Three-dimensional assessments are necessary to determine the true, spatially-resolved composition of tissues

Abstract

Methods for partially resolved cellular profiling has enabled in-depth quantitative tissue mapping via thinly cut sections to study inter-patient and intra-patient differences in normal human anatomy and disease onset and progression. These methods often profile extremely limited spatial regions, which may impact the evaluation of heterogeneity due to tissue sub-sampling. Here, we applied CODA, a deep learning-based tissue mapping platform, to reconstruct the 3D microanatomy of surgically resected human pancreas biospecimens obtained from patients diagnosed with pancreatic cancer. To compare differences in the inter- and intra-tumoral heterogeneity, we assessed the bulk and spatially resolved tissue composition of a cohort of two-dimensional (2D) whole slide images (WSIs), and a cohort of 3D serially sectioned and reconstructed tissues of pancreata. Here, we show the strength of using 3D as the gold standard, by measuring the information loss and sampling problems when using WSIs and TMAs. We demonstrate that spatial correlation in microanatomical tissue content decays significantly within a span of just a few microns within tumors. As a corollary, hundreds of TMAs and tens of WSIs are required to estimate spatial bulk tumor composition with <10% error in any given pancreatic tumor. In sum, we demonstrate that 3D assessments are necessary to accurately assess tumor burden and tissue composition. These preliminary results show the importance of rate of sampling necessary to more reliably assess spatially resolved tissue composition.

Generative interpolation and restoration of images using deep learning for improved 3D tissue mapping

“…We are entering an era in which 3D imaging of biomedical samples has become a requirement as 2D assessments are not sufficient in capturing the content and morphology of multi-cellular structures, rare events, and spatial relationships among different cell types. 1 Various models have been developed to leverage 2D biomedical image stacks of histology slides, MRI images, cryo-EM slides, and tissue-cleared light-sheet images to reconstruct volumes of microanatomical structures and whole organs. Such models rely on the quality of individual 2D images within the image stacks for the accurate reconstruction of volumes.…”

“…Novel three-dimensional (3D) imaging techniques and algorithms designed to integrate large, multimodal datasets have improved our ability to assess normal anatomy and tissue heterogeneity using anatomical, molecular, -omic probes. [1][2][3][4][5][6][7] Across 3D image modalities, a common challenge emerges: a lack of resolution due to mechanical or financial constraints, or due to the presence of damaged or distorted tissue. Here, we introduce a methodology to repair and enhance 3D biological imaging data using generative artificial intelligence (AI) image interpolation.…”

A disrupted compartment boundary underlies abnormal cardiac patterning and congenital heart defects

“…Serial sections were stained with hematoxylin and eosin (H&E) and digitized at 20x magnification. CODA, a technique to 3-D reconstruct serial tissue images, was used to map the microanatomy of the hearts 35,52 . The various steps of CODA can be broken down into image registration, nuclear detection, and tissue labelling.…”

COEXIST: Coordinated single-cell integration of serial multiplexed tissue images

“…This comparison assumes uniformity across the sections and is conducted either at the pixel level by marker intensity or at the population level by cell phenotyping [3][4][5][6] . However, tissue sections are inherently heterogeneous, driven by factors including the sectioning process, the complexity of the tissue structure, or the presence of rare cell populations [7][8][9][10] . A recent study applying cyclic immunofluorescence imaging (CyCIF) to 40μm-thick melanoma sections found more than 90% of the nuclei were not fully intact within 5μm subsections 11 .…”