IRP Focus: Imaging, Sensing & Networked Systems: Research Projects
Engineering Faster MR Imaging
Todd Constable, Dept. of Biomedical Eng., Yale School of Medicine
Magnetic resonance imaging has been used in medical imaging for 30 years. Engineers working to improve the technology today address two main issues: imaging speed and quality. Faster imaging would reduce costs by enabling higher use of MRI machines, and improvements in image resolution would enhance a physician’s ability to make a proper diagnosis.
Todd Constable, professor of biomedical engineering, diagnostic radiology and neurosurgery, leads a team of engineers who are working to both accelerate and improve the quality of MR imaging by spatially encoding the data an MR imager records.
Most recent improvements in the speed of MRI have been made by introducing arrays of small coils that receive signals from the body. Data from the coils are spatially encoded by magnetic field gradients—that’s what produces the loud clanging sound a patient hears inside the magnet. Instead of adding more coils, the Constable team’s approach has been to change the gradient encoding strategy. They’ve switched from using linear to nonlinear magnetic field gradients to encode spatial information in a manner that's maximally complementary to what is provided by receiver coil arrays.
Indeed, the approach increases the efficiency of data acquisition and allows excellent images to be obtained using less data and, therefore, less time. At current rates, an MRI imaging sequence may take up to 10 minutes to acquire. For complete scans, patients often must lie still in the machine for a full hour. The Constable Lab’s new method could produce images from 2 to 16 times faster. Alternatively, instead of speeding up the acquisition time, the technology could be used to produce higher resolution images than what are currently available.
Constable and colleagues have filed two patents on the approach and are working on a licensing agreement with Siemens Medical to incorporate it into a clinical platform. They estimate that within five years their technology could be built into all clinical magnetic resonance scanners, doubling throughput and reducing medical costs, or producing higher resolution images at current times.
Publications: Stockmann JP, Ciris PA, Galiana G, Tam L, Constable RT, O-Space imaging: Highly efficient parallel imaging using second-order nonlinear fields as encoding gradients with no phase encoding, Magn. Reson. Med., 64(2): 447-456, 2010.
Computational Tools to Aid Conspiracy Surveillance
Andreas Savvides, Dept. of Electrical Eng., Dept. of Comp. Sci.
If you’ve ever seen the HBO crime drama “The Wire,” you know the lengths to which criminals will go to confound police surveillance. Footage from cameras hidden over a busy public plaza do little to aid detectives in differentiating illegal activity from innocent routine interactions.
Yale electrical engineering professor Andreas Savvides is expanding on previous work in behavior modeling to develop computational tools that could augment surveillance technologies by automatically analyzing group activities and detecting patterns and aberrations. The technology, which Savvides is developing with colleagues at Johns Hopkins University Applied Physics Laboratory, could be deployed to help combat terrorist threats, or monitor people and vehicles in high-risk locations, as well as to assess more benign situations such as shopping patterns in a mall.
Their Group Activity Network Analysis software leverages APL software previously developed for rapid, iterative query refinement against a social-network database. Relying on a set of algorithms that parse sequences of interactions into layers, the software classifies observed entities such as objects—a weapon or a package, for instance—individuals, vehicles, or buildings. Resulting data populates a database that can then be speedily queried to return records that display group activities as graphs. Designed to be “sensor-independent,” the tools can be implemented with non-video surveillance systems too, such as cell phone or e-mail communications.
To test their tools, the researchers, have in fact, applied them to an open-air drug deal scenario inspired by “The Wire.” Savvides says the next step will be to implement, test, and refine strategies for more robustly specifying group activities and to develop new multi-point sensing modalities that can identify individual activities and interactions in more detail.
