Efrat Shimron
Postdoctoral Fellow
Department of Electrical Engineering and Computer Sciences (EECS)
UC Berkeley
About
New! I'm currenlty joining the Technion - Israel Institute of Technology as an assistant professor, with dual affiliation to the Electrical & Computer Engineering (ECE) department & Biomedical Engineering departments. I'm excited to open my medical AI lab at the Technion. My research focuses on developing computational techniques for solving inverse problems in medical imaging, primarily for Magnetic Resonance Imaging (MRI). It combines computer vision, AI, physics, and medical sciences.
I was previously a postdoc in the Department of Electrical Engineering and Computer Sciences (EECS) at UC Berkeley, where I worked with Prof. Michael (Miki) Lustig (UC Berkeley) and Prof. Shreyas Vasanawala (Stanford) on developing computational techniques for dynamic (temporal) MRI and investigating AI bias. I also have an ongoing collaboration with the group of Prof. Matthew Rosen (Harvard) regarding the emerging technology of low-field MRI.
My work was recently published in PNAS, highlighted in a PNAS commentary article, highlighted in the NIBIB New Horizons plenary talk at the ISMRM'22 conference, and covered in the Berkeley News, Berkeley Engineering Magazine, and UT Austin News.
I recently received several career awards: Rising Stars in Electrical Engineering and Computer Sciences (2023), Outstanding Emerging Investigator, and the Weizmann Institute Women’s Postdoctoral Career Development Award in Science (2022-2024).
I am an advocate of reproducible research - all my papers have open-source code (links below). I was also the 2021-2023 Trainee Representative of the ISMRM Reproducible Research Study Group (RRSG) Governing Committee.
For students/postdocs: I have open positions for all levels (senior undergrads/MSc/PhD/postdoc). If you're interested in joining my team, please reach out.
Research
This work was published in PNAS and highlighted in a PNAS commentary article.
It was also covered in the Berkeley News, Berkeley Engineering Magazine, and UT Austin News.
It also received a Magna cum Laude Award at the ISMRM 2021 Conference.
Here is the git repo.
20-min talk from the MRI Together workshop (2021)
5-min talk from ISMRM'21 conference (this link requires registration to conference).
BladeNet: Dynamic Abdominal MRI using PROPELLER and Deep Learning
I gave oral presentations about this work in the ISMRM conference in London (May 2022) & the Sedona Workshop on Data Sampling and Image Reconstruction (Jan 2023).
This work introduces a new acquisition-reconstruction framework for rapid dynamic (temporal) abdominal MRI with high spatio-temporal resolution and built-in motion correction.
ISMRM 2022 abstract 0684 - here (requires access).
Sedona 2023 abstract - here (requires access).
K-band: a novel strategy for fast MRI
New! The paper is on arXiv: https://arxiv.org/abs/2308.02958
This work was accepted for oral presentations in the 2023 Sedona conference (abstract here - this requires access) and the 2023 ISMRM conference.
A new approach for rapid data acquisition and self-supervised training of MRI reconstruction algorithms w/o ground truth data.
Theoratical convergence guarantees
This work was done by two wonderful students that I have mentored, Frederic Wang & Han Qi (co-first).
TED - Temporal Differences Compressed Sensing for fast dynamic MRI
TED is a reconstruction method that I developed for dynamic (temporal) MR imaging.
Here, TED was developed for temperature monitoring, which is required for real-time MR monitoring of high-intensity focused ultrasound. Note that TED is more general and can be suitable for other dynaic MRI applications.
Synthesizing Complex Multi-coil MRI data from Magnitude-only Images.
New paper! Here.
DL reconstruction techniques are data-hungry, but raw MRI databases are scarce. We propose a method to leverage the huge amounts of magnitude-only data stored in hospitals.
This GAN-based method enables synthesizing complex-valued multi-coil data from magnitude data. The results indicate that neural nets trained on the synthesized data produce competitive state-of-the-art results.
This work is part of my collaboration with the group of Prof. Peder Larson from UCSF.
Accelerating Ultra-Low Field MRI with Compressed Sensing
ISMRM 2022 abstract #88.
This work is part of my collaboration with Dr. David Waddington (University of Sydney) and Prof. Matthew Rosen (The Martinos center, Harvard)
Rigorous Uncertainty estimation for MRI Reconstruction
ISMRM 2022 abstract #749.
Presented in an oral presentation at the ISMRM conference in London (May 2022).
This work is led by Ke Wang (I'm a co-author).
CORE-PI
CORE-PI is a method that I developed in my PhD studies. It is a parallel imaging MRI reconstruction method, suitable for 2D Cartesian scans. The main advantage of CORE-PI is that it is parameter-free, i.e. there's no to calibrate any hyperparamer in order to use it.
