Speaker
Description
Whole Slide Imaging (WSI) has become a crucial modality in modern histopathology, enabling the digital conversion of tissue on glass slides into high-resolution virtual slides. WSI supports the reconstruction of three-dimensional (3D) volumes from serial histological sections, providing new opportunities to study the microstructural organization of complex tissues. However, 3D reconstruction from serial WSI is challenged by structural artifacts introduced during tissue preparation and imaging. These inconsistencies can accumulate across sections, degrading volumetric integrity. The level-set method (LSM) is a numerical technique that embeds surface information into a signed distance function to capture complex geometric and topological changes. The proposed workflow mitigates residual distortions by integrating LSM into the non-linear stage of an established pipeline.
This study utilized in-vivo MRI and histological data from a single adult female marmoset brain. A T1-weighted anatomical scan was acquired at 300 μm isotropic resolution. Following imaging, the brain was cryosectioned coronally at 20 μm thickness, and sections were co-stained with Hoechst 33342 and Myelin Basic Protein. All image registrations were performed using Advanced Normalization Tools (ANTs). The MRI volume was rigidly and affinely registered to the blockface volume to align with the histological cutting plane. Blockface images were aligned using a four-step coarse-to-fine framework consisting of pairwise, sequential, and high-pass filtered transformations to preserve global shape while improving local slice-to-slice alignment. Histological sections were reconstructed using the blockface volume as reference. A final deformable non-linear registration step corrected localized distortions using level-set maps generated from segmented white and gray matter.
The coarse-to-fine reconstruction achieved an NMI of 0.4470 and SSIM of 0.8022. The reference reconstruction pipeline achieved an NMI of 0.4617 and SSIM of 0.8030. The proposed method achieved an NMI of 0.4946 and SSIM of 0.8297, demonstrating improved inter-modality alignment and structural consistency. This framework addresses spatial information loss introduced during histological processing by providing a flexible, structurally guided approach for 3D reconstruction.
| Keyword-1 | Level-Set-Method |
|---|---|
| Keyword-2 | Image Reconstruction |
| Keyword-3 | Neuroimaging |