|About this Abstract
||2019 TMS Annual Meeting & Exhibition
||Computational Approaches for Big Data, Artificial Intelligence and Uncertainty Quantification in Computational Materials Science
||M-5: Efficacy of a Mathematical Model in Mimicking Trabecular Bone Structures Using Deep Learning Techniques
||Neda Shafiei, Joel Gomez, Edward Guo, Xiaodu Wang
|On-Site Speaker (Planned)
We intended to verify the efficacy of a probabilistic digital model of trabecular bone we developed in mimicking trabecular bone structures, by testing whether the anatomic differences in the microstructures could be detected using deep-learning techniques.
Based on the information of trabecular bones (trabecular size, orientation and spatial distributions) from vertebral body and femoral neck, five mathematical models were generated for each location to mimic the trabecular structures using a MATLAB code. From each model, 255 cross sectional images were obtained. The images were split into two groups (ratio of 8:2) for training and validation of the convolutional-neural-network (CNN), respectively. Afterwards, the efficacy of the CNN model was tested using real bones.
The CNN could recognize anatomic location by > 99% accuracy and <1.0% loss after 60 iterations. Moreover, the trained model could distinguish between anatomic locations of real trabecular bone images (accuracy and loss of 0.99 and 0.005, respectively).
||Planned: Supplemental Proceedings volume