Han Cui Email and Phone Number

Conduct research and develop key innovations in model-based image reconstruction, focusing on correcting Diffusion-Weighted Imaging(DWI) artifacts and improving network efficiency.

Built a UNet-based network with spatial and temporal convolution to reconstruct under-sampled Magnetic Resonance(MR) images acquired by a multi-phase single-shot fast spin echo (SSFSE) PROPELLER pattern.Applied Temporal-Deformable Alignment Network with spatial-temporal convolution attention mechanism to aid the accurate motion correction and clean reconstruction.

Took inspiration from JPEG compression and developed an unsupervised routine for training deep neural networks to predict image quality.Generalized the application of image quality prediction network from RGB images to MR images.Designed versatile image quality-driven metric using quality estimate network to evaluate and improve the performance of image restoration tasks such as JPEG artifact removal and MRI reconstruction.

Reproduced recent work that simplifies the MRI reconstruction using deep networks by training with DICOM data, a magnitude-only data storage format for MRI.Demonstrated that the resulting reconstruction networks are overly optimistic and failed on real-world complex MRI tasks due to the artifacts introduced in the inverse Fourier transform over the magnitude MR images.

Developed interactive application that calculates the levelized cost of energy (LCOE) in Mexico based on geological data provided by the National Renewable Energy Laboratory. Assisted local families in estimating the potential cost savings from installing solar panels for home using data-driven LCOE analysis.

Analyzed energy consumption of California’s disadvantaged communities by utilizing unsupervised clustering with geographic information system (GIS).Visualized geographical and statistical distribution to facilitate resource planning, community development, and poverty alleviation.