The Human Organ Atlas

Abstract

We present the Human Organ Atlas (HOA), an open data repository making accessible multiscale 3D imaging of human organs. The repository also provides software tools and training resources enabling worldwide access, sharing, and analysis of these datasets, facilitating further research and the continued expansion of the HOA. The images are generated using a synchrotron imaging technique - Hierarchical Phase-Contrast Tomography (HiP-CT) that uses the ESRF's Extremely Brilliant Source, spanning whole organ imaging at around 20 μm/voxel with local volumes of interest within the intact organs imaged down to ~ 1 μm/voxel. This offers a comprehensive exploration of human anatomy, providing unparalleled insights into intricate structures and spatial relationships. The Human Organ Atlas offers researchers, clinicians, and educators a valuable resource for anatomical study, image analysis, medical education, and large-scale data mining.

Figure 1:

Overview of the Human Organ Atlas (HOA), demonstrating features including open data and registered hierarchical datasets. “Applications” show some of possible uses that the HOA has and will enable for researchers around the world, including anatomical visualization and training, studying multisystemic disease e.g. COVID-19, hypertension etc., machine learning image segmentation challenges (panel adapted from [19]), and mapping of texture features such as white matter orientation in the brain.

Figure 2:

A) Overview of the HiP-CT method pipeline (adapted from [9]). B) Diagram showing the key components of the imaging setup at ESRF BM18 (adapted from [14]). C) Example of the hierarchically aligned volumes of image data from HiP-CT scan of the Lung of S-20–29. In each case there is an overview scan, and then various other resolution scans aligned within it. In this example there are medium resolution and a high-resolution scan at 6 and 2 μm/voxel respectively. Note how multiple medium and high-resolution scans can be made within a single sample, and all can be aligned to the overview scan.

Figure 3:

Overview of the HOA portal features. A) The HOA home page with the tabs for different pages listed across the top. To discover datasets, users can either go to B) the ‘Search’ tab to search based on medical, scan and demographic metadata., or C) the ‘Explore’ tab. Within the ‘Explore’ tab (C) datasets can be ordered first by organ (C, left column) or by donor (C, right column). From both ‘Search’ and ‘Explore’ tabs the user is then presented with D) a list of datasets, and when one dataset is chosen taken to E) the dataset page with in-browser visualisation and download options for the various down sampled datasets and metadata.

Figure 4:

Overview of data currently in the HOA; A) the voxel size of each dataset against dataset size, with organs separated by marker style and colour. Many datasets are over 200GB and some are over 1TB, highlighting the importance of having in-browser viewing and down sampled datasets available for download. B) An overview of key medical conditions represented in the atlas. These medical conditions are provided as meta-data and allow researchers with an interest in a specific disease to find and utilise data that are relevant to their research questions. Note the higher proportion of age-related diseases such as cancer and hypertension as well as recently widespread diseases e.g. COVID-19, and some rarer diseases such as Dandy-Walker-Variant.

Figure 5:

Using registered hierarchical datasets to perform a super-resolved segmentation. A) Overview of the approach. Bi) Hierarchical dataset for human left kidney LADAF-2020–27. Bii) Renal glomerulus images at 1.29 μm. Ci-iii) The glomeruli outline at each imaging resolution (orange dashed line). Note how the glomeruli cannot be reliably distinguished at 25 μm/voxel. Di-Div) The manual annotations, predictions, and inference across the resolutions with the final predicted glomeruli distribution across the whole kidney. Panel adapted from results in [41], where the multiscale segmentation pipeline implementation can be accessed at: https://github.com/UCL-MSM-Bio/2025-zhou-hipct-hierarchical-segmentation and annotated high-resolution data are at: https://doi.org/10.5281/zenodo.15397768.

Figure 6:

Application of HiP-CT data to creating anatomical teaching materials. A) A full human heart imaged at 19.85 μm/voxel, allowing detailed visualization of the ventricular myocardium within the intact organ. Ai displays a virtual longitudinal section with anatomical landmarks labelled, including the left (LV) and right (RV) ventricles, mitral valve (MV), papillary muscle, and chordae tendineae. Aii and Aiii show hierarchical zoom-ins of the left ventricular wall, imaged at 6.36 μm/voxel and 2.26 μm/voxel respectively, revealing the detailed myocardial microstructure such as cardiomyocyte fibres. B) Showing an intact human kidney imaged at 25 μm/voxel where the arterial tree has been segmented prior to rendering. Each anatomically defined vessel is shown in a real example from the renal artery ca. 2cm in diameter down to the interlobular arterioles ca. 50 μm radius