Xiao Wang

High Performance MBIR (parallel-beam)

This is super fast code for parallel beam CT reconstruction on single node, multi-core CPUs. The main idea to achieve fast computations and convergence is super-voxels. Experiments show that super-voxels can accelerate MBIR CT reconstructions by more than 100 times, compared to the state-of-the-art implementations. In addition, this code implements a plug-and-play framework that allows users to use advanced prior models, such as a Non-Local Mean denoiser, and a convlutional neural network prior.

Here is the github repository for the super-voxel code, written in C programming language:

super-voxel code

Here is a python wrapper for the above code if you want to use python:

python_wrapper

If you do use the code and demos, please cite the following papers:

[1] Xiao Wang, Amit Sabne, Sherman Kisner, Anand Raghunathan, Charles Bouman, and Samuel Midkiff, "High Performance Model Based Image Reconstruction," 21st ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP '16), March 12-16, 2016.

[2] Xiao Wang, Amit Sabne, Putt Sakdhnagool, Sherman J. Kisner, Charles A. Bouman, and Samuel P. Midkiff, "Massively Parallel 3D Image Reconstruction," Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC'17), November 13-16, 2017. (One of three finalists for 2017 ACM Gordon Bell Prize.)

Baseline MBIR Open Source

My collaborators and I developed a baseline Model Based Iterative Reconstruction (MBIR) software for parallel-beam CT reconstruction. The github link to the open-source code can be found at:

https://github.com/cabouman/OpenMBIR

Feel free to use the code for education and research purpose. You are also welcome to contribute to the ongoing MBIR software by creating a new branch on github.

Cloud-Based Inversion Engine

Over the years, my fellow co-workers and I develped a cloud server system for the high performance CT Model-Based Iterative Reconstruction (MBIR). This cloud server, named the " Inversion Engine", is developed based on the Super-Voxel research idea. By using the Inversion Engine, users can upload their own dataset and get image reconstruction result in real-time. At the same time, users can treat the image reconstruction system as a black box and don't need to know how reconstruction algorithm works.

contact me if you want to use the cloud-based inversion engine.

Welcome!

My name is Xiao Wang (王蕭). I recently started a new job at Oak Ridge National Laboratory as research scientist in machine learning and medical imaging. My new contact email is wangx2@ornl.gov

Previously, I was a postdoc researcher at Harvard Medical School and Boston Children's Hospital. working with Professor Simon Warfield from 2017 to 2021.

I was a graduate student in Prof. Charles Bouman and Samuel Midkiff's research group at ECE department, Purdue University, and I received my PhD degree there in 2017. Before that, I got two bachelor's degrees from St. John's University, one in Mathematics and one in Computer Science.

My research interests include: High Performance Computing; Image Processing; Medical Physics; Machine Learning; and Numerical Methods.

My research direction focuses on applying the domain knowledge in the above mentioned fields to computed tomographic (CT) , Positron Emission Tomography (PET) and Magnetic Resonance (MR) reconstruction, so that the reconstruction algorithm tightly integrates with CT or PET or MR detector sensor design and data acquisition physics, making it possible to observe phenomena at a super-resolution previously difficult or impossible to measure with traditional approaches.

Contact:
email: wangx2@ornl.gov

Xiao Wang

Postdoctoral Fellow

Softwares

High Performance MBIR (parallel-beam)

Baseline MBIR Open Source

Cloud-Based Inversion Engine