Johns Hopkins APL Technical Digest, Volume 30, Number 1, 2011
A. Bamis, J. Fang and A. Savvides, Detecting Interleaved Sequences and Groups in Camera Streams for Human Behavior Sensing, Proceedings of the Third ACM/IEEE International Conference on Distributed Smart Cameras, ICDSC, Como, Italy, August - September 2009
A. Bamis, J. Fang and A. Savvides, A Method for Discovering Components of Human Rituals from Streams of Sensor Data, Proceedings of the 19th ACM International Conference on Information and Knowledge Management, CIKM, Toronto, Canada, October 26-30, 2
Designing the Computational Fabric of Green Buildings
Andreas Savvides, Dept. of Electrical Eng., Dept. of Comp. Sci.; Michelle Addington, Yale School of Architecture, Yale School of Forestry & Environmental Studies; Juliana Wang, Yale Climate & Energy Institute
In the parlance of today’s construction industry, a “green building” is one that maintains a small carbon footprint by engaging alternative energy sources such as solar, wind, and fuel cells sustainably and with economic precision. No building can be a lean user of energy in isolation: resources must be coordinated with adjacent buildings and the power grid. To enable a building to efficiently manage its energy load, adapt to its occupants’ behaviors, and be an intelligent inhabitant of the “smart grid,” engineers must possess a detailed understanding of a building’s spatio-temporal properties. That requires ongoing observation of the state of the building.
One innovative method engineers have devised for tracking such data is to embed sensing and computing fabric across a building to sense, decide, and react in an autonomous manner. Yale’s Embedded Networks and Applications Lab (ENALAB), led by electrical engineering and computer science professor Andreas Savvides, has partnered with research groups across campus to use Yale’s buildings as a living lab to study how buildings consume energy. Together with researchers in the School of Architecture, the School of Forestry & Environmental Studies, and the Yale Climate and Energy Institute, Savvides’ team is developing metrics for better characterizing building performance. The research partnership is also part of a new multinational and multi-university consortium being formed to work closely with industry to develop technologies for Next Generation Intelligent Buildings.
One of the main challenges Savvides and collaborators are addressing is the creation of a unified computing fabric from a large number of wirelessly interconnected, low-cost sensors, processors, and actuators that can meet, as one unit, all of a community’s sensing, timing, and reliability constraints. This entails smooth coordination between the physical and cyber worlds, across digital and analog boundaries, and discrete computations to satisfy the constraints of a continuous regime.
D. Jung and A. Savvides, Estimating Building Consumption Breakdowns using ON/OFF State Sensing and Incremental Sub-Meter Deployments, Proceedings of the 8th ACM Conference on Embedded Sensor Systems (SenSys), Zurich, Switzerland, November 3-5, 2010
A. Savvides, I. Paschalidis and M. Caramanis, Cyber-Physical Systems for Next Generation Intelligent Buildings, Proceedings of the 2nd International Conference on Cyber Physical Systems, Chicago, IL April 12- 14, 2011
Developing In Vivo Microscopy for Fundamental Neuroscience
Michael Levene, Dept. of Biomedical Eng.
Major areas of experimental neuroscience, from micro-circuitry to gene function to basic anatomy, are best studied optically. Yet conventional microscopy approaches can only reach about 5 percent of the intact mouse brain. One goal of the Levene Lab is to image the remaining 95 percent using optical microscopy.
Levene and colleagues have engineered a technique using micro-optics,which overcomes the major hurdle to viewing the recesses of the live brain: the inability to get adequate optical access in vivo due to the strong scattering of light. Their method works by examining the brain longitudinally through a microscope that shines light through a tiny prism embedded in the brain of a live mouse. The prism allows a view deep into the brain, where firing neurons can be viewed. The approach is like using a periscope to peer into the brain, and the Levene team has been able, using calcium-sensitive dyes, to witness neurons responding when the mouse’s whiskers are tweaked.
The technique has implications for basic science as well as therapeutic interventions. Deep in vivo imaging could lead to a greater fundamental understanding of how the brain works and also has implications for understanding diseases such as epilepsy. Potential clinical applications include image-guided surgery. Levene and team continue to work with colleagues in Yale’s Neurosurgery Department to develop these ideas.
Enabling Wireless Communications without Infrastructure
Sekhar Tatikonda, Dept. of Electrical Eng., Dept. of Statistics, Dept. of Comp. Sci.
