Data-driven image mechanics (D2IM): A deep learning approach to predict displacement and strain fields from undeformed X-ray tomography images – Evaluation of bone mechanics

Abstract

The recent advent of deep learning (DL) has enabled data-driven models to pave the way for the full exploitation of rich image datasets from which physics can be learnt. Here we propose a novel data-driven image mechanics (D²IM) approach that learns from digital volume correlation (DVC) displacement fields of vertebrae, predicting displacement and strain fields from undeformed X-ray computed tomography (XCT) images. D²IM successfully predicted the displacements in all directions, particularly in the cranio-caudal direction of the vertebra, where high correlation (R²=0.94) and generally minimal errors were obtained compared to the measured displacements. The predicted axial strain field in the cranio-caudal direction of the vertebra was also consistent in distribution with the measured one, displaying generally reduced errors in the regions within the vertebral body. The application of D²IM to lower resolution imaging in initial testing provides promising results indicating the future viability of integrating this technology into a clinical setting. This is the first study using experimental full-field measurements on bone structures from DVC to inform DL-based models such as D²IM, which represents a major contribution in the prediction of displacement and strain fields based only on the greyscale content of undeformed XCT images. In future, D²IM will incorporate a range of biological structures and loading scenarios for accurate prediction of physical fields, aiming at clinical translation for improved diagnostics.

Data Availability

Code for preparing dataset, training D²IM model and visualising/analysing results has been hosted on GitHub: https://github.com/PeterSoar/D2IM_Prototype

The dataset used for this study can be found on Figshare: https://doi.org/10.6084/m9.figshare.25404220.v1