Xiang "Shaun" Li, Ph.D.

Instructor in Investigation
Gordon Center for Medical Imaging, Research Institute
Instructor in Radiology
Harvard Medical School
Research Staff
Radiology, Massachusetts General Hospital
PhD University of Georgia 2016
Bachelor of Engineering Shanghai Jiaotong University 2006
artificial intelligence; big data; brain computer interface; brain mri images; cardiac imaging; ct; functional neuroimaging; machine learning; medical image analysis; multimodal neuroimaging; neuroimaging As an instructor at MGH/HMS, my research includes on two main fields which are interlinked: 1) development of new machine learning and artificial intelligence methodologies, mainly for the purpose of medical image analysis, and 2) applying these methods for solving practical clinical/scientific problems. With more than 10 years of experience in algorithm development/implementation and a doctoral degree in computer science, I have been working on new tools to support efficient large-scale data analytics, including matrix decomposition under different regularizations, new network architectures for deep learning models and distributed computing platforms. My work presented in 2016 ACM SIGKDD conference, "Scalable Fast Rank-1 Dictionary Learning for fMRI Big Data Analysis", is the first system that performs computation of terabytes-level neuroimaging data without relying on supercomputers. My current work on methodology development focuses on building a unified framework to integrate the analysis across multi-modal, multi-scale and multi-view images, as well as co-analysis of those images in a single model. This framework is particularly useful for medical image analysis, as usually there exists multiple scans of the same patient under different protocols (e.g. modalities, scanning parameters, etc.) and it is often desired for analyzing them together.

From the application perspective, I have been working on the processing and analysis of medical images in two main categories: lung CT and brain MR. By collecting and analyzing CT images >300 patients within the institution, I have developed a deep learning-enabled system for pneumothorax detection (i.e. screening) with high sensitivity and very fast running speed. The system is among the four finalists of the 2018 NVIDIA Global Impact Award, for its potential impact on application of AI in healthcare. Currently I’m working on the alpha testing for the system’s integration into the clinical workflow of MGH, as well as extending its applicability to other critical conditions such as pulmonary embolism. On the other hand, analysis of brain MR images especially functional MRI data is a continuation of my PhD research. Being among the first researchers who utilized mathematic models to characterize brain functional dynamics in both normal and abnormal populations (since 2011), my works on diagnosis of post-traumatic stress disorder (PTSD), attention deficit hyperactivity disorder (ADHD), and mild cognitive impairment (MCI) are the pioneers in this field and inspired many later works. My current project on functional brain modeling involves using dynamic graph representation combined with geometric deep learning to achieve holistic and efficient characterization of functional brain states.
Center for Advanced Medical Computing and Analysis Publications
New Chardon #25
25 New Chardon Street
Boston, MA 02114-2696