Research Statement – Sean I Young, PhD

Page 1 Back to the top

Biography. I am Instructor of Radiology at the Martinos Center, Harvard Medical

School (supported by a NIH K99/R00 Career Development Award) and Research

Affiliate with the Computer Science and Artificial Intelligence Lab (CSAIL) at the

Massachusetts Institute of Technology, where I design novel computational imaging

methods for radiology. Previously, I was a Postdoctoral Scholar in the Department

of Electrical Engineering, Stanford University, where I worked on computational

imaging and model compression. I received my PhD in electrical engineering from

the University of New South Wales, Sydney, NSW, Australia. My research expertise

lies in the design of novel methods for computational imaging and, in particular, 3D image reconstruction

and related inverse problems in medical imaging. In 2018, I received the Australian Pattern Recognition

Society (APRS)’s best paper award for my work on “fast optical flow extraction from compressed video”.

Research Statement. I seek to improve healthcare outcomes by making fundamental contributions to the

science of computational radiology. Examples of computational radiology problems I have solved include

accurate MRI of moving subjects from a single MR slice stack; registration of medical images of different

contrast to deci-voxel accuracy; and supervised image reconstruction with few labeled images. Coming from

an imaging and signal processing background, I am also interested in solving more general computational

imaging problems in hopes that the developed techniques will find use in radiology one day. Examples of

computational imaging problems I have solved include 100x faster motion estimation using filtering; non-

line-of-sight surface reconstruction using Cholesky–Wiener filtering; and 30x smaller convolutional neural

networks using transform quantization for real-time imaging. I detail some of my work in chronological

order below and conclude by discussing future directions for my research.

PREVIOUS RESEARCH

100x Faster Motion Estimation (–2019). My computational imaging research began in with my PhD work

on video compression, where I investigated the use of high-order parametric displacement models for more

efficient representation of video. In video compression, one way to reduce data redundancy is to transmit

an earlier video frame, followed by the displacement and difference fields that warp the earlier frame and

synthesize a later frame—the same principles which underlie change detection in medical imaging. My US

patent

filed in 2018 demonstrates the utility of higher-order (generalization of affine) displacement models

for video compression, accompanied by a procedure that can recover the values of displacement parameters

to a high precision.

Concurrently with my work on parametric displacement estimation techniques, I investigated the use of

dense deformation fields to improve the performance of video compression systems. I demonstrated that

deformation estimation and many related inverse problems in imaging and vision can be solved with high-

dimensional Gaussian filtering to obviate the need for slow, iterative optimization procedures. For certain

tasks, this filtering technique accelerated displacement estimation by a factor of 100 compared to iterative

approaches.

My doctoral work resulted in a PhD thesis on non-linear optimization and regularization

techniques for inverse problems in imaging, and four first-author journal publications

4,6,7,8

as well as smaller

conference and workshop contributions.

Fast Non-Line-of-Sight Surface Reconstruction (2019). After my PhD, I moved to Stanford University and

continued to work on efficient methods for computational imaging problems. Collaborating closely with

Gordon Wetzstein, I worked on the non-line-of-sight imaging (“looking around the corner”) problem and

developed an efficient method

that can resolve surfaces of hidden scene objects with an unprecedented

level of detail by recovering both albedo and surface normals, in addition to a 1000-fold speed-up over the

state-of-the-art methods. Similar to my filter-based motion estimation approach, I formulate the solution of

Page 2 Back to the top

the underlying inverse problem as filtering of the observed signal and apply “Cholesky–Wiener filtering” to

evaluate the analytic solution in a non-iterative manner. This work led to a first author publication

in the

IEEE CVPR (oral presentation) among other unpublished contributions.

30x Smaller Convolutional Neural Networks (2021). While at Stanford, I also worked with Bernd Girod

to develop compression techniques for convolutional neural networks

. With the rise of portable medical

imaging devices, neural network compression continues to be an important problem in radiology due to the

demand for real-time image processing on embedded imaging devices. I first developed a theory of rate and

distortion for linear operators, together with decorrelating transforms for provably optimal rate–distortion

performance. The weight tensors were quantized in the transform domain again subject to rate–distortion

constraints, and the resulting network model was optionally fine-tuned, for a 30x reduction in network size

all without loss of accuracy. This work led to one first-author publication in the IEEE Trans Pattern Anal

Mach Intell

among other conference publications.

Few-Shot Supervised Image Reconstruction (2022). In the wake of the COVID-19 pandemic, I became

determined to pursue research that can touch people’s lives. I moved to Boston in late 2020 for medical

imaging research in the labs of Bruce Fischl (Martinos Center, Harvard Medical School) and Polina Golland

(CSAIL, MIT) to work on semi-supervised imaging

, which allows neural networks to be trained on a mix

of labeled and unlabeled images—a scenario that is frequently encountered in medical imaging. I formulate

semi-supervision as the machine learning equivalent of the “regularization by denoising” (RED) formalism

used to solve inverse problems, and alternate denoising and weight update steps for supervision. This work

led to one first author publication in the IEEE Trans Pattern Anal Mach Intell

as well as other workshop

publications with collaborators at the University of Maryland.

