Improved Post-Processing Framework for Whole Slide Imaging and 3D Reconstruction

Whole slide imaging (WSI) enables high-resolution digitization of histological sections, but 3D reconstruction from serial slides remains hindered by illumination artifacts, tile inconsistencies, and section-specific geometric distortions. We present an enhanced reconstruction framework that addresses both tile-level intensity inhomogeneity and instability in deformable registration. First, we introduce an overlap-aware illumination correction that refines an initial flat-field estimate by adding an intensity equality constraint in overlapping tile regions. Second, intensity normalization of sections was performed before reconstruction to ensure consistent intensity profiles among serial sections. Lastly, we incorporate level-set method (LSM) constraints into an existing volumetric reconstruction framework to stabilize alignment. The framework was developed for marmoset brain histology, including within-specimen blockface photography and magnetic resonance imaging (MRI). The LSM reconstruction was validated using a synthetic banana image with a known initial shape and with mouse histology reconstructed using an external 3D MRI reference. Overlap-aware correction in the marmoset WSI significantly reduced residual gridding artifacts across 112 sections (Friedman χ²(2) = 890.04, p < 2.2 × 10^-16). Within the marmoset dataset, intensity normalization reduced inter-slice intensity variability (coefficient of variation: 0.392 to 0.246) and incorporation of LSM consistently improved registration accuracy by approximately 10% in Normalized Mutual Information scores and 3%-5% in Structural Similarity Index Measure. LSM also yielded improved structural correspondence in both gray matter and white matter structures.