In the modern world of wireless communications, a dynamic network is one where parameters change due to the mobility and traffic variations among users, and an ad hoc network is one in which devices organize themselves to create a wireless communication network. Dynamic ad hoc networks are highly useful in situations such as emergency rescue settings, military communications, sensor networks, and personal networking where only a lightweight infrastructure is possible, control of the network is not centralized, and there is a need for rapid deployment.
The engineering challenges to creating such networks, however, are numerous. The Tatikonda electrical engineering lab is working to optimize dynamic wireless ad hoc networks for more widespread use in situations where users’ parameters are constantly changing—where nodes are moving and the type of transmission, data or voice, is variable.
Simple algorithms exist for scheduling time and assigning frequencies for a small number of users, but Tatikonda and team are tackling what is known as the graph coloring problem. With different colors corresponding to different frequencies, they assign colors to each node so that there is no contention when there are many thousands of users. The Yale electrical engineers test their work in the field in collaboration with robotics engineers at the University of Texas, Austin, using fleets of all-terrain, semi-autonomous robots.
Their goal is to develop algorithms that are scalable, efficient, robust to changes, and local in order to accommodate many users without exhausting the available number of frequencies and without requiring communication across the entire network. Ideally, Tatikonda’s color graphing solution will enable each node in an ad hoc network to look only at its neighbors and determine what frequency is available.
In the News:
Is This Frequency Available? The Next Generation of Communication Networks
Applying Fluctuation Spectroscopy to Clinical Use
Michael Levene, Department of Biomedical Engineering
A technique known as fluctuation spectroscopy has been around since 1970. Widely used in the biophysics community, it is useful for measuring absolute concentrations of particles in fluids.
The Levene Lab is pursuing the use of fluorescence fluctuation spectroscopy in the clinic to quickly and effectively measure the concentration of proteins in blood. Specifically, their approach could be useful for diagnosing von Willebrand Disease, a common inherited clotting disorder, and other blood coagulation disorders
Because von Willebrand Disease is caused by a defect in the protein known as von Willebrand factor that is necessary for blood clotting, fluorescence fluctuation spectroscopy can be used to label the proteins and measure the concentration and size of those proteins in the blood.
While existing tests for von Willebrand Disease involve running a gel that can take 16 hours, involves radioactivity, and often returns poor results, the Levene Lab’s fluorescence fluctuation spectroscopy technique can provide accurate and reproducible results quickly.
It could have applications in diagnosing, classifying, and monitoring disease, as well as in elucidating specifics about the pathogenetic mechanism of thrombotic thrombocytopenia purpura, and as a new tool for assessment of coagulation or bleeding risk in a variety of common systemic conditions.
Frontiers in Optics
Interpreting and Quantifying Images with Computers
Hemant Tagare, Yale School of Medicine, Dept. of Biomedical Eng., Dept. of Electrical Eng.; Frederick Sigworth, Yale School of Medicine, Dept. of Biomedical Eng.; Hongwei Wang, Dept. of MB&B
Teaching a computer to understand and interpret an image is a difficult problem that engineers have long been trying to solve with sophisticated mathematics and computational power. The problem becomes especially challenging when the image is noisy, out of focus, or contains confusing extraneous information.
Hemant Tagare’s lab is developing algorithms that process biological or medical images to filter out the noise, detect weak signals, and quantify their geometric content. One example is the lab's new effort in creating three-dimensional protein structure from two-dimensional electron microscopy images of the protein. Electron microscopy protein images can be as unclear as the view through a window in a heavy rainstorm, but Tagare’s team takes noisy images of the protein from a variety of angles and processes them to recreate the protein's 3D structure.
In collaboration with Yale professors Frederick Sigworth and Hongwei Wang, the Tagare team is trying to create 3D models that capture the flexible structure of the Dicer protein. Dicer is known for its ability to chop up viral RNA that enters a cell. Modeling its structure could help explain how it does this, and perhaps enable scientists to manipulate or mimic that mechanism in other proteins. What sets Tagare’s efforts apart from engineers elsewhere who are trying to model Dicer is the interdisciplinary approach that relies on close collaboration among biophysicists, biochemists, and image processing experts.
Application of Systems Engineering To Human Motor Control
Kumpati Narendra, Dept. of Electrical Eng.
