OHSU

Adaptive Systems Lab

The Adaptive Systems Laboratory is home for research in machine learning, adaptive and statistical signal processing, statistical pattern recognition, image processing and analysis, biologically-inspired computation, stochastic dynamics, and modeling of biological systems.

Faculty
Todd Leen
Misha Pavel
Patrick Roberts
Izhak Shafran
Xubo Song
Kemal Sonmez


Research Projects
  • Augmented Cognition (Misha Pavel)
    The AugCog project aims to create a closed loop brain-computer interface system that manipulates the interface between a human operator and the computer system to modify the task difficulty in order to maximize task performance. The closed-loop control signal relies on the estimated cognitive load of the human subject as assessed from ambulatory EEG measurements. The main challange of this project is ambulatory EEG signal processing and cognitive state classification, since the mobile nature of most everyday tasks induce motion artifacts that are detremental to traditional techniques developed for laboratiry settings. A major aspect of our research is identifying task-relevant robust EEG channels and features using the maximum mutual information principle. Funded by DARPA (subcontract from Honeywell).

  • Clinical Laboratory Error Detection (Todd Leen, Deniz Erdogmus, and Steve Kazmierczak)
    Hospital clinical laboratory tests are a major source of medical information used to diagnose, treat, and monitor patients. Test errors only affect about 0.5% of samples, but lead to delays, additional expense, and sometimes to erroneous treatments that increase risk to patients. Such errors compromise clinical utility, cost effectiveness and patient safety. Current methods to detect errors are either so insensitive to errors that they do not detect sample faults reliably, or so prone to false-alarms that their alerts are routinely disregarded.

    This project develops and uses statistical machine learning technology to reliably detect errors in hospital clinical laboratory tests, using data derived from patient samples. A primary obstacle to developing reliable statistical detectors for lab errors is the cost of labeling samples combined with the low error rate. This research will provide algorithms for clinical lab error detection that will extend to tests used in other disease entities (for example diabetes and heart failure).

  • Cardiac Imaging and Motion Quantification (Xubo Song)
    The rapid adoption of aggressive treatments for heart failure, including resynchronization pacing, has renewed interest in the fundamental electromechanical properties of the left ventricle. Related to MRI, this has resulted in research efforts aimed at the development and application of methods to assess left ventricular mechanics, including magnetic resonance tissue tagging, phase velocity myocardial mapping and cine DENSE displacement mapping in 3D space. By comparison, there have been fewer efforts to study right ventricular mechanics, even though in certain conditions such as congenital heart diseases, pulmonary hypertension and cor pulmonale, RV function can be more important than LV function.

    In echocardiography, the development of tissue Doppler strain rate methods have been by now followed by newer speckle tracking methods capable of showing multiple types of strain deformation without angle dependency, including circumferential shortening strain, radial thickening strain in systole and twisting and untwisting. Two-dimensional methods can define, to some extent, segmental motion in cross-sectional views, but are compounded by through-plane motion. For both echo and MRI, it will likely require a robust 3D method to comprehensively define regional ventricular wall motion and deformation.

    This project is aimed at development and validation of a method to obtain spatially-dense displacement fields of both left and right ventricular deformation maps from ultrasound scans performed on an advanced 3D echocardiographic system; a method that should be capable of showing strain displacement fields and torsional deformation for both ventricles. This proposal is aimed at further developing the techniques we have pioneered and validating them in a unique dynamic cardiac phantom and in open-chest pigs using sonomicrometry as a gold standard. We will also evaluate the efficacy of our new approach and translate it to the clinical setting by studying patients with congenital heart disease and compare our results from our method MRI tagging, and wall velocity encoded strain measurements.

