Welcome back everybody to Beyond the Patterns.
So today I have the great pleasure to introduce a really great speaker and personal friend
Jong Cholye from KAIST in South Korea.
And you know Jong is a professor of the department of bio and brain engineering and a junk professor
at the department of mathematical sciences of the Korea's advanced institute of sciences
and technology KAIST.
He received his bachelor and master degrees from Seoul National University Korea and a
PhD from Purdue University West Lafayette.
Before joining KAIST he was a senior researcher at Philips research, GE global research in
New York and a postdoctoral fellow at the University of Illinois at Urbana-Champaign.
He has served as associate editor of IEEE transactions on image processing, IEEE transactions
on computational imaging and an editorial board member for magnetic resonance in medicine.
He is currently an associate editor for IEEE transactions on medical imaging and a senior
editor of the IEEE signal processing magazine.
He is an IEEE fellow, chair of the IEEE SPS computational imaging TC and IEEE EMBS distinguished
lecturer.
He was general co-chair for 2020 at the IEEE symposium on biomedical imaging, ESB together
with Matthew Jacob.
His current research is focusing on deep learning theory and algorithms for various imaging
reconstruction problems in x-rays, CT, MRI, optics, ultrasound, remote sensing and the
like.
So it's a great pleasure to have him here in the series today and he will present on
generative adversarial networks for medical image reconstruction.
John great pleasure to have you here.
The stage is yours.
Great thank you very much for the kind introduction and kind invitation.
It's really great my honor and pleasure to be here to present our work.
And as I introduced the title is again for medical image reconstruction.
Here I mainly talk about some of the unsupervised learning approaches based on the generative
adversarial network models.
As you know that we are currently living in the era of deep learning for medical imaging
but not only for diagnosis purpose for example diabetic disease from the fundus imaging or
skin cancer disease and also there are a lot of startup companies currently available working
on various aspects of the medical diagnosis.
And not only those kinds of diagnosis purpose and now deep learning is a state of art for
various image processing tasks such as segmentation and image registration.
And this kind of task usually the images are generated from the scanner and then using
the deep learning approaches inspired from the computer vision approaches we are utilizing
the we are performing the diagnosis and the image analysis.
However listen the new trend of deep learning is actually for the inverse problem or image
reconstruction.
In this case is from the sensor data from the scanning and scanners we are interested
in forming the high resolution high sensitive images using the deep neural network.
And this is the main topic where I'm going to talk.
In fact we actually we are quite proud of that we are one of the pioneers in this area
because in 2016 in AAPM LODO CT grant challenges we proposed world first deep learning approaches
for the LODO CT reconstruction.
At the time this was a world in the second prize instead of the first one.
The first one is actually still from our former students who is now in the MGH.
He actually proposed a model based architecture reconstruction approaches but our solution
Presenters
Zugänglich über
Offener Zugang
Dauer
01:25:09 Min
Aufnahmedatum
2020-12-04
Hochgeladen am
2020-12-04 17:08:45
Sprache
en-US
It’s a great pleasure to welcome Prof. Dr. Jong Chul Ye from KAIST for a presentation to our lab!
Title: GAN for Medical Image Reconstruction
Prof. Dr. Jong Chul Ye, KAIST
Abstract: Although deep neural networks have been widely studied for medical imaging applications, most of them are supervised learning framework that requires matched label data. Unfortunately, in many medical imaging applications, high-quality label data is often difficult to obtain, so the need for unsupervised learning is increasing. In this talk, I will mainly focus on unsupervised learning in medical image reconstruction problems such as low-dose X-ray CT, accelerated MRI, ultrasound, optics, etc., when the matched target data are not available. In particular, we introduce a recent advance of generative models, in particular optimal transport-driven CycleGAN framework, which has a strong mathematical background and can be readily incorporated with the imaging physics. The use of optimal transport driven cycleGAN for low-dose X-ray CT, CT metal artifact removal, accelerated MRI, MR motion artifact removal, ultrasound imaging artifact removal, etc., which have been pioneered by our lab, will be also introduced.
Short Bio: Jong Chul Ye is a Professor of the Dept. of Bio/Brain Engineering and Adjunct Professor at Dept. of Mathematical Sciences of Korea Advanced Institute of Science and Technology (KAIST), Korea. He received the B.Sc. and M.Sc. degrees from Seoul National University, Korea, and the Ph.D. from Purdue University, West Lafayette. Before joining KAIST, he was a Senior Researcher at Philips Research, GE Global Research in New York, and a postdoctoral fellow at the University of Illinois at Urbana Champaign. He has served as an associate editor of IEEE Trans. on Image Processing, IEEE Trans. on Computational Imaging, and an editorial board member for Magnetic Resonance in Medicine. He is currently an associate editor for IEEE Trans. on Medical Imaging, and a Senior Editor of IEEE Signal Processing Magazine. He is an IEEE Fellow, Chair of IEEE SPS Computational Imaging TC, and IEEE EMBS Distinguished Lecturer. He was a General Co-chair for 2020 IEEE Symp. On Biomedical Imaging (ISBI) (with Mathews Jacob). His current research focus is deep learning theory and algorithms for various imaging reconstruction problems in x-ray CT, MRI, optics, ultrasound, remote sensing, etc.
References
Continuous Conversion of CT Kernel using Switchable CycleGAN with AdaIN
S Yang, EY Kim, JC Ye
arXiv preprint arXiv:2011.13150 2020
Unsupervised MR Motion Artifact Deep Learning using Outlier-Rejecting Bootstrap Aggregation
G Oh, JE Lee, JC Ye
arXiv preprint arXiv:2011.06337
Unpaired Training of Deep Learning tMRA for Flexible Spatio-Temporal Resolution
E Cha, H Chung, EY Kim, JC Ye
IEEE Transactions on Medical Imaging
Unpaired deep learning for accelerated MRI using optimal transport driven cycleGAN
G Oh, B Sim, HJ Chung, L Sunwoo, JC Ye
IEEE Transactions on Computational Imaging 6, 1285-1296
AdaIN-Switchable CycleGAN for Efficient Unsupervised Low-Dose CT Denoising
J Gu, JC Ye
arXiv preprint arXiv:2008.05753
Two-Stage Deep Learning for Accelerated 3D Time-of-Flight MRA without Matched Training Data
H Chung, E Cha, L Sunwoo, JC Ye
arXiv preprint arXiv:2008.01362 2020
Cyclegan with a blur kernel for deconvolution microscopy: Optimal transport geometry
S Lim, H Park, SE Lee, S Chang, B Sim, JC Ye
IEEE Transactions on Computational Imaging 6, 1127-1138
Unsupervised CT Metal Artifact Learning using Attention-guided beta-CycleGAN
J Lee, J Gu, JC Ye
arXiv preprint arXiv:2007.03480
Assessing the importance of magnetic resonance contrasts using collaborative generative adversarial networks
D Lee, WJ Moon, JC Ye
Nature Machine Intelligence 2 (1), 34-42
Optimal transport, cyclegan, and penalized ls for unsupervised learning in inverse problems
B Sim, G Oh, S Lim, JC Ye
arXiv preprint arXiv:1909.12116
This video is released under CC BY 4.0. Please feel free to share and reuse.
For reminders to watch the new video follow on Twitter or LinkedIn. Also, join our network for information about talks, videos, and job offers in our Facebook and LinkedIn Groups.
Music Reference: Damiano Baldoni - Thinking of You