Linking morphometric variations in human cranial bone to mechanical behavior using machine learning.
Authors
Affiliations (5)
Affiliations (5)
- Department of Mechanical Engineering, The University of Alberta, Edmonton, AB T6G 2R3, Canada. Electronic address: [email protected].
- Department of Mechanical Engineering, The University of Alberta, Edmonton, AB T6G 2R3, Canada.
- Division of Anatomy - Department of Surgery, Faculty of Medicine & Dentistry, The University of Alberta, Edmonton, AB T6G 2R3, Canada.
- Defence Research and Development Canada, Valcartier Research Centre, 2459, Route de la Bravoure, Quebec QB, G3J 1X5, Canada.
- Department of Mechanical Engineering, The University of Alberta, Edmonton, AB T6G 2R3, Canada; School of Dentistry, The University of Alberta, Edmonton, AB T6G 1C9, Canada.
Abstract
With the development of increasingly detailed imaging techniques, there is a need to update the methodology and evaluation criteria for bone analysis to understand the influence of bone microarchitecture on mechanical response. The present study aims to develop a machine learning-based approach to investigate the link between morphology of the human calvarium and its mechanical response under quasi-static uniaxial compression. Micro-computed tomography is used to capture the microstructure at a resolution of 18μm of male (n=5) and female (n=5) formalin-fixed calvarium specimens of the frontal and parietal regions. Image processing-based machine learning methods using convolutional neural networks are developed to isolate and calculate specific morphometric properties, such as porosity, trabecular thickness and trabecular spacing. Then, an ensemble method using a gradient boosted decision tree (XGBoost) is used to predict the mechanical strength based on the morphological results, and found that mean and minimum porosity at diploë are the most relevant factors for the mechanical strength of cranial bones under the studied conditions. Overall, this study provides new tools that can predict the mechanical response of human calvarium a priori. Besides, the quantitative morphology of the human calvarium can be used as input data in finite element models, as well as contributing to efforts in the development of cranial simulant materials.