CV

M.S.

ShanghaiTech University, Shanghai, China

• Computer Science and Technology
• GPA 3.63/4.3 (Major 3.73/4)
• Advisor Yuyao Zhang
• Core Course
  • Deep Learning, Medical Image Processing and Analysis, Digital Image Processing

Bachelor

ShanghaiTech University, Shanghai, China

• Computer Science and Technology
• GPA 3.39/4.3

Longitudinal Brain Atlases Construction via implicit neural representation.

• Alleviated the temporal inconsistency issue caused by independently averaging brain images at discrete time points in existing longitudinal atlas construction methods.
• Formulated the time inconsistency issue as a 4D image denoising task, and used implicit neural representation to construct continuous and noise-free longitudinal brain atlases.
• Improved temporal consistency while maintaining accurate representation of brain structures on two modalities of brain atlases (QSM and fetus atlases).
• Generated finer 4D atlases with higher temporal resolution (e.g., 0.5-week interval).

Robust Self-supervised 3D Fetal Brain MRI Reconstruction.

• Tackled the issue of corrupted reconstruction of the fetal brain caused by slice misalignment and blurring of the brain anatomy due to severe motion during MR data collection.
• Combined the MRI acquisition model and a Deep Decoder network to effectively reduce the image artifacts resulting from slice misalignment and motion.
• Outperformed SOTA methods (SVRTK, NiftyMIC, and SSGNN) in five metrics, including a 24% improvement in PSNR, on both simulated and clinical data.

Self-supervised Slice-to-Volume Registration for Severe Fetal Motion.

• Focused on the challenging task of Slice-to-Volume Registration (SVR), which aligns the slices with severe inter-slice motion to the correct position in the volume.
• Incorporated the MRI acquisition model into the SVR network to accurately predict the spatial transformation matrix aligning 2D slices to 3D volumes.
• Achieved SOTA accuracy of SVR, and improved the performance of downstream fatal MRI reconstruction (based on NeSVoR, etc.) on both simulated and clinical data.

Longitudinal Infant Brain MRI Segmentation.

• Investigated different segmentation model designs and extended DenseNet to 3D volumetric data to perform 3D volumetric segmentation.
• Achieved dice similarity coefficient (DSC) of 90.325% of three brain tissues across five time points.

Research Intern

NSB BLCTO Technology Vision&Architecture, Bell Labs

• Proposed a behavior recognition method based on meta-learning using WiFi channel state information(CSI).
• Adapted to new environments rapidly, as demonstrated by its superior performance on two public datasets and real-world data compared to traditional supervised learning methods.