Transforming Semiautomatic Patient-Specific Model Building Workflows Into Autonomous Imaging-Through-Analysis Tools

Abstract: 

Transforming Semiautomatic Patient-Specific Model Building Workflows Into Autonomous Imaging-Through-Analysis Tools

While finite element simulations are well established in medical research, their potential for everyday use in clinical practice still lies largely idle. One of the major reasons is the process of building patient-specific computational models, including transfer of diagnostic imaging data to explicit surfaces, geometry cleanup, and boundary-fitted mesh generation. Although powerful software solutions to streamline this process are available today, many simulation workflows involving complex physiological geometries still require the intervention of specially trained analysts. The associated cost and time implications do not fit into many clinical processes characterized by tight budgets and urgent decision-making.

This research program envisions seamless imaging-through-analysis procedures that enable the full automation of predictive biomedical simulations from reading in diagnostic imaging data to the output of clinically relevant simulation results. The goal of the project is to initiate research activities that provide a pathway to a closer integration of predictive simulation in clinical decision-making and help unlock its potential in clinical routines, with transformative impact on improving public healthcare. Clinical applications the group is currently working on include liver perfusion, heart valve hemodynamics, and multiscale bone fracture and osteoporosis diagnosis. They have also successfully started to translate some imaging-through-analysis procedures to applications in materials science, where imaging technologies play a significant role, e.g., to characterize complex microstructure in the degradation analysis of lithium-ion batteries during cyclic charging/discharging.

The group implements practicable cyberinfrastructure frameworks for fully automated simulation workflows. They strive to minimize the implementation effort by consistently leveraging and integrating existing software, in particular open-source tools for image processing (e.g., ITK), adaptive mesh generation (e.g., Netgen), parallelization (Trilinos, PetSc), and visualization (ParaView). The basic entity of their pipeline from geometric parameterization to multiphysics simulation to visualization is an adaptive finite element mesh, which facilitates interoperability between single components. They target both medium-scale computing clusters affordable to hospitals and clinics for simulations in clinical practice, as well as high performance computing environments for extreme-scale research simulations. The heterogeneous features of Mesabi make it the ideal environment to test the potential of the group's software frameworks at different scales (from medium-scale computations that could potentially be carried out on site in a hospital to extreme scale computations), exploiting the local availability of GPUs at each processing unit for extremely efficient operations.

Return to this PI's main page.

Group name: 
schillin