CaDS Seminar 2023 - August 15

Xin Liu (SDL Fluids & Solids Engineering)

Optimization of Super-Resolution pipeline on CT images for increasing the accuracy of respiratory flow simulations

Abstract:

In this seminar, I will present an optimized super resolution processing pipeline to be integrated into the automatic workflow for rhinology using CT data. The network was based on U-net with residual learning, in which optimized hyperparameters were determined by Bayesian optimization and asynchronous successive-halving algorithm in DeepHyper package. A data pruning method was used together with image tiling to further increase accuracy and convergence speed of the network by biasing SGD towards important samples after a short warm-up phase. Increased accuracy of physical property simulations is shown at in- and outlet area and centerline of airway. The current study can help to increase the number of CT recordings that are available for flow simulations, and therefore improve CFD-based diagnoses and treatments of pathologies in the human respiratory system.