Leveraging artificial neural networks to mesh frog gastrocnemius muscle from digital photography

There has been a lot of development in realistic muscle modeling based on finite elements in the last decade. However, one of the challenges in this area remains to be custom or specimen-specific meshing of the relevant muscle, mainly due to the scarcity of the DT-MRI infrastructure and expertise. The purpose of this work is to capitalize on the Bayesian regularization backpropagation based artificial neural networks, to transform digital photographic imagery into a finite element mesh, and the accompanying internal fiber orientation data. A gastrocnemius muscle was extracted from a frog and utilized to conduct the proposed work. Due to the highly nonlinear nature of the resulting finite element model, from both metric and material considerations, as well as the limited suitability of available elements for meshing in this case, a custom hexahedral-type mesh topology was selected for the meshing procedure. Results indicate a very good agreement between the geometries of the sample muscle and its mesh. Furthermore, fiber orientations were approximated as following the fusiform geometry of the muscle. The proposed framework can be used to overcome the pre-processing requirements of subject-specific muscles, including hexahedral-type meshing and extraction of internal fiber orientation data.