Related Experiment Video
Updated: Aug 21, 2026
Outer-Boundary Assisted Segmentation and Quantification of Trabecular Bones by an Imagej Plugin
Published on: March 14, 2018
BONE DATA LAKE: A SCALABLE PLATFORM FOR LARGE-SCALE RADIOGRAPHIC TRABECULAR BONE TEXTURE ANALYSIS
1Tribology Laboratory, School of Mechanical and Civil Engineering, Curtin University, Perth, Australia.
Introduction:
Trabecular bone (TB) texture analysis of hand and knee X-ray images may serve as a biomarker for detection, prediction and monitoring of OA. However, the scale and heterogeneity of imaging data pose challenges for data storage, standardization, cataloguing, security, and efficient OA analysis.
Objective:
Our objective is to address this problem through the development of a platform, called the Bone Data Lake (BDL) that enables the secure storage of large numbers of X-ray images, TB texture regions and TB texture parameters, independent of their format, size and source.
Methods:
The BDL was developed as a cloud-based platform (Wolski et al. 2025; Figure 1) using Amazon Web Services. It comprises three components: a raw data storage, a processed data storage, and data reference system. The first component operates as a repository of data in its raw original format. The second component reformats the data into a standardized format. The last component is the central catalog, which holds metadata information, such as storage locations of original and reformatted data. The BDL's performance was evaluated using a dataset of 20,000 knee and hand X-ray images of varying formats (DICOM, PNG, JPEG, BMP, and compressed TIFF) and sizes (0.3 MB to 66.7 MB). The dataset represents convenience samples of X-ray images downloaded from the ReadMyXray website (https://readmyxray.curtin.edu.au). Images were uploaded into BDL and automatically converted to a standardized TIFF format. TB regions of interest (ROIs) were selected on the standardized images, and their catalog was automatically constructed. Fractal signature texture parameters were calculated for the ROIs and stored in Microsoft Excel files, which were automatically transformed into CSV files and cataloged.
Results:
The BDL efficiently transforms X-ray images and catalogues TB texture regions and parameters. For 20, 000 images, BDL achieved approximately tenfold reduction in processing time compared with the workstation (8 minutes versus 82 minutes). The data catalog successfully stored metadata including image ID, timestamps, resolutions, image sizes, and ROI orientation. Schema of data in the Excel files was correctly detected and the corresponding CSV file was generated.
Conclusion:
The BDL is a scalable and flexible data storage platform for X-ray and TB bone texture images and parameters. This platform provides a foundation for future OA research infrastructure and supports the development of reliable, secure, and collaborative systems for the detection and prediction of OA from radiographs.