Controlling the human body is a formidable task. It takes the newborn baby countless experiments with its limbs to acquire motor control skills needed to ambulate. Most people take for granted the ability to control movement, unless of course they are afflicted with a movement disorder.
Since 2006, Kumpati Narendra has been collaborating with Peter Reeves (formerly a graduate student at Yale, now at Michigan State University) to apply systems engineering and control theory to the study of human motor control. For the past six years, they have been investigating the effects of control impairments on common conditions such as back pain; more recently they have initiated a study of Parkinson’s Disease (PD).
Research on Back Pain: Interest in controlling human movement has had a long and distinguished history. For instance, Giovanni Alfonso Borelli applied mechanistic theory in the 17th century to predict how much effort was required by muscles in the spine to support the body and external loads. Others developed more sophisticated models of the spine to predict loading on the spine and related it to risk of injury. However, it wasn’t until fairly recently that researchers started to investigate how a motor control error, an unintentional mistake in muscle recruitment, could lead to injury under relatively nominal loads.
To elaborate on this work, Reeves and Narendra examined, in a series of papers, how the central nervous system maintains stability of the spine, and suggested how different types of control impairments may be responsible for back pain.
N. P. Reeves, K. S. Narendra, and J. Cholewicki, “Spine Stability: The six blind men and the elephant.” Clinical Biomechanics, vol. 22, pp. 266-274, 2007.
N. Reeves, J. Cholewicki, and K. S. Narendra, “Effects of reflex delays on postural control during unstable seated balance.” Journal of Biomechanics, vol. 42, pp. 164-170, 2009.
N. P. Reeves, K. S. Narendra, and J. Cholewicki, “Spine Stability: lessons from balancing a stick.” Clinical Biomechanics, vol. 26, pp. 325-330, May 2011.
Research on Parkinson’s Disease
Jean Martin Charcot, father of neurology, coined the term Parkinson’s Disease, after a London doctor, James Parkinson, who wrote a paper entitled “An Essay on the Shaking Palsy” in 1817. However, it was only after Sherrington’s work on the nervous system was well established that PD was related to malfunctions within motor circuits.
At present PD is considered as a set of neurodegenerative conditions affecting humans. While no single cause is known to be responsible, it is attributed to the composite result of defects or changes in a number of benign biological processes. Narendra and Reeves believe (following the significant work done at the Hamilton Institute in Maynooth, Ireland), that a systems approach can be a most effective one for the study of PD. They are investigating systematically the effects of small changes in different cellular and metabolic subsystems on motor control. They also believe that the same models can be used to propose corrective actions that may be best suited for an individual.
An alternative form of treatment for PD is through external electrical stimulation. In the 1990s, experiments were described which showed spectacular results using such an approach. Stimulation of the subthalamic nucleus was shown to dramatically reduce Parkinson’s Disease symptoms and restore normal motor function. Since that time, this form of stimulation has evolved into a second form of treating PD. Recently, Narendra and Reeves have also initiated research on the effect of external stimulation on electro-chemical signals in biological systems.
Investigating Atomic Precision at the Oxide Interface
Eric I. Altman, Dept. of ChE & EnvE; Sohrab Ismail-Beigi, Dept. of Applied Physics, Dept. of Physics
For many materials scientists, what goes on at the interface between two different materials is where the excitement is. Long gone are the days when electronics developers glued materials together by hand. Today’s materials are sandwiched together with atomic-scale precision – an advancement that has led to the control of exotic solid-state phenomena, such as magnetism and superconductivity at the nanoscale and a promise of applications that will have broad-sweeping impact on the technologies of our time.
The Atomic Scale Design, Control, and Characterization of Oxide Structures Interdisciplinary Research Group at the Center for Research on Innovative Structures and Phenomena (CRISP), led by Yale Professor of Chemical and Environmental Engineering Eric Altman and Associate Professor of Applied Physics Sohrab Ismail-Beigi is investigating the novel chemical, electronic, and magnetic properties that emerge at interfaces between oxides. The group’s work revolves around crystalline oxides –common compounds that can exhibit nearly every possible effect seen in solid-state physics.
