slices[cur_idx]
cur_aligned_slices, _= st.align.morpho_align(
    models=[reference_neighbor_slices, transformed_slice],
    spatial_key=spatial_key,
    key_added=key_added,
    device=device,
    rep_layer='X_pca',
    rep_field='obsm',
    dissimilarity='cos',
    # parameters that related to nonrigid
    beta=1e-1,
    lambdaVF=1e1,
    K = 50,
)
st.pl.overlay_slices_2d(slices = cur_aligned_slices, spatial_key = spatial_key, height=4, overlay_type='backward')
st.pl.overlay_slices_2d(slices = cur_aligned_slices, spatial_key = key_added + '_nonrigid', height=4, overlay_type='backward')

If the result is improved, you can insert the results into original slice:

aligned_slices[cur_idx].obsm[spatial_key][:,:2] = cur_aligned_slices[1].obsm[key_added + '_nonrigid']

2. Perform Spateo linear interpolation using the previous and next slices, and then use the interpolated result as the reference. This will comming soon in the tutorial.

The related content will also be updated in the tutorial soon. I hope this helps!

[yuling999666]

Hi @yuling999666, apologies, the current link is indeed invalid. You can refer to this page instead, and we will update it as soon as possible.

Regarding when to use non-rigid alignment in 3D reconstruction, my opinion is that if a slice is significantly different from both the previous and the next slices, it may indicate the need for non-rigid alignment to restore its original shape. You can use the st.pl.overlay_slices_2d function to plot overlapping slices for better judgment.

As for how to perform non-rigid alignment, we generally use two approaches. Suppose the aligned slices from previous pairwise rigid alignment are stored in aligned_slices, you can do following:

1. Directly concatenate surrounding slices as a reference.

spatial_key = 'align_spatial'
key_added = 'align_spatial_stage2'
neighbor_size = 1 # neighborhood size to concat
cur_idx = 1 # the slice that need to use nonrigid alignment
min_idx = max(0, cur_idx - neighbor_size)
max_idx = min(cur_idx + neighbor_size + 1, len(slices))
reference_neighbor_slices = ad.concat([aligned_slices[i] for i in range(min_idx,max_idx) if i != cur_idx])
transformed_slice = aligned_slices[cur_idx]
cur_aligned_slices, _= st.align.morpho_align(
    models=[reference_neighbor_slices, transformed_slice],
    spatial_key=spatial_key,
    key_added=key_added,
    device=device,
    rep_layer='X_pca',
    rep_field='obsm',
    dissimilarity='cos',
    # parameters that related to nonrigid
    beta=1e-1,
    lambdaVF=1e1,
    K = 50,
)
st.pl.overlay_slices_2d(slices = cur_aligned_slices, spatial_key = spatial_key, height=4, overlay_type='backward')
st.pl.overlay_slices_2d(slices = cur_aligned_slices, spatial_key = key_added + '_nonrigid', height=4, overlay_type='backward')

If the result is improved, you can insert the results into original slice:

aligned_slices[cur_idx].obsm[spatial_key][:,:2] = cur_aligned_slices[1].obsm[key_added + '_nonrigid']

2. Perform Spateo linear interpolation using the previous and next slices, and then use the interpolated result as the reference. This will comming soon in the tutorial.

The related content will also be updated in the tutorial soon. I hope this helps!

Thanks for clarification. It really helps a lot!

[yuling999666]

Hi Yifan, one more question. When you did 3D reconstruction in figure 3 for human metastatic lymph node, did you use non-rigid alignment? Thanks.

[YifanLu2000]

Hi @yuling999666, I think we didn't use the nonrigid alignment for 3D reconstruction for human metastatic lymph node dataset

[yuling999666]

Hi @yuling999666, I think we didn't use the nonrigid alignment for 3D reconstruction for human metastatic lymph node dataset

Thanks for quick response. I appreciate it.