Multiscale and multimodal imaging for three-dimensional vascular and histomorphological organ structure analysis of the pancreas

Gabriel Alexander Salg, Verena Steinle, Jonas Labode, W. Wagner, Alexander Studier‐Fischer, Jochen Reiser, Elyes Farjallah, Jonas Albers, Tim Hilgenfeld, Nathalia A. Giese, Wolfram Stiller, Felix Nickel, Martin Loos, Christoph Michalski, H.-u. Kauczor, T. Hackert, Christian Dullin, Philipp Mayer, Hannes KenngottHPB 2024 · DOI

TL;DR

This paper introduces a novel multiscale, multimodal imaging pipeline to create detailed 3D reconstructions of the pancreatic vascular network, bridging the gap between macroscopic clinical imaging and high-resolution microscopic views. This method allows for unprecedented analysis of pancreatic microanatomy, paving the way for better understanding of diseases and potential applications in bioengineering.

What Problem Does This Paper Address?

The pancreas, with its intricate exocrine and endocrine functions, relies heavily on its vascular system for communication and nutrient supply. However, our understanding of this complex network is limited, largely due to the reliance on 2D histology, which provides only snapshots and destroys tissue. Existing 3D imaging techniques often lack the resolution to capture fine vascular details or the field of view to contextualize them within the whole organ. This study aims to overcome these limitations by developing a comprehensive imaging process that integrates multiple techniques to visualize the pancreatic vasculature from macroscopic to near-cellular levels.

Methods

The core of this research is a multiscale imaging workflow that combines several techniques:

  • Radiopaque Vascular Casting (RVC): A specialized casting agent is injected into the vasculature to permanently stain blood vessels for X-ray-based imaging. This method is crucial as it preserves tissue integrity for subsequent histological analysis.
  • Clinical Computed Tomography (CT) & Digital Volume Tomography (DVT): These techniques are used for whole-organ imaging, providing macroscopic context and identifying larger vascular structures. DVT offers higher resolution than standard CT.
  • Micro-Computed Tomography (µCT): This technique allows for mesoscopic imaging, enabling the visualization of smaller arterioles and venules, bridging the gap between macroscopic and microscopic scales.
  • Synchrotron-based Propagation-Based Imaging (PBI): This high-resolution technique is applied to tissue sections, providing near-cellular level detail and allowing for the identification of capillary structures.
  • Histology & Immunohistochemistry: Standard histological staining (e.g., anti-insulin, anti-CD31) is used to correlate imaging data with cellular and tissue morphology, validating the vascular segmentation and identifying specific tissue types like islets of Langerhans.
  • Data Analysis: Sophisticated algorithms are employed for 3D reconstruction, segmentation of vascular networks, and quantitative analysis of vessel diameter and branching patterns (e.g., Strahler order, Gaussian Mixture Models).

Datasets

The study utilized porcine pancreatic specimens. While no specific public datasets were used for training, the paper presents a comprehensive dataset generated from these specimens, which is made available as 3D surface meshes of CT, DVT, µCT, and PBI data. This generated dataset serves as a benchmark for multiscale pancreatic vascular analysis.

Key Insights and Ideas

  • Bridging Scales: The most significant insight is the successful integration of imaging modalities that span vastly different resolutions, allowing for a continuous, traceable view of the vascular network from large arteries down to near-capillary structures.
  • Virtual Histology: PBI, in particular, offers a non-destructive "virtual histology" approach, enabling detailed 3D structural analysis of soft tissues without the loss of information inherent in traditional sectioning.
  • Islet Vasculature Identification: The high-resolution imaging, especially PBI, allowed for the identification of dense vascular networks within pancreatic tissue, which, when correlated with insulin staining, were confirmed to be intra-islet capillaries. This provides a new avenue for studying islet function and pathology.
  • Quantitative Vascular Analysis: The development and application of advanced segmentation and analysis algorithms enable objective, quantitative characterization of vascular morphology and arborization, moving beyond qualitative descriptions.

Key Results

  • The imaging pipeline successfully visualized pancreatic vasculature across multiple scales, from macroscopic arteries and veins (CT/DVT) down to arterioles, venules, and partially capillaries (µCT/PBI).
  • CT and DVT imaging provided whole-organ context, with mean vessel diameters of 2.72 mm and 2.34 mm, respectively.
  • µCT imaging resolved structures down to 9 µm, identifying arterioles and venules with a mean diameter of 0.71 mm.
  • PBI achieved the highest resolution (2.0 µm voxel size), revealing capillary structures with a mean vessel diameter of 0.05 mm.
  • Correlation with immunohistochemistry confirmed the identification of islets of Langerhans based on their dense vascularization.
  • Quantitative analysis revealed distinct vascular branching patterns and diameter distributions across different imaging modalities.

Limitations and Future Work

  • Casting Agent Shrinkage: The radiopaque casting agent can shrink during tissue processing (formalin fixation, paraffin embedding), potentially leading to an underestimation of vessel diameters. Future work could involve quantifying this shrinkage factor.
  • Capillary Filling Incompleteness: While PBI identified capillaries, a uniform end-to-end representation was not always achieved, possibly due to incomplete filling or resolution limitations.
  • Tissue Preparation Artifacts: Reprocessing tissue for histology can lead to shrinkage and dissolution of the casting agent.
  • AI for Segmentation: The authors suggest that AI-driven segmentation tools could replace semi-automatic methods, improving efficiency and objectivity.
  • Data Integration: Achieving seamless, dynamic integration of data across different modalities, accounting for sample deformations, remains a computational challenge.
  • Human Pancreas Studies: While porcine pancreata are similar to human ones, future studies should investigate human specimens to analyze pathologies directly.
  • Bioartificial Organ Design: The developed methods hold promise for generating organ-specific rule-based frameworks for bioartificial vascular networks, aiding in the design of functional engineered tissues.

Summary updated

BibTeX
@inproceedings{salg2024multiscale,
  author = {Gabriel Alexander Salg and Verena Steinle and Jonas Labode and W. Wagner and Alexander Studier‐Fischer and Jochen Reiser and Elyes Farjallah and Jonas Albers and Tim Hilgenfeld and Nathalia A. Giese and Wolfram Stiller and Felix Nickel and Martin Loos and Christoph Michalski and H.-U. Kauczor and T. Hackert and Christian Dullin and Philipp Mayer and Hannes Kenngott},
  title = {{Multiscale and multimodal imaging for three-dimensional vascular and histomorphological organ structure analysis of the pancreas}},
  booktitle = {HPB},
  year = {2024},
  doi = {10.1016/j.hpb.2024.03.864},
}
Figure