Check out the code.
CORE-Deblur
CORE-Deblur is another method that I developed in my PhD studies. It is a Parallel Imaging Compressed Sensing (PI-CS) method.
Our paper introduced the concept of MR image reconstruction by deblurring using Compressed Sensing (CS), and showed that CORE-Deblur can expedite CS computations: the number of required iterations is reduced by a factor of 10.
News
Upcoming talks:
I'm honored to be a keynote speaker in Israel's Machine Vision Conference (IMVC'23).
I will talk about our k-band method in the IEEE Asilomar conference on signals, systems and computers (October 2023, CA).
Our k-band paper is out: https://arxiv.org/abs/2308.02958
I recently gave a talk at the Martinos center (Harvard/MGH). The recording is available on youtube, here.
Recent papers:
Efrat Shimron, Or Perlman "AI in MRI: Computational Frameworks for a Faster, Optimized, and Automated Imaging Workflow". This is an editorial paper for the special issue that we co-edited, titled "AI in MRI: Frontiers and applications", which includes 17 papers that showcase the state-of-the-art in the field.
Nikhil Deveshwar, Abhejit Rajagopal, Sule Sahin, Efrat Shimron, Peder E.Z. Larson
"Synthesizing Complex Multicoil MRI Data from Magnitude-only Images" Bioengineering journal
ISMRM 2023 (Toronto) :
Oral talk - "K-band: A self-supervised strategy for training deep-learning MRI reconstruction networks using only limited-resolution data". This is work that I supervised, the two co-first-authors are my students. I'm very proud of our team! Tuesday, June 6, 08:15, in the session "ML/AI for Acquisition & Reconstruction ".
Invited educational talk - "Fundamentals of deep learning". I'll give this talk on Sunday, June 4, in the 13:00-17:00 course.
Oral talk - Waddington, Shimron, Shan, Koonjoo, Shen, Rosen. "Accelerated Imaging at Ultra-low Magnetic Fields: A comparative study of traditional and neural-network-based reconstruction approaches". This is part of my collaboration with David Waddington and the group of Matthew Rosen from MGH.
I'll moderate these sessions:
“ML/AI Emerging Algorithms”. Monday, June 5, 13:45-15:45, room 714A/B.
"I got the grant, now what?" Sunrise session, Wed, June 7, 07:00-08:00, room 717A/B.
Two oral talks in the ISMRM Sedona Workshop on Data Sampling and Image Reconstruction (Arizona, 2023):
Shimron et al., "BladeNet: an acquisition-reconstruction framework for free-breathing dynamic MRI"
Wang, Qi, De Goyeneche, Lustig, Shimron, "K-band: Training self-supervised reconstruction networks using limited-resolution data"
I recently received two career awards (August 2022):
Rising Stars in Electrical Engineering and Computer Sciences (EECS) (2023). The Rising Stars annual event was launched at MIT in 2012, and since then every year it is hosted by a different university. This year it is hosted by UT Austin.
Women’s Postdoctoral Career Development Award in Science (national Israeli fellowship) (2022-2024), given by the Weizmann Institute of Science.
I'm co-editing a special issue titled "AI in MRI: Frontiers and applications" for the Bioengineering journal, together with Dr. Or Perlman. This special issue already includes 17 papers that showcase the state-of-the-art in AI for MRI!
I co-organized a session with Prof. Florian Knoll for the upcoming in-person BASP Frontiers conference (Switzerland, February 2023). Session title: "Potential Pitfalls of Deep Learning in Medical Image Reconstruction".
April 2022: My work was highlighted in a PNAS commentary article.
March 2022: my Data Crimes paper was published in PNAS and covered in the Berkeley News, Berkeley Engineering Magazine and UT Austin News (March 2022).
Collaborations
Prof. Jon Tamir, Electrical and Computer Engineering Department (ECE), UT Austin.
Prof. Matthew Rosen, Martinos Center for Biomedical Imaging, Harvard.
Prof. Peder Larson, Univeristy of California, San Francisco (UCSF)
Prof. William Grissom, Biomedical Engineering Department, Vanderbilt University, TN.
Dr. David Waddington - The University of Sydney
Prof. Andrew Webb, Leiden University Medical Center (LUMC), Leiden, The Netherlands.
Prof. Martin Uecker, Department of Diagnostic and Interventional Radiology, University Medical Center Göttingen.
Prof. Stella Yu, Director of the ICSI Vision Group at the Berkeley Institute for Data Science.
Students (mentees)
Han Cui
Max Lister
Alma Harlangen
(c) website copyright Efrat Barak Shimron, 2022