Oxide materials range from the very common, such as sand, to esoteric materials that include high temperature superconductors and materials that change from insulators to metals when placed near magnets. Because of their lattice structure, crystalline oxides of different chemical composition can be stacked together, allowing for atomic-scale sandwiching of a variety of materials. What goes on at the interface between materials is of great interest and sometimes surprising, as was the case when researchers found superconducting properties could be displayed between two insulating materials. It can take years to design and grow materials with atomic precision, but CRISP has some of the top “growers” in the field, a team of theorists, and four oxide molecular beam epitaxy (MBE) machines—the sophisticated vacuum systems that grow materials a single atomic layer at a time– as well as state-of-the-art characterization tools uniquely suited to determining the positions and identities of all of the atoms at the interface and their chemical bonding.
The group’s research focuses on designing new materials with unique physical properties; creating new computing, communication, and sensing devices enabled by the novel properties of oxide interfaces; and understanding and manipulating the interactions between electrons that give rise to the novel properties.
In the news:
The Center for Research on Interface Structures and Phenomena
At the Oxide Interface: Where Experimentalists Play
Wireless Sensing Systems for Monitoring Behavior
Andreas Savvides, Dept. of Electrical Eng,, Dept. of Comp. Sci.
Sensing and surveillance technologies capable of not only monitoring but also interpreting human behavior could have a variety of uses in defense, public safety, and anti-terrorism. There are also potential uses in the home.
Electrical engineers in the Embedded Networks and Applications Lab (ENALAB) led by Andreas Savvides are investigating many uses for a sensing technology they have developed called the BehaviorScope. One focus is its effectiveness in monitoring health and wellness factors in the aging population. The BehaviorScope architecture integrates a variety of inputs from a large number of sensors into a single reasoning framework that can process, present, and communicate a subject’s behavior patterns in space and time in a given environment automatically. The technology hosts a collection of algorithms that provide new ways of interpreting human behavior data.
The goal of the architecture is to put low-level sensor data into high-level, more meaningful semantic forms that will allow the data to be further processed in the context of a host of applications and services. Ultimately, this technology could be employed to predict, prevent, and respond to precipitous events experienced by seniors and others.
19th ACM International Conference on Information and Knowledge Management
Proceedings of the Third ACM/IEEE International Conference on Distributed Smart Cameras
Special Issue of International Journal on Personal and Ubiquitous Computing
In the news:
Researchers are pioneering the use of ‘smart cameras’ to help monitor elderly people who live alone
Assisted Living: “Smart” Cameras and “Intelligent” Sensor Networks Provide Independence
Extending the Impact of Multiphoton Microscopy
Michael Levene, Dept. of Biomedical Eng.
From the first microscopes to Zernike's phase contrast and modern laser scanning fluorescence microscopes, each advance in microscopy technology has opened up new windows onto the mechanisms of life. A team of engineers led by Professor Michael Levene is working to extend the impact and reach, in particular, of multiphoton microscopy and optical spectroscopy through elegant technological innovations, many of which are ready for rapid adoption by biologists who are otherwise naive in optical physics.
One breakthrough from the Levene biomedical engineering lab has been the development of 3D models of whole, intact mouse organs. Combining multiphoton microscopy with “optical clearing,” which uses a solution that renders tissue transparent, the engineers are able to scan mouse organs and create high-resolution, 3D, images of the brain, small intestine, large intestine, kidney, lung, and testicles. They then create 3D models of the complete organs.
Until now, this was only possible by slicing the organs into thin sections or destroying them in the process. When combined with optical clearing, multiphoton microscopy—so called because it uses two photons to excite naturally fluorescent cells within the tissue—can image a larger field-of-view at much greater depths. Once the tissue is cleared, the researchers shine different wavelengths of light on it to excite the inherently fluorescent tissue. The fluorescence is displayed as different colors that highlight the different structures and tissue types.
Now that the concept has been tested on mouse organs, Levene’s team is moving to human samples, focusing on prostate, liver, and breast tissue. Long term, the engineers hope to translate the technology to be used for 3D virtual biopsies to detect or rule out cancer.