Can DXA image-based deep learning model predict the anisotropic elastic behavior of trabecular bone?

Data-Driven Based Characterization of Anisotropic Mechanical Properties for Cancellous Bone From Low-Resolution CT Images

OBJECTIVES

Exploring the anisotropic mechanical behavior of cancellous bone is crucial for in-vivo bone biomechanical analysis. However, it is challenging to characterize anisotropic mechanical behaviors under low-resolution (LR) clinical CT images due to a lack of microstructural information. The data-driven method proposed in this article accurately characterizes the anisotropic mechanical properties of cancellous bone from LR clinical CT images.

METHODS

The trabecular bone cubes of sheep are used to obtain a high-resolution (HR) micro-CT and an LR clinical CT image dataset. First, an auto-encoder model is trained using HR image data. Microstructural features are extracted by the encoder. A fast super-resolution (FSR) model is trained to map LR bone cubes to the features extracted from corresponding HR samples. The pretrained FSR model is used to convert LR clinical CT images to encoded microstructural features. The features are later used to predict target histomorphological parameters, anisotropic elastic tensors, and fabric tensors based on a fully connected neural network.

RESULTS

The data-driven model accurately predicts the elastic tensor and fabric tensor of trabecular bones with LR CT images with 0.6 mm/pixel spatial resolution. It was verified that LR clinical CT images could generate microstructural information using a generative deep-learning model and an up-sampling operation.

SIGNIFICANCE

This study proves that clinical medical images of cancellous bone can be used for analysis of complex mechanical properties using a data-driven method, which is useful for real-time bone defect diagnosis and personalized bone prosthesis design in clinical application.

Inverse design of anisotropic bone scaffold based on machine learning and regenerative genetic algorithm

Introduction: Triply periodic minimal surface (TPMS) is widely used in the design of bone scaffolds due to its structural advantages. However, the current approach to designing bone scaffolds using TPMS structures is limited to a forward process from microstructure to mechanical properties. Developing an inverse bone scaffold design method based on the mechanical properties of bone structures is crucial. Methods: Using the machine learning and genetic algorithm, a new inverse design model was proposed in this research. The anisotropy of bone was matched by changing the number of cells in different directions. The finite element (FE) method was used to calculate the TPMS configuration and generate a back propagation neural network (BPNN) data set. Neural networks were used to establish the relationship between microstructural parameters and the elastic matrix of bone. This relationship was then used with regenerative genetic algorithm (RGA) in inverse design. Results: The accuracy of the BPNN-RGA model was confirmed by comparing the elasticity matrix of the inverse-designed structure with that of the actual bone. The results indicated that the average error was below 3.00% for three mechanical performance parameters as design targets, and approximately 5.00% for six design targets. Discussion: The present study demonstrated the potential of combining machine learning with traditional optimization method to inversely design anisotropic TPMS bone scaffolds with target mechanical properties. The BPNN-RGA model achieves higher design efficiency, compared to traditional optimization methods. The entire design process is easily controlled.

Experimentally characterizing the spatially varying anisotropic mechanical property of cancellous bone via a Bayesian calibration method

Traditional experimental tests for characterizing bone's mechanical properties usually hypothesize a uniaxial stress condition without quantitatively evaluating the influence of spatially varying principal material orientations, which cannot accurately predict the mechanical properties distribution of bones in vivo environment. In this study, a Bayesian calibrating procedure was developed using quantified multiaxial stress to investigate cancellous bone's local anisotropic elastic performance around joints as the spatial variation of main bearing orientations. First, the bone cube specimens from the distal femur of sheep are prepared using traditional anatomical axes. The multiaxial stress state of each bone specimen is calibrated using the actual principal material orientations derived from fabric tensor at different anatomical locations. Based on the calibrated multiaxial stress state, the process of identifying mechanical properties is described as an inverse problem. Then, a Bayesian calibration procedure based on a surrogate constitutive model was developed via multiaxial stress correction to identify the anisotropic material parameters. Finally, a comparison between the experiment and simulation results is discussed by applying the optimal model parameters obtained from the Bayesian probability distribution. Compared to traditional uniaxial methods, our results prove that the calibration based on the spatial variation of the main bearing orientations can significantly improve the accuracy of characterizing regional anisotropic mechanical responses. Moreover, we determine that the actual mechanical property distribution is influenced by complicated mechanical stimulation. This study provides a novel method to evaluate the spatially varying mechanical properties of bone tissues enduring complex mechanical loading accurately and effectively. It is expected to provide more realistic mechanical design targets in vivo for a personalized artificial bone prosthesis in clinical treatment.

Designing anisotropic porous bone scaffolds using a self-learning convolutional neural network model

The design of bionic bone scaffolds to mimic the behaviors of native bone tissue is crucial in clinical application, but such design is very challenging due to the complex behaviors of native bone tissues. In the present study, bionic bone scaffolds with the anisotropic mechanical properties similar to those of native bone tissues were successfully designed using a novel self-learning convolutional neural network (CNN) framework. The anisotropic mechanical property of bone was first calculated from the CT images of bone tissues. The CNN model constructed was trained and validated using the predictions from the heterogonous finite element (FE) models. The CNN model was then used to design the scaffold with the elasticity matrix matched to that of the replaced bone tissues. For the comparison, the bone scaffold was also designed using the conventional method. The results showed that the mechanical properties of scaffolds designed using the CNN model are closer to those of native bone tissues. In conclusion, the self-learning CNN framework can be used to design the anisotropic bone scaffolds and has a great potential in the clinical application.

Influence of the parameters of the convolutional neural network model in predicting the effective compressive modulus of porous structure

In recent years, the convolutional neural network (CNN) technique has emerged as an efficient new method for designing porous structure, but a CNN model generally contains a large number of parameters, each of which could influence the predictive ability of the CNN model. Furthermore, there is no consensus on the setting of each parameter in the CNN model. Therefore, the present study aimed to investigate the sensitivity of the parameters in the CNN model for the prediction of the mechanical property of porous structures. 10,500 samples of porous structure were randomly generated, and their effective compressive moduli obtained from finite element analysis were used as the ground truths to construct and train a CNN model. 8,000 of the samples were used to train the CNN model, 2000 samples were used for the cross-validation of the CNN model and the remaining 500 new structures, which did not participate in the CNN training process, were used to test the predictive power of the CNN model. The sensitivity of the number of convolutional layers, the number of convolution kernels, the number of pooling layers, the number of fully connected layers and the optimizer in the CNN model were then investigated. The results showed that the optimizer has the largest influence on the training speed, while the fully connected layer has the least impact on the training speed. Additionally, the pooling layer has the largest impact on the predictive ability while the optimizer has the least impact on the predictive ability. In conclusion, the parameters of the CNN model play an important role in the performance of the CNN model and the parameter sensitivity analysis can help optimize the CNN model to increase the computational efficiency.