Speaker
Description
Porto-Sinusoidal Vascular Disease (PSVD) is a life-threatening chronic liver disorder, primarily driven by severe portal hypertension complications [1]. Currently, definitive clinical diagnosis relies on the identification of specific and non-specific non-cirrhotic histological microvascular lesions, such as the progressive obliteration of intrahepatic portal venules (OPV) and portal vein stenosis, detectable through Light Microscopy (LM) [2]. Yet, the precise cellular and structural mechanisms underlying this disease remain poorly characterised, leaving a critical gap in both understanding and clinical management [3]. Addressing this gap requires mapping the liver’s highly specialised hierarchical vascular architecture. While being useful for lesion detection, conventional 2D LM fails capture this 3D organization: it requires tissue slicing and staining, which may introduce artifacts and only offers 2D perspectives of a continuous 3D vascular architecture [4]. Synchrotron X-ray Phase-Contrast Imaging (X-PCI) overcomes these limitations by enabling non-destructive 3D visualisation of intact biopsies with high soft tissue contrast and sub-micron resolution (0.65 μm) [5]. Nevertheless, imaging the full multi-scale vascular architecture generates massive terabyte-scale datasets that overload computational memory when processed entirely at maximum resolution, demanding specialised processing strategies.
To address this computational challenge, we developed an efficient multiscale AI-based pipeline (Figure 1 in attachments) applied to liver biopsies from control and PSVD rat models [6] imaged at the FaXToR beamline (ALBA Synchrotron) [7]. First, we performed bit depth reduction and standardised the reconstructed measurements through contrast normalization. Next, we stitched the samples using an in-house adaptation of different resolutions levels: Binning 4 (2.6 µm), 2 (1.3 µm) and 1 (native, 0.65 µm). We utilised the resulting stitching parameters to map the global macrovascular tree. Resolution-specific nnU-Net models [8] then segmented the vascular architecture at each scale: a binary model mapped large vessel walls at low resolution (DSC = 0.988), while a multiclass model delineated terminal sinusoids and suppressed background at higher resolutions, producing a continuous vascular representation from macrovessels to sinusoids.
As a preliminary result, a resolution-dependent logical subtraction isolated the exclusive microvascular component, revealing a severe 28.33% density reduction in the diseased sample. This capillary dropout is completely hidden under downsampled regimes, proving that native maximum resolution is mandatory to detect microvascular pruning. Furthermore, a spatial translation framework integrated the 3D surface meshes from multiple resolutions into a single coordinate system, enabling the simultaneous visualisation shown in Figure 1 (in attachments). Together, these results establish a non-destructive multiscale pipeline for characterising the structural remodelling underlying PSVD, with potential applicability to other vascular anatomies including placental, pulmonary and coronary networks.
References:
[1] A. De Gottardi, C. Sempoux, and A. Berzigotti, “Porto-sinusoidal vascular disorder,” Journal of Hepatology, vol. 77, no. 4, pp. 1124–1135, Oct. 2022, doi: 10.1016/j.jhep.2022.05.033.
[2] A. De Gottardi et al., “Porto-sinusoidal vascular disease: proposal and description of a novel entity,” The Lancet Gastroenterology & Hepatology, vol. 4, no. 5, pp. 399–411, May 2019, doi: 10.1016/S2468-1253(19)30047-0.
[3] G. Campreciós, B. Bartrolí, C. Montironi, E. Belmonte, J. C. García-Pagán, and V. Hernández-Gea, “Porto-sinusoidal vascular disorder,” in Sinusoidal Cells in Liver Diseases, Elsevier, 2024, pp. 445–464. doi: 10.1016/B978-0-323-95262-0.00022-X.
[4] R. Xuan et al., “Phase-contrast computed tomography: A correlation study between portal pressure and three dimensional microvasculature of ex vivo liver samples from carbon tetrachloride-induced liver fibrosis in rats,” Microvascular Research, vol. 125, p. 103884, Sep. 2019, doi: 10.1016/j.mvr.2019.103884.
[5] A. Patera et al., “FaXToR: the hard X-ray micro-tomography beamline at the Spanish synchrotron ALBA,” J Synchrotron Rad, vol. 33, no. 1, pp. 207–217, Jan. 2026, doi: 10.1107/S160057752500997X.
[6] G. Campreciós et al., “Interspecies transcriptomic comparison identifies a potential PORTO‐SINUSOIDAL vascular disorder rat model suitable for in vivo drug testing,” Liver International, vol. 44, no. 1, pp. 180–190, Jan. 2024, doi: 10.1111/liv.15765.
[7] A. Patera, A. G. Zippo, A. Bonnin, M. Stampanoni, and G. E. M. Biella, “Brain micro‐vasculature imaging: An unsupervised deep learning algorithm for segmenting mouse brain volume probed by high‐resolution phase‐contrast X‐ray tomography,” Int J Imaging Syst Tech, vol. 31, no. 3, pp. 1211–1220, Sep. 2021, doi: 10.1002/ima.22520.
[8] H. Goharbavang, A. T. Ashitkov, A. Pillai, J. D. Wythe, G. Chen, and D. Mayerich, “Segmentation and modeling of large-scale microvascular networks: a survey,” Front. Bioinform., vol. 5, p. 1645520, Oct. 2025, doi: 10.3389/fbinf.2025.1645520.