Back to Events

Seminar Series

April 24

Metric learning in diffeomorphic registration and a new fidelity measure based on unbalanced optimal transport.

Francois-Xavier Vialard
University Paris-Dauphine, France

Part Of

4:00P
- 5:00P

Location

32-D507
Add to Calendar 2017-04-24 16:00:00 2017-04-24 17:00:00 America/New_York Metric learning in diffeomorphic registration and a new fidelity measure based on unbalanced optimal transport. Diffeomorphic registration is an ill-posed inverse problem which isoften solved via the minimization of an energy functional whichinvolves a control on the deformation and a fidelity measure.Motivated by the design of these two terms which is crucial forpractical applications, my talk will be divided into two parts: thefirst one will focus on the design of the regularization of thedeformation in the LDDMM (large deformation by diffeomorphisms)setting. Starting from the simple sum of kernels, we show how toextend it to spatially varying metrics using left-invariantmetrics. We then present a method to estimate this metric in atemplate/population context. The second part of the talk willintroduce the use of optimal transport as a fidelity measure. Afterrecalling standard optimal transport, we present an extension ofoptimal transport to the case of unbalanced measures. Building onrecent numerical advances, we discuss the use of fast scalingalgorithms to approximate the problem via entropicregularization. Preliminary results on synthetic data are shown. 32-D507

April 04

Mapping human brain growth before birth using MRI

Colin Studholme
University of Washington

Part Of

4:00P
- 5:00P

Location

32-D507
Add to Calendar 2017-04-04 16:00:00 2017-04-04 17:00:00 America/New_York Mapping human brain growth before birth using MRI This talk will review some of the computational imaging techniquesbeing used to collect high resolution MRI of the fetal brain as partof the UW fetal brain database project. In particular, we will coversome of the key problems caused by extreme fetal head motion duringmulti-slice imaging and how these can be overcome by alternativeimaging and retrospective motion correction techniques. Examples ofstructural, functional and diffusion tensor imaging in the presence ofmotion will be included. 32-D507

March 16

Universal microbial diagnostics using random DNA probes

Amirali Aghazadeh
Rice University

Part Of

11:30A
- 12:30P

Location

32-D451
Add to Calendar 2017-03-16 11:30:00 2017-03-16 12:30:00 America/New_York Universal microbial diagnostics using random DNA probes Early identification of pathogens is essential for limiting development oftherapy-resistant pathogens and mitigating infectious disease outbreaks.Most bacterial detection schemes use target-specific probes todifferentiate pathogen species, creating time and cost inefficiencies inidentifying newly discovered organisms. In this talk I will present a noveluniversal microbial diagnostics (UMD) platform to screen for microbialorganisms in an infectious sample, using a small number of random DNAprobes that are agnostic to the target DNA sequences. Our platformleverages the theory of sparse signal recovery (compressive sensing) toidentify the composition of a microbial sample that potentially containsnovel or mutant species. We validated the UMD platform in vitro using fiverandom probes to recover 11 pathogenic bacteria. We further demonstrated insilico that UMD can be generalized to screen for common human pathogens indifferent taxonomy levels. UMD’s unorthodox sensing approach opens the doorto more efficient and universal molecular diagnostics. 32-D451

March 09

Simit and Taco: A Language and Compiler for Computing on Sparse Systems when Performance Matters

Fredrik Kjolstad
MIT CSAIL

Part Of

11:30A
- 12:30P

Location

32-D451
Add to Calendar 2017-03-09 11:30:00 2017-03-09 12:30:00 America/New_York Simit and Taco: A Language and Compiler for Computing on Sparse Systems when Performance Matters It is hard to write fast code, especially when data is sparse. We believelanguages and compilers can help. In this talk I will introduce Simit andTaco, a new language and a compiler for computing on sparse systems usinglinear and tensor algebra. Simit lets programmers seamlessly compute onsparse system both as individual elements of a hypergraph, and as thebehavior of the entire system in the form of global vectors, matrices andtensors. Simit provides a novel assembly construct that makes itconceptually easy and computationally efficient to move between these twoabstractions. As a result, Simit programs are typically shorter than Matlabprograms yet are competitive with hand-optimized code and also run on GPUs.Taco is a compiler for sparse linear and tensor algebra that compiles anytensor expression to fast loops. Together, Simit and Taco providesproductivity and efficiency for sparse computations. 32-D451

