Medical Image Analysis and Spinal Disc Degeneration Estimation
1. Labeling bone joints of hands (Independent work)
Reliable and accurate labeling and segmentation of medical images are expected to benefit clinical decision support systems for diagnosis, surgery planning, and population-based analysis of pathology. However, the high costs of image acquisition and the extensive manual annotation effort limit the availability of training data for many medical image analysis applications. To mitigate this challenge, this work implements the SpatialConfiguration-Net (SCN) inspired by existing literature, for anatomical landmark localization. The SCN architecture addresses the localization task by dividing it into two sub-problems, thereby reducing the need for large training datasets. This simplification is achieved through the multiplication of heatmap predictions from its two components, allowing for end-to-end training. Empirical evaluations on open-source hand radiographs demonstrate that the SCN maintains high performance even with a limited number of training images.
2. In vivo subject-specific estimation of cervical intervertebral disc degeneration
Progressive deterioration of cervical intervertebral discs narrows the spinal canal, leading to chronic spinal cord compression and disability. This study proposes a computational method to estimate the in vivo material degradation of cervical intervertebral discs. The kinematics of the cervical spinal column is modeled using a nonlinear finite element method, incorporating the latest nonlinear material properties from experiments. In vivo intervertebral motion data, acquired from biplane radiography for a given subject, were used to inversely calibrate the disc material properties and validate the spine mechanical model. The calibrated disc properties successfully revealed a reduction in the stiffness of the annulus and nucleus in the patient's cervical spine.
Efficient Solution and Extension of Lattice Discrete Particle Model
1. An efficient static solver for the lattice discrete particle model
The lattice discrete particle model (LDPM) has been proven to be one of the most appealing computational tools to simulate fracture in quasi-brittle materials. Despite tremendous advancements in the definition and implementation of the method, solution strategies are still limited to dynamic algorithms, resulting in prohibitive computational costs and challenges related to solution accuracy for quasi-static conditions. This study presents a novel static solver for LDPM, introducing several fundamental innovations to overcome convergence issues. Additionally, graph coloring of the Jacobian operator, automatic differentiation, and advanced matrix manipulations are utilized to improve computational efficiency by up to 1000 times.
2. Lattice discrete particle modeling for coupled mechanical and mass transport simulation of reinforced concrete
Although the Lattice Discrete Particle Model (LDPM) has excelled in simulating cracking patterns of particle-reinforced materials, its potential for modeling mass transport through cracks and steel-reinforced materials remains underexplored. This study introduces three new elements to the LDPM: mass transport elements through crack channels, steel elements, and steel-to-concrete bond elements. These enhancements enable the model to simulate chloride transport in damaged reinforced concrete. Notably, the bond element efficiently represents the effects of rebar ribs. The integrated model uses a fully dynamic solution strategy: reinforced concrete is solved with a dynamic explicit method and central difference scheme, while mass transport models use a dynamic implicit method with the Crank-Nicolson scheme and Newton-Raphson iterations. Calibration and validation have been conducted with rational accuracy, though the work is ongoing.
Jia, D., Brigham, J. C., & Fascetti, A. (2024). An efficient static solver for the lattice discrete particle model. Computer-Aided Civil and Infrastructure Engineering, 1-21.Jia, D., Zhu, Y., Brigham, J. C., & Fascetti, A. (2024). A novel dual lattice discrete particle model for multiphysics simulation of coupled mechanical and transport behavior in concrete members subjected to long-term loading. In 16th World Congress on Computational Mechanics and 4th Pan American Congress on Computational Mechanics (WCCM-PANACM).
