I am currently a Ph.D candidate in Electrical Engineering at Department of EECS at the University of Michigan. I am co-advised by Prof. Jeffrey A. Fessler from EECS and Prof. Heang-Ping Chan from the Department of Radiology.
I obtained my M.S. degree in Electrical Engineering from University of Michigan. During that time I worked with Prof. Yuni Dewaraja on SPECT imaging. I obtained my Bachelor’s degree in Physics from Peking University, China, where I worked with Prof. Qi Ouyang on reverse engineering of Boolean networks.
Here is a link to my LinkedIn profile.
My current research topics include:
Detector Blur and Correlated Noise Modeling for Digital Breast Tomosynthesis Reconstruction
Jiabei Zheng, Jeffrey A. Fessler, Heang-Ping Chan, IEEE Transactions on Medical Imaging, 2017
[Link] [PDF] [Supplementary Material]
The journal article of our work on modeling detector blur and correlated noise for tomosynthesis. Please contact me if you are interested in more details of the implementation.
Segmented Separable Footprint Projector for Digital Breast Tomosynthesis and its application for Subpixel Reconstruction
Jiabei Zheng, Jeffrey A. Fessler, Heang-Ping Chan, Medical Physics, 2017
We developed the SG projector, by far the fastest and most accurate projector for tomosynthesis applications. The projector is the core component for tomographic reconstructions. Its accuracy affects the quaility of reconstructed images. The speed of the SG projector enables practical application of advanced image reconstruction techniques to reduce X-ray dose in tomosynthesis.
Effects of Detector Blur and Correlated Noise on Digital Breast Tomosynthesis Reconstruction
Jiabei Zheng, Jeffrey A. Fessler, Heang-Ping Chan, Proc. SPIE, Medical Imaging, 2017
[Link] [PDF] [Presentation Slides]
This conference paper is a further step of our previous study presented at the 4th CT Meeting (2016). We improved the regularization method and demonstrated that the effectiveness of our previous reconstruction method requires all model components to be incorporated at the same time.
Digital Breast Tomosynthesis Reconstruction with Detector Blur and Correlated Noise
Jiabei Zheng, Jeffrey A. Fessler, Heang-Ping Chan, Proc. 4th International Conference on Image Formation in X-Ray CT, vol. CT-Meeting 2016, pp. 21-24, 2016
[Link] [PDF] [Presentation Slides]
Model-based image reconstruction (MBIR) refers to modeling the physics (e.g., X-ray spectrum, beaming-hardening, detector spatial response, focal spot size) of medical scanners in the iterative reconstruction. We conducted the first study to apply the idea of MBIR to tomosynthesis. The model in the paper considers detector blur and correlated noise. The effectiveness of the method indicates a great potential of improving tomosynthesis image quality with MBIR.
Digital Breast Tomosynthesis Reconstruction using Spatially Weighted Non-convex Regularization
Jiabei Zheng, Jeffrey A. Fessler, Heang-Ping Chan, Proc. SPIE, Medical Imaging, 2016
[Link] [PDF] [Poster]
Microcalcification cluster is an early sign of breast cancer. We proposed the spatially weighted non-convex regularization method for digital breast tomosynthesis and solved the corresponding optimization problem with a majorize-minimize separable quadratic surrogate algorithm. The average contrast-to-noise ratio of microcalcifications is enhanced by over 200% compared with simultaneous algebraic reconstruction technique (SART).
Optimization and Application of Non-local Means (NLM) Filtering of SPECT/CT
Jiabei Zheng, Se Young Chun, Jeffrey A. Fessler, and Yuni K. Dewaraja, Oral Presentation at SNMMI 2014 Annual Meeting
We designed an adaptive strategy to adjust the parameters of the non-local means filtering in its application in SPECT-CT scan for the best estimation of absorbed radiation dose in Yttrium-90 microsphere therapy. The effectiveness of the method is supported by both simulated and experimental studies.
Subspace Clustering with Missing Entries
Lianli Liu, Dejiao Zhang, Jiabei Zheng, a project of Machine Learning (EECS 545)
We derived and implemented the expectation-maximization algorithm to achieve the high-rank matrix completion assuming columns of the matrix belong to a union of subspaces. The algorithm is then applied to recover missing pixel values of images.
Image Deblurring with Blurred/Noisy Image Pairs
Huichao Ma, Buping Wang, Jiabei Zheng, Menglian Zhou, a project of Image Processing (EECS 556)
We implemented a method to use a pair of blurred/noisy image to generate a high-quality denoised image. We first estimated the blur kernel by solving a regularized least-square optimization. Then the gain-controlled Richardson-Lucy algorithm is implemented for deblurring with less ringing artifacts. The image details are then enhanced by extracting a detail layer with bilateral filtering.