Single-Stack Slice-to-Volume Reconstruction (2023). At Harvard Medical School and MIT, I developed

a fully convolutional approach to slice-to-volume reconstruction (SVR), enabling imaging of subjects with

uncontrollable motion (such as in the case of the fetal population) from a single stack of MR slices acquired

using e.g., HASTE or SSFSE sequences.

Previous SVR methods require multiple slice stacks for accurate

reconstruction, precluding their use in applications such as fetal fMRI, where the time-sensitive nature of

the acquisition often prohibits the use of multiple slice stacks. This work led to a first author submission to

the IEEE CVPR (under review)

CURRENT RESEARCH

Universal Image Registration (K99, 2023–). Certain changes in human organs are strongly predictive of

early disease processes (such as cortical atrophy in the case of Alzheimer’s Disease). However, using existing

image processing tools to detect this anatomical change in clinical MR scans taken across time has proven

difficult due to numerous MR effects and subject motion, which can mask the anatomical change of interest

relevant to diagnosis and research. My K99 project aims of the proposed project are to leverage the recent

advances in deep learning to design, develop and evaluate an imaging and analysis framework which can

resolve anatomical changes with high accuracy, aiding in the early diagnosis and intervention before the

onset of dysfunction in the case of neurological diseases, for example.

Research Outlook. My expertise in computational imaging and modeling coupled with a newly discovered

aptitude for deep learning in MRI makes me an extremely suitable candidate for a junior faculty position

in radiology. My mentors from Harvard and MIT have already provided guidance and mentoring in the

fields of neurology and radiology to ensure that appropriate methods are used to address clinically relevant

questions. The NIH K99/R00 award has not only provided me with an opportunity to work with mentors

and solve a problem of utmost importance but has also supported me to grow as an independent researcher

and ultimately raise the next generation of researchers to carry on our collective scientific legacy.

Transitioning to faculty, I will start my own research lab focused on developing novel computational

imaging methods for radiology. I will utilize my experience with a range of computational imaging problems

and a deep understanding of signal processing theory to spearhead the computational imaging and vision

research of my future lab in the R00 phase. I see the increase in demand for engineering academics with

Page 3 Back to the top

medical knowledge as an opportunity for translational research, either in an engineering department with

strong ties to a clinic or a radiology department having strong ties to engineering. Transition to faculty will

also allow me to pursue my passion as an educator to motivate and inspire the next generation of researchers

and pass on to them very carefully distilled knowledge in hopes that they will take our scientific discoveries

and innovations further.

REFERENCES

1. Sean I. Young, David Taubman. Rate-distortion optimized optical flow estimation. Proc. ICIP; 2015.

p. 1677–1681.

2. Sean I. Young, Reji K. Mathew, David Taubman. Optimizing block-coded motion parameters with

block-partition graphs. Proc. ICIP; 2016. p. 2037–2041.

3. Sean I. Young, Xiaoyu Xiu, Yuwen He, Rahul Vanam. Higher-order motion models and graduated

motion parameter estimation for video coding. United States Patent US201762504963P; 2018.

4. Sean I. Young, Aous T. Naman, David Taubman. COGL: Coefficient Graph Laplacians for

Optimized JPEG Image Decoding. IEEE Trans Image Process. 2019 Jan;28(1):343–355.

5. Sean I. Young, Aous T. Naman, Bernd Girod, David Taubman. Solving vision problems via filtering.

Proc. ICCV; 2019. p. 5592–5601.

6. Sean I. Young, Aous T. Naman, David Taubman. Graph Laplacian regularization for robust optical

flow estimation. IEEE Trans Image Process. 2020;29:3970–3983.

7. Sean I. Young, Bernd Girod, David Taubman. Fast optical flow extraction from compressed video.

IEEE Trans Image Process. 2020;29:6409–6421.

8. Sean I. Young, Bernd Girod, David Taubman. Gaussian lifting for fast bilateral and nonlocal means

filtering. IEEE Trans Image Process. 2020;29:6082–6095.

9. Sean I. Young, David B. Lindell, Bernd Girod, David Taubman, Gordon Wetzstein. Non-line-of-

sight surface reconstruction using the directional light-cone transform. Proc. CVPR; 2020. p. 1407–

1416.

10. Sean I. Young, Wang Zhe, David Taubman, Bernd Girod. Transform quantization for CNN

compression. IEEE Trans Pattern Anal Mach Intell. 2021;1–1.

11. Sean I. Young., Adrian V. Dalca, Enzo Ferrante, Polina Golland, Christopher A. Metzler, Bruce

Fischl, and Juan Eugenio Iglesias. Supervision by Denoising. IEEE Trans Pattern Anal Mach Intell.

2023;1–1.

12. Sean I. Young, Yaël Balbastre, Bruce Fischl, Polina Golland, Juan Eugenio Iglesias. Fully

convolutional slice-to-volume reconstruction for single-stack MRI. Proc. CVPR; submitted.