3D Mapping of Serial Histology Sections with Anomalies Using a Novel Robust Deformable Registration Algorithm

Tward, D, Li, X, Huo, B, Lee, B, Mitra, P, Miller, M (January 2019) 3D Mapping of Serial Histology Sections with Anomalies Using a Novel Robust Deformable Registration Algorithm. In: Multimodal Brain Image Analysis and Mathematical Foundations of Computational Anatomy.

DOI: 10.1007/978-3-030-33226-6_18


The neuroimaging field is moving toward micron scale and molecular features in digital pathology and animal models. These require mapping to common coordinates for annotation, statistical analysis, and collaboration. An important example, the BRAIN Initiative Cell Census Network, is generating 3D brain cell atlases in mouse, and ultimately primate and human. We aim to establish RNAseq profiles from single neurons and nuclei across the mouse brain, mapped to Allen Common Coordinate Framework (CCF). Imaging includes (Forumala Presented). 500 tape-transfer cut 20 (Forumala Presented). m thick Nissl-stained slices per brain. In key areas 100 $$\upmu $$ m thick slices with 0.5–2 mm diameter circular regions punched out for snRNAseq are imaged. These contain abnormalities including contrast changes and missing tissue, two challenges not jointly addressed in diffeomorphic image registration. Existing methods for mapping 3D images to histology require manual steps unacceptable for high throughput, or are sensitive to damaged tissue. Our approach jointly: registers 3D CCF to 2D slices, models contrast changes, estimates abnormality locations. Our registration uses 4 unknown deformations: 3D diffeomorphism, 3D affine, 2D diffeomorphism per-slice, 2D rigid per-slice. Contrast changes are modeled using unknown cubic polynomials per-slice. Abnormalities are estimated using Gaussian mixture modeling. The Expectation Maximization algorithm is used iteratively, with E step: compute posterior probabilities of abnormality, M step: registration and intensity transformation minimizing posterior-weighted sum-of-square-error. We produce per-slice anatomical labels using Allen Institute’s ontology, and publicly distribute results online, with several typical and abnormal slices shown here. This work has further applications in digital pathology, and 3D brain mapping with stroke, multiple sclerosis, or other abnormalities.

Item Type: Conference or Workshop Item (Paper)
Subjects: Investigative techniques and equipment > brain atlas
organism description > animal > mammal > rodent > mouse
CSHL Authors:
Communities: CSHL labs > Mitra lab
SWORD Depositor: CSHL Elements
Depositing User: CSHL Elements
Date: 1 January 2019
Date Deposited: 06 May 2021 17:50
Last Modified: 06 May 2021 17:50
URI: https://repository.cshl.edu/id/eprint/40016

Actions (login required)

Administrator's edit/view item Administrator's edit/view item
CSHL HomeAbout CSHLResearchEducationNews & FeaturesCampus & Public EventsCareersGiving