  • Position Tracking and Mobility Assessment of Elders (Misha Pavel, Eric Wan)
    This project aims to improve upon the recent advances in state estimation techniques for nonlinear dynamical systems. These include the unscented Kalman filter, the particle filter, and the sigma-point filters. Existing methods rely heavily on the accurate model knowledge for the specific state estimation domain, whereas many problems involve uncertainty about the dynamic equations and the noise distributions. Our purpose is to develop robust state estimation algorithms based on information theoretic estimation principles that can handle such uncertainties as well as outliers and sensor failures. Specifically, we are interested in the application of these techniques for unobtrusive monitoring of elderly in their homes using motion sensors and RFID transmitters in cooperation with the Point-of-Care Laboratory.

  • Stochastic Learning (Todd Leen)
    The discovery in recent years that synaptic plasticity is mediated by processes sensitive to the relative timing (in milliseconds) between pre- and post-synaptic events overturned models of synaptic change based on average activity levels (so-called rate-dependent models). The latter formed the dominant conceptual framework since the work by Donald Hebb published in 1949. This experimental discovery requires different theoretical and computational tools. Individual synaptic events have inherent random variability, so computational synaptic dynamics in the new paradigm must be based in the theory of stochastic processes. Previous work modeling the stochastic dynamics of neural systems uses approximation tools -- the nonlinear Fokker-Planck equation (FPE) -- known to be deeply flawed and potentially misleading. The situation recalls the use of the FPE by machine learning theorists in the early to mid 1990s; indeed the dynamics of spike-timing-dependent plasticity in neural systems and those of stochastic approximation algorithms in the machine learning literature are very similar.

    This project establishes rigorous tools for treating the stochastic dynamics of learning systems based on spike-timing-dependent synaptic plasticity. It develops well-grounded approximation techniques (and exact solutions where available) and applies them to synaptic dynamics in natural and artificial learning systems. The new methods are compared to those employed in the recent literature to provide insight into the accuracy and appropriateness of the various methods. The project is relevant to both the computational neuroscience and machine learning communities.

  • Four-Dimensional Imaging and Analysis for the Study of Immune Responses (Xubo Song, James Rosenbaum, Stephen Planck)
    The recent development of video-microcropy technology for imaging immune responses within the eye is revolutionizing the way researchers are studying and understanding the immune mechanism. Such technology enables the visualization of T cell behavior in disease models without resorting to surgical trauma. The motion patterns of T cells are directly related to the cellular and chemical environment at the site of inflammation and thus can reveal the underlying disease mechanism. However, such technology needs the capability for computer-aided image processing and analysis due to images that are compromised by motion artifacts that obfuscate the true T cell motion. In addition, the current practice of manual tracking of the cell locations is a prohibitive task for massive amount of images. The goal of this proposal is to develop image-processing techniques for tracking and characterizing cell motion in microscopic video of the ocular uveal tract. The specific aims are: (1) to develop techniques for image registration in order to stabilize images plagued with motion artifacts; (2) to develop computer-based image processing techniques to track cell motion; and (3) to statistically characterize cell motion. This research is a multidisciplinary collaboration that involves OHSU OGI and SOM investigators, who bring their expertise in ophthalmology, immunology, microscopy and image analysis to this effort. The data from these studies will constitute preliminary data for NIH proposals to study immune responses in four-dimensional microscopic imaging of the eye.

  • Sensory Processing and Learning (Patrick Roberts and Todd Leen)
    A key challenge in neuroscience is to understand how the central nervous system processes and stores spatial-temporal patterns of sensory information. The electrosensory lateral line lobe (ELL) of the mormyrid electric fish presents a special opportunity for examining this influence because much of the physiology and anatomy is well characterized. Until now, the weak link in the experimental study of the electrosensory processing has been the inability to apply sophisticated, precisely tailored, realistic spatial-temporal stimulus patterns while recording from the brain. This project develops and exercises a computer-controlled electro-sensory platform that allows the experimenter to present stimulus patterns that precisely probe the system, and thus allow testing of qualitative hypotheses and quantitative circuit models of ELL processing.