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

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: March 10, 2024

ABSTRACT The development of novel imaging platforms has improved our ability to collect and analyze large three-dimensional (3D) biological datasets. Advances in computing have led an extract complex spatial information from these data, such as the composition, morphology, interactions multi-cellular structures, rare events, integration multi-modal features combining anatomical, molecular, transcriptomic (among other) information. Yet, accuracy quantitative results is intrinsically limited by quality input images, which can contain missing or damaged regions, be poor resolution due mechanical, temporal, financial constraints. In applications ranging intact (e.g. light-sheet microscopy magnetic resonance imaging) sectioning based serial histology section transmission electron microscopy), data become paramount. Here, we address challenges leveraging frame interpolation for image motion (FILM), a generative AI model originally developed temporal interpolation, range 3D types. Comparative analysis demonstrates superiority FILM over traditional linear produce functional synthetic its better preserve including microanatomical cell counts, well quality, contrast, variance, luminance. repairs tissue damages images reduces stitching artifacts. We show that decrease time synthesizing skipped images. demonstrate versatility method with wide modalities (histology, tissue-clearing/light-sheet microscopy, imaging, species (human, mouse), healthy diseased tissues (pancreas, lung, brain), staining techniques (IHC, H&E), pixel resolutions (8 nm, 2 µm, 1mm). Overall, potential improving resolution, throughput, datasets, enabling imaging.

Language: Английский

---

CODAvision: best practices and a user-friendly interface for rapid, customizable segmentation of medical images

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: April 14, 2025

Abstract Image-based machine learning tools have emerged as powerful resources for analyzing medical images, with deep learning-based semantic segmentation commonly utilized to enable spatial quantification of structures in images. However, customization and training algorithms requires advanced programming skills intricate workflows, limiting their accessibility many investigators. Here, we present a protocol software automatic images guided by graphical user interface (GUI) using the CODAvision algorithm. This workflow simplifies process microanatomical enabling users train highly customizable models without extensive coding expertise. The outlines best practices creating robust datasets, configuring model parameters, optimizing performance across diverse biomedical image modalities. enhances usability CODA algorithm ( Nature Methods , 2022) streamlining parameter configuration, training, evaluation, automatically generating quantitative results comprehensive reports. We expand beyond original implementation serial histology demonstrating numerous modalities biological questions. provide sample data types including histology, magnetic resonance imaging (MRI), computed tomography (CT). demonstrate use this tool applications metastatic burden vivo deconvolution spot-based transcriptomics datasets. is designed researchers interest rapid design basic understanding anatomy.

Language: Английский

---

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

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 5, 2024

Abstract Failure of septation the interventricular septum (IVS) is most common congenital heart defect (CHD), but mechanisms for patterning IVS are largely unknown. We show that a Tbx5 + /Mef2cAHF progenitor lineage forms compartment boundary bisecting IVS. This coordinated population originates at first- and second field interface, subsequently forming morphogenetic nexus. Ablation progenitors cause disorganization, right ventricular hypoplasia mixing lineages. Reduced dosage CHD transcription factor TBX5 disrupts position integrity, resulting in defects (VSDs) defects, including Slit2 Ntn1 misexpression. Reducing NTN1 partly rescues cardiac mutant embryos. Loss or causes VSDs perturbed septal distributions. Thus, we identify essential cues direct to pattern proper septation, revealing new birth defects.

Language: Английский

---

Power-law growth models explain incidences and sizes of pancreatic cancer precursor lesions and confirm spatial genomic findings

Science Advances, Journal Year: 2024, Volume and Issue: 10(30)

Published: July 26, 2024

Pancreatic ductal adenocarcinoma is a rare but lethal cancer. Recent evidence suggests that pancreatic intraepithelial neoplasia (PanIN), microscopic precursor lesion gives rise to cancer, larger and more prevalent than previously believed. Better understanding of the growth-law dynamics PanINs may improve our ability understand how miniscule fraction makes transition invasive Here, using three-dimensional tissue mapping, we analyzed >1000 found size distributed according power law. Our data suggest in bulk, PanIN can be predicted by general growth behavior without consideration for heterogeneity microenvironment or an individual’s age, history, lifestyle. models intraductal spread fusing lesions drive observed distribution. This analysis lays groundwork future mathematical modeling efforts integrating incidence, morphology, molecular features tumorigenesis demonstrates utility combining experimental measurement with dynamic tumorigenesis.

Language: Английский

---

Combined assembloid modeling and 3D whole-organ mapping captures the microanatomy and function of the human fallopian tube

Science Advances, Journal Year: 2024, Volume and Issue: 10(39)

Published: Sept. 27, 2024

The fallopian tubes play key roles in processes from pregnancy to ovarian cancer where three-dimensional (3D) cellular and extracellular interactions are important their pathophysiology. Here, we develop a 3D multicompartment assembloid model of the tube that molecularly, functionally, architecturally resembles organ. Global label-free proteomics, innovative assays capturing physiological functions (i.e., oocyte transport), whole-organ single-cell resolution mapping used validate these assembloids through multifaceted platform with direct comparisons tissue. These techniques converge at unique combination parameters highest similarity reference tube. This work establishes (i) an optimized human for vitro studies pathophysiology (ii) iterative customized models organs microanatomically accurate by combining tunable tissue methods.

Language: Английский

---

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

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: May 7, 2024

ABSTRACT Multiplexed tissue imaging (MTI) and other spatial profiling technologies commonly utilize serial sectioning to comprehensively profile samples by each section with unique biomarker panels or assays. The dependence on sections is attributed technological limitations of MTI panel size incompatible multi-assay protocols. Although image registration can align serially sectioned MTIs, integration at the single-cell level poses a challenge due inherent biological heterogeneity. Existing computational methods overlook both cell population heterogeneity across modalities information, which are critical for effectively completing this task. To address problem, we first use Monte-Carlo simulations estimate overlap between 5μm-thick sections. We then introduce COEXIST, novel algorithm that synergistically combines shared molecular profiles information seamlessly integrate level. demonstrate COEXIST necessity performance several applications. These include combining improved profiling, rectification miscalled phenotypes using single panel, comparison platforms resolution. not only elevates platform validation but also overcomes constraints MTI’s limitation full nuclei slide, capturing more intact in consecutive thus enabling deeper lineages functional states.

Language: Английский

---

3D histology reveals that immune response to pancreatic precancers is heterogeneous and depends on global pancreas structure

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 6, 2024

SUMMARY Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal cancer for which few effective therapies exist. Immunotherapies specifically are ineffective in pancreatic cancer, part due to its unique stromal and immune microenvironment. intraepithelial neoplasia, or PanIN, the main precursor lesion PDAC. Recently it was discovered that PanINs remarkably abundant grossly normal pancreas, suggesting vast majority will never progress cancer. Here, through construction of 48 samples cm 3 -sized human pancreas tissue, we profiled microenvironment 1,476 3D at single-cell resolution better understand early evolution tumor determine how inflammation may play role progression. We found bulk strongly correlates PanIN cell fraction. response around heterogeneous, with distinct hotspots cold spots appear disappear span tens microns. Immune generally mark locations higher grade dysplasia near acinar atrophy. The composition these dominated by naïve, cytotoxic, regulatory T cells, associated fibroblasts, macrophages, little similarity less-inflamed PanINs. By mapping FOXP3+ cells 3D, present density larger lesions compared smaller PanINs, initiation not exhibit an immunosuppressive response. This analysis demonstrates while common pancreases most individuals, pivotal role, both microscopic scale, demarcating regions significance

Language: Английский

---