March 06

Continuous Representations of Brain Connectivity

Daniel Moyer
USC

Part Of

4:00P
- 5:00P

Location

32-D507
Add to Calendar 2017-03-06 16:00:00 2017-03-06 17:00:00 America/New_York Continuous Representations of Brain Connectivity Brain connectivity ("connectomics") has recently become a popularframe of analysis in medical imaging. Almost synonymous with this is agraph-theoretic representation of brain, in which gray matter regionsare nodes and the observed interactions between these regions areedges. While graph theory is a convenient abstraction, this view ofconnectivity requires the choice of a specific set of regions apriori, and thus is not able to easily model uncertainty betweenregion choices. In general popular graph summary statistics are alsonot robust to changing region sets.This talk will focus on recent work that instead introduces acontinuous representation of connectivity. I will start by presentinga point process model of structural connectivity, as well a KDE methodfor direct estimation of model parameters. I will then show aconnection to a subset of the discrete graph representations(exponential random graphs), and recent work on connectivity basedparcellation (region selection) exploiting this connection. 32-D507

February 10

Occlusion-aware face image analysis

Bernhard Egger
Universitaet Basel, Switzerland

Part Of

11:00A
- 12:00P

Location

32-D451
Add to Calendar 2017-02-10 11:00:00 2017-02-10 12:00:00 America/New_York Occlusion-aware face image analysis Analysis-by-Synthesis is a conceptually elegant and powerful approach forimage analysis. The idea is to start with a generative, probabilistic modelof the image, and to adapt the model parameters such that the generatedinstances are close to the observed image.In practical applications, the approach has proven to be difficult toapply. One of the big challenges is missing data and outliers which oftenoccur in real-world scenarios.This talk will focus on face image analysis where this challenge arises asocclusions, caused by various objects such as facial hair, glasses or evenobjects not directly related to faces. I present an occlusion-aware andfully probabilistic approach for adaptation of a three-dimensionalstatistical model of faces to single 2D images.I will start by presenting the 3D Morphable Face Model which is a combinedGaussian shape and color model. To generate images a camera andillumination model is incorporated in a computer graphics pipeline. Toanalyze a face image we search for model parameters rendering an imageclose to the observed target image. This posterior distribution ofsuitable instances is explored by a Markov chain Monte Carlo method. Wehandle occlusions by integrating Markov Random Field segmentation into faceand non-face during the model adaptation process. The segmentation of faceand non-face is solved together with the Morphable Model adaptation usingan EM-like algorithm. 32-D451

January 09

GIFT-Surg - Transforming surgery on unborn Babies

Sebastien Ourselin
University College London

Part Of

10:00A
- 11:00A

Location

Seminar Room D507
Add to Calendar 2017-01-09 10:00:00 2017-01-09 11:00:00 America/New_York GIFT-Surg - Transforming surgery on unborn Babies This talk will provide an overview of our Guided Instrumentation for Fetal Therapy and Surgery project (GIFT-Surg). In collaboration with KU Leuven, GIFT-Surg is working towards transforming surgery on unborn babies. Our ambition within GIFT-Surg is to deliver a new platform for fetal therapy and surgery through a unique combination of innovative interventional imaging systems and advanced surgical tools offering new levels of visualisation, flexibility and precision. This talk will highlight the combination of multidisciplinary expertise already involved in our project on the development of the main building blocks for this platform. Results on the design of new surgical manipulators, new intra-operative imaging devices including miniature photo-acoustics probe, new data fusion, tracking and segmentation tools and new surgical planning and guidance solutions will be presented. Seminar Room D507