Segmenting functional tissue units across human organs using community-driven development of generalizable machine learning algorithms

Abstract

The development of a reference atlas of the healthy human body requires automated image segmentation of major anatomical structures across multiple organs based on spatial bioimages generated from various sources with differences in sample preparation. We present the setup and results of the Hacking the Human Body machine learning algorithm development competition hosted by the Human Biomolecular Atlas (HuBMAP) and the Human Protein Atlas (HPA) teams on the Kaggle platform. We create a dataset containing 880 histology images with 12,901 segmented structures, engaging 1175 teams from 78 countries in community-driven, open-science development of machine learning models. Tissue variations in the dataset pose a major challenge to the teams which they overcome by using color normalization techniques and combining vision transformers with convolutional models. The best model will be productized in the HuBMAP portal to process tissue image datasets at scale in support of Human Reference Atlas construction.

Functional Tissue Units in the Human Reference Atlas

“…Within the HRA effort, we are using this catalog of FTUs for several use cases. For example, we ran Kaggle competitions to develop scalable and generalizable segmentation code to identify FTUs in images 2,14 . We use the segmentation code to automatically count the number of FTUs per unit area and to compute general biomarker expression values for genes, proteins, lipids, and metabolites.…”

“…The human body serves as the root node in the center, the largest anatomical structures (organs) are placed further out, and smaller sub-structures and tissue types of branch outwards from the organs' leaf nodes denoting cell types. (2) The second graph contains all the blood vessels in the HRA, with the chambers of the heart in the center, and increasing smaller vessels more distal to the heart again branching outwards from the center. Nodes in the two radial tree graphs meet at points where the HRA indicates a vessel supplies or drains the corresponding anatomical structure.…”

“…Therefore, tissues and organs in the body consist of many repeating FTUs to perform the overall function of the organ. We previously generalized this original definition by describing an FTU as "the smallest tissue organization that performs a unique physiologic function and is replicated multiple times in a whole organ" 2 . Here, we extend the definition of FTUs by allowing them to form a nested hierarchy to accomplish various functions.…”

“…In this perspective, we present a catalog of the first 22 FTUs for the HRA. The current version of this atlas contains (1) the length and diameter of each FTU; (2) the vasculature pathways that connect FTUs to the heart and amongst each other; (3) the cell types within each FTU; and (4) 2D illustrations of each FTU showing their prototypical cell types, cell shapes, sizes, and spatial arrangement. We map all anatomical structures to their corresponding Uber-anatomy Ontology (UBERON) ID 3 and all cell types to their Cell Ontology (CL) ID 4 to enable linkage to other datasets.…”