About me

I am currently a postdoc fellow working with Dr. Kilian Pohl and Dr. Edith V. Sullivan at the Department of Psychiatry and Behavioral Sciences, Stanford University.

I finished my Ph.D. at the Department of Computer Science, University of North Carolina at Chapel Hill. My principal advisor was Prof. Stephen Pizer. I was co-advised by Prof. Marc Niethammer and Prof. Ron Alterovitz. Before joining UNC, I recieved my B.S. degree from Shanghai Jiao Tong University.

I am generally interested in using image analysis and machine learning techniques to improve detection, diagnosis and treatment of diseases. My research interests lie in the areas of non-linear statistics, machine learning and neuroscience.

Contact: qingyuz -at- stanford dot edu
[Google Scholar][CV][LinkedIn]

Recent Work

Longitudinal Analysis of Brain Functional Connectivity

Longitudinal neuroimaging studies have become increasingly prevelant these days. Longitudinal analysis of functional connectivity, however, still relies on cross-sectional procedures (computational methods), which neglect intra-subject dependencies of longitudinal fMRI data. We are developing novel methods for estimating subject-specific intrinsic functional networks (resting-state functional networks) that reflect biologically plausible longitudinal effects. Longitudinal analysis on the measures produced by these methods are shown to have improved statistical power.

Joint Classification/Confounding-factor-removal

Alcohol use effects on adolescent brain development revealed by simultaneously removing confounding factors, identifying morphometric patterns, and classifying individuals. [Open Access]

Group Analysis on a Riemannian Manifold: Application to Functional Connectivity Analysis

Functional connectivity (rs-fMRI) matrices live on the manifold of positive-definite cone. A subject-specific longitudinal trajectory of functional connectivity then defines a tangent vector. We demonstrate the intrinsic ambiguity in performing group analysis on these vectors in the context of functional connectivity analysis. Several methods based on parallel-transport, Lie group action, and latent p-value theory, are proposed to tackle the issue [pdf]. Toy example source code [github]

Joint Data-Disparity/Registration Estimation via Monte-Carlo EM

Surface disparity situations, such as missing data, topology change and resolution difference, can severely impact registration accuracy. It is desirable to identify incompatible regions between surfaces and to exclude those regions in the registration process. We developed two variants of the EM algorithm, namely Monte-Carlo EM and hard EM, to jointly estimate disparity and registration. [pdf]

Anisotropic Stiffness Learning via Maximum-a-posteriori

Estimating material elastic properties has long been a challenging problem, especially when force measurements are unavailable. A statistical framework provides an alternative for stiffness estimation from a probabilistic point of view. We investigate a learning model that can estimate inhomogeneous and anisotropic stiffness paramters of the material from a set of known deformations. This could potentially open up a new perspective to shape analysis. [pdf]

Geometry Fusion of Multiple Single-frame 3D Reconstructions from an Endoscopic Video

Due to the constant motion of human tissues during endoscopy, we have to handle non-rigid deformations across all the frames in the task of endoscopic 3D reconstruction. One way to achieve this is to fuse all the single-frame reconstructions into a unified surface. We propose a template-free groupwise surface registration method that can effectively handle partially overlapping data, missing surface patches and topology change. [pdf]

Surface Registration for CT/Endoscope Fusion

The clinical problem we want to tackle here is to transfer the tumor information from a 2D endoscopic movie frame into the 3D CT space for radiation treatment planning. The solution is to first reconstrut a 3D surface model from the video and then register that surface to the CT image. We investigate both deformation-based [pdf] and spectral-graph-matching-based [pdf] methods for this registration problem.

Local Metric Learning in 2D/3D Registration for Abdomen Radiation Treatment Planning

2D/3D registration is often used in Image-Guided Radiation Therapy (IGRT) for tracking target motion during treatment delivery. A challenge in disease sites subject to respiratory motion is that organ deformation may occur, thus requiring incorporation of deformation in the registration process. An improved metric-learning algorithm is designed for this purpose. We are the first ones to study this problem in the abdomen. [pdf]