Ultra High Field 7.0T Magnetic Resonance Imaging (MRI) and their applications to neuroscience, especially to the areas of language and cognitive sciences will be discussed.  Ultra-high field MRI began to provide super-resolution tractographic images delineating the fine fiber  structures such as the sub-components of the superior longitudinal fasciculus (SLF), among others, suggesting potential applications of these fiber track information, for example, to analysis of language circuitry and other cognitive and behavioral sciences.
 In short, some recent results of the new tractographic imaging obtained with 7.0T MRI and its applications to language and cognitive sciences will be discussed and highlighted.
 

 In this talk, recent developments in Positron Emission Tomography (PET) / Magnetic Resonance Imaging (MRI) are explored and the challenges of simultaneous imaging as well as the opportunities afforded by the two modalities are discussed.  The unique sensitivity of PET (picomolar) and its quantitative capabilities can be associated with the superb spatial and temporal resolution of MR as well as its excellent soft tissue contrast to provide an ideal imaging modality for many cancers as well as cardiac and brain explorations.
 Specifically, the role of dual probes and other nanoparticles that can be used to probe simultaneously under the same conditions different physiological processes using a PET and an MR or an optical signal are presented.  Improvements in image quality and diagnostic accuracy are illustrated in specific patient studies and synergies between PET and MR spectroscopy are discussed in the context of guiding radiotherapy.  Beyond oncology, applications in cardiac (viability, perfusion) and brain imaging (neurodegenerative disease, traumatic brain injury) are presented including mapping of mitochondrial membrane potential and simultaneous PET/fMRI for mapping dopaminergic and serotoninergic neurotransmission.

 Ultrasound diagnosis is routinely used in obstetrics and gynecology for fetal biometry, and owing to its time-consuming process, there has been a great demand for automatic estimation. However, the automated analysis of ultrasound images is complicated because they are patient-specific, operator-dependent, and machine-specific. Among various types of fetal biometry, the accurate estimation of abdominal circumference (AC) is especially difficult to perform automatically because the abdomen has low contrast against surroundings, non-uniform contrast, and irregular shape compared to other parameters. The convolutional neural network (CNN) method, which has recently shown great successes in object recognition, was also applied in fetal biometry to analyze high-level features from ultrasound image data. However, this method has faced obstacles in the clinical environment: (i) it is difficult to collect sufficient data for training, and (ii) it is difficult to cope with serious artifacts including shadowing artifacts. We propose a specially designed CNN, which takes account of doctors' decision process, anatomical structure, and the characteristics of the ultrasound image.  This method increases classification performance with relatively small number of data and also deals with artifacts by including ultrasound propagation direction as well as multiple scale patches as inputs.  This  machine learning  method for fetal biometry shows good performance in most cases and even for ultrasound images deteriorated by shadowing artifacts. This talk is about  a joint work with Jaeseong Jang (NIMS), Bukweon Kim, Sung Min Lee (CSE, Yonsei), Yejin Park, Ja-YoungKwon (College of Medicine, Yonsei).

 Biomedical imaging plays a central role in medicine and biology with its range of applications and level of performance having increased dramatically during the past decade. After a brief recall of the classical inversion techniques (filtered backprojection (FBP), Tikhonov regularization and LMMSE estimation), we present a panorama of the more recent developments in image reconstruction. In particular, we review sparsity-based methods associated with compressed sensing. The role of advanced signal processing there is obvious and rather dramatic, as it allows reconstructing images from lesser views, which translates into faster imaging and/or a reduction of the radiation dose for the patient. We provide theoretical arguments (new representer theorems) that explain the limitations of conventional l2-norm minimization and the better performance of l1-regularization in the sub-sampled regime. We then discuss the emergence of 3rd generation methods that incorporate some form of learning. We present experimental results and comparisons with the state-of-the-art, including a novel algorithm that results from the combination of classical FBP and Deep ConvNets. We conclude the presentation with a discussion of some of the pitfalls and a list of challenges for the future.

Title: Adaptive regularization methods for dynamic MRI image reconstruction

 Dynamic MRI image reconstruction is an inherently under-determined inverse problem because the object is changing as the data is collected. (There is no such thing as "fully sampled data" in dynamic MRI.) The ill-posed nature of dynamic MRI requires some form of regularization (signal models) to distinguish among candidate solutions.  Traditional k-space data sharing methods (like keyhole imaging) use implicit signal models, whereas modern regularized methods use explicit signal models. Typical regularization methods are based on simple mathematical signal models such as wavelets.  This talk will focus on newer methods that are adaptive, where the signal model is learned either from training data or concurrently with reconstruction of the dynamic image sequence.  Machine learning ideas underly these approaches and I will discuss challenges and opportunities. This is joint work with Sai Ravishankar.

 In this talk, we will discuss a class of models and techniques that can effectively model and extract rich low-dimensional structures in high-dimensional data such as images and videos, despite nonlinear transformation, gross corruption, or severely compressed measurements. This work leverages recent advancements in convex optimization for recovering low-rank or sparse signals that provide both strong theoretical guarantees and efficient and scalable algorithms for solving such high-dimensional combinatorial problems. We illustrate how these new mathematical models and tools could bring disruptive changes to solutions to many challenging tasks in computer vision, image processing, and pattern recognition. We will also illustrate some emerging applications of these tools to other data types such as 3D range data, web documents, image tags, bioinformatics data, audio/music analysis, etc. In the end, we will discuss some extensions of such low-dimensional models, and their connections with other popular data-processing models such as deep neural networks.
 This is joint work with John Wright of Columbia, Emmanuel Candes of Stanford, Zhouchen Lin of Peking University, Shenghua Gao of ShanghaiTech, and my former students Zhengdong Zhang, Xiao Liang of Tsinghua University, Arvind Ganesh, Zihan Zhou, Kerui Min of UIUC etc.