Datasets for acquiring 3D bone shapes: a systematic review.

Abstract

Three-dimensional (3D) human bone shapes are essential for various engineering, preclinical, and clinical applications. Despite their importance, they are not easily findable because existing datasets are scattered across multiple sources with inconsistent documentation. This study aimed to systematically identify publicly available datasets that provide 3D bone shapes or CT/MRI scans suitable for generating 3D bone shapes, to extract their specifications, and analyze the specifications to provide an overview of the current landscape of these datasets. A systematic search was conducted across academic databases (Scopus, Web of Science, PubMed), data repositories (Zenodo, Mendeley Data), and the Grand Challenge platform. Dataset specifications, including available 3D bone shapes and their representation formats, imaged regions and modalities, subject metadata, and data usage policies, were extracted and analyzed. A total of 152 datasets covering over 371,000 subjects were identified. 44 datasets included 3D bone shapes, of which 30 provided voxel segmentation masks, 5 included only surface meshes, and 9 contained both; none offered computer-aided design (CAD) models. Among 3D bone shapes, the spinal column was the most covered (28 datasets), and the thorax was the least covered (10 datasets). In conclusion, 3D bone shape datasets are scattered across various sources, making them difficult to find. To address this challenge, we compiled their information into a guide github.com/BoneHub/Public-Datasets. The scarcity of mesh and CAD models highlights a gap that limits progress in orthopaedic, biomechanical, and engineering research and design. This review consolidates detailed specifications of 3D bone shape datasets to assist in identifying, comparing, and utilizing existing resources more efficiently.