Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Jul 19;6(1):717.
doi: 10.1038/s42003-023-04848-5.

Segmentation of human functional tissue units in support of a Human Reference Atlas

Yashvardhan Jain  1 , Leah L Godwin  2 , Yingnan Ju  2 , Naveksha Sood  2 , Ellen M Quardokus  2 , Andreas Bueckle  2 , Teri Longacre  3 , Aaron Horning  4 , Yiing Lin  5 , Edward D Esplin  6 , John W Hickey  7 , Michael P Snyder  6 , Nathan Heath Patterson  8 , Jeffrey M Spraggins  8   9 , Katy Börner  10
Affiliations

Segmentation of human functional tissue units in support of a Human Reference Atlas

Yashvardhan Jain et al. Commun Biol. .

Abstract

The Human BioMolecular Atlas Program (HuBMAP) aims to compile a Human Reference Atlas (HRA) for the healthy adult body at the cellular level. Functional tissue units (FTUs), relevant for HRA construction, are of pathobiological significance. Manual segmentation of FTUs does not scale; highly accurate and performant, open-source machine-learning algorithms are needed. We designed and hosted a Kaggle competition that focused on development of such algorithms and 1200 teams from 60 countries participated. We present the competition outcomes and an expanded analysis of the winning algorithms on additional kidney and colon tissue data, and conduct a pilot study to understand spatial location and density of FTUs across the kidney. The top algorithm from the competition, Tom, outperforms other algorithms in the expanded study, while using fewer computational resources. Tom was added to the HuBMAP infrastructure to run kidney FTU segmentation at scale-showcasing the value of Kaggle competitions for advancing research.

PubMed Disclaimer

Conflict of interest statement

M.P.S. declares the following competing interests: M.P.S. is cofounder and a member of the scientific advisory board of Personalis, Qbio, January AI, SensOmics, Protos, Mirvie, NiMo, and Oralome. He is on the scientific advisory board of Danaher, GenapSys, and Jupiter. All other authors declare no competing interests.

Figures

Fig. 1
Fig. 1. Overview of algorithm comparison setup using HuBMAP and HPA data.
Five different training and test strategies were implemented. Strategy 1: In the “HuBMAP - Hacking the Kidney” Kaggle competition, the HuBMAP kidney dataset was split into 15 WSIs for training, 10 for test, and 5 for validation. Strategy 2: To showcase generalizability, winning algorithms are trained on 5 HuBMAP colon WSIs and then tested on 2 colon WSIs. Strategy 3: Transfer learning uses 15 kidney WSIs (pretrained model from Strategy 1 is used as initial weights), 5 colon WSIs to train the model, and 2 colon WSIs to test—all data is from HuBMAP. Strategy 4 and 5 use HPA data for kidney and colon data, respectively, to test inference-time generalizability across stains without further training; Strategy 4 uses the models trained in Strategy 1 and tests on 99 HPA kidney WSIs; Strategy 5 uses the models trained in Strategy 3 and tests on 58 HPA colon WSIs.
Fig. 2
Fig. 2. FTU datasets.
a The 30 kidney and 7 colon tissue datasets were registered into the corresponding male/female, left/right HuBMAP 3D reference organs for kidney and colon to capture the size, position, and rotation of tissue blocks. b Sample kidney WSI (scale bar: 2 mm) with zoom into one glomerulus annotation (scale bar: 50 µm). c Sample colon WSI (scale bar: 500 µm) with zoom into a single crypt annotation (scale bar: 20 µm). d Metadata for 37 WSIs sorted top-down by vertical location within the reference organs; test datasets are given in bold.
Fig. 3
Fig. 3. Algorithm performance results on HuBMAP data.
Violin plots show performance for kidney on the left (one dot per 2038 glomeruli) and transfer-learning performance for colon data (one dot for each of the 160 crypts) on the right. a Dice coefficient. b Recall. c Precision. Mean Dice values per WSI for all algorithms are provided in Supplementary Table 7. Interactive versions of these graphs are at https://cns-iu.github.io/ccf-research-ftu/hubmap_violin.html.
Fig. 4
Fig. 4. Algorithm performance results on HPA data.
Violin plots show performance for kidney on the left (one dot per 337 glomeruli) and colon data (one dot for each of the 3107 crypts) on the right. a Dice coefficient. b Recall. c Precision. Mean Dice values for all algorithms are provided in Supplementary Tables 7 and 8. Interactive versions of these graphs are at https://cns-iu.github.io/ccf-research-ftu/hpa_violin.html.
Fig. 5
Fig. 5. Performance metrics terminology.
a Datasets: ground truth, predicted set, and false negatives, true positives and false positives. b Metrics: Dice coefficient, recall, and precision.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Snyder MP, et al. The human body at cellular resolution: the NIH Human Biomolecular Atlas Program. Nature. 2019;574:187–192. - PMC - PubMed
    1. Börner K, et al. Anatomical structures, cell types and biomarkers of the Human Reference Atlas. Nat. Cell Biol. 2021;23:1117–1128. - PMC - PubMed
    1. de Bono B, Grenon P, Baldock R, Hunter P. Functional tissue units and their primary tissue motifs in multi-scale physiology. J. Biomed. Semant. 2013;4:22–22. - PMC - PubMed
    1. Hayman, J. M., Jr., Martin, J. W., Jr. & Miller, M. Renal function and the number of glomeruli in the human kidney. Arch. Intern. Med. 64, 69–83 (1939).
    1. HuBMAP + HPA - Hacking the Human Body. https://kaggle.com/competitions/hubmap-organ-segmentation (2022).

Publication types