Browsing by Subject "signal processing"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Artifact Reduction In The Integration Of Neural Electrodes And Extracellular Recording With Ultrahigh Field Magnetic Resonance Imaging
    (2020-05) Cruttenden, Corey
    The understanding of brain function in humans and animals can be greatly improved by simultaneous recording of electrical neural signals and acquisition of ultrahigh field (UHF) magnetic resonance images. Such simultaneous recordings will enable the study of neurovascular coupling with functional structure specificity and improve our understanding of the circuits and brain function changes associated with deep brain stimulation (DBS). However, integrating neural electrodes with UHF MRI is technically challenging due to electromagnetic interactions between the electrodes and the MRI magnetic fields. These interactions include magnetic field distortions by the electrodes that create image artifacts in MRI, as well as electromagnetic inductive coupling that introduces noise and interferences in the extracellular neural signal recordings made at UHF. This dissertation develops solutions to address both types of interferences. The MRI image artifact issue is addressed through novel neural electrode designs, and the electrical interferences in extracellular neural recordings are addressed using software filtering and estimation techniques. Two types of implantable neural electrodes are presented that significantly reduce artifacts in UHF MRI images due to their improved properties including magnetic susceptibility that better matches the surrounding brain tissue. The improved magnetic susceptibility match with brain tissue reduces distortions to the static magnetic field, which consequently reduces MRI image artifacts. First, carbon nanotube film electrodes are shown to substantially reduce image artifacts in UHF MRI, and second, a novel gold-aluminum microwire neural electrode is developed that is very easy to construct and provides multiple channels for recording or stimulation with reduced artifacts at UHF. The ease of constructing the gold-aluminum microwire electrode increases its potential impact to the field because it can be quickly adopted by other research groups and applied to any MRI field strength. The ability of both types of electrodes to reduce UHF MRI image artifacts is demonstrated using both phantom tissue and in vivo animal studies, and also verified using numerical computations of the magnetic field distortions around a two-dimensional electrode model embedded in a brain tissue substrate. Three major artifact removal algorithms are developed for cleaning neural signal recordings in a UHF MRI scanner. These are for the specific tasks of removing artifacts due to periodic motions including breathing and hardware vibration, eliminating severe scanning artifacts from the gradients of the MRI system during fMRI acquisition, and extracting action potential spike waveforms that are otherwise well below detection thresholds in an UHF 16.4 T MRI animal research scanner. First, reference-free adaptive filtering is implemented for removal of periodic motion induced interferences in extracellular local field potential (LFP) recordings at UHF. Second, a new approach to estimate severe fMRI gradient-induced artifacts using a coefficient shrinkage algorithm based on the first difference of the extracellular neural signal is presented. The first difference of the extracellular neural signal is found to have interesting statistical properties that benefit MRI artifact estimation including a Gaussian probability density function, a near-white power spectral density, an approximate Dirac delta function autocorrelation, and an upper bound on its singular value distribution. Finally, an adaptive virtual referencing approach based on adaptive least mean square (LMS) filters is shown to reduce the noise floor in extracellular unit recordings made both on the benchtop and at UHF. The reduced noise floor allows for identification of additional action potential waveforms that were previously below the detection threshold. This advancement allowed us to detect and classify action potential waveforms during fMRI data acquisition inside a 16.4 T MRI animal research scanner, the highest field strength horizontal-bore small animal scanner currently available. The artifact estimation and removal algorithms developed to improve the extracellular neural signal quality recorded during UHF fMRI could be beneficial in other applications as well. For example, periodic motion artifacts caused by breathing, heart pulsation, chewing, and blinking are often present in extracellular neural recordings made outside of MRI scanners, particularly if they are made in awake/behaving animals. Furthermore, the coefficient shrinkage algorithm used for removing fMRI gradient-induced artifacts could benefit the removal of stimulation artifacts from neural recordings made during electrical or optogenetic brain stimulation. The findings related to the properties of the first difference of the extracellular neural signal that provided statistical advantages in noise estimation and removal might apply to other signals with a 1/f power spectral density profile as well. Finally, adaptive virtual referencing for extraction of action potential spike waveforms is shown in this dissertation to benefit benchtop recordings in addition to recordings made at UHF. The technical contributions of this dissertation enable preclinical animal studies to be undertaken involving the simultaneous use of neural electrodes together with UHF MRI. Such studies can answer questions about the relationship between neuronal signaling and the functional MRI contrast with cortical layer and column specificity. Further, such studies will enable us to conduct fMRI imaging while using DBS electrodes which will help us understand the underlying mechanisms of how DBS successfully treats multiple brain disorders.
  • Thumbnail Image
    Item
    Improving the Specificity of Medical Ultrasound Imaging Using Scattering Spectrum Matching
    (2021-06) Al-Hussayen, Anas
    Medical ultrasound imaging systems continue to evolve at a fast pace enabled by advances in transducer technology, analog mixed-signal electronics, and high-performance computing. Ultrasound imaging is used in numerous applications, including echocardiography and peripheral vascular imaging in addition to specific anatomical imaging such as liver and kidney. In all these applications, ultrasound offers major advantages of portability, real-time visualization and measurements, and relative ease of use with and without the use of ultrasound contrast agents (UCA). One of the main limitations of this modality is the speckle phenomenon, which limits the contrast resolution. The contrast resolution is further compromised when imaging soft (low-scattering) tissue targets through or in the presence of strongly-scattering objects, which could introduce reverberation and/or clutter artifacts. Modern ultrasound employs ultrawideband transducers with fractional bandwidths larger than 70% being typical. This provides the opportunity to improve the image quality using frequency compounding methods, which have been traditionally used in speckle reduction at the expense of spatial resolution. However, this should be done with some understanding of the underlying scatterer spectrum in order to realize the promise of improved contrast. For example, in a uniform speckle region, almost any filterbank decomposition and compounding of the echo data results in an improved contrast ratio. For example, multiband audio signal processing is often based on the use of filterbanks with constant bandwidth B or constant quality factor Q. Such decomposition approaches are useful to gain insight into the spectrum characteristics. Complex scattering regions such as the surroundings of blood vessels may require a different approach for spectrum decomposition. This thesis research proposes the scattering spectrum matching (SSM) approach to the decomposition and compounding ultrasound echo data. The SSM approach seeks to decompose the echo signal spectrum based on an autoregressive (AR) model of the echo data rather than a preconceived filterbank design of the multiband signal processing architecture, e.g. constant B or constant Q. This statistical approach to spectrum decomposition accounts for the fact that the scattering spectrum is a `mixture' of scattering modes defined by the tissue organization in a given region of interest. It paves the way for adaptive learning algorithms for improving the specificity of medical ultrasound data. The SSM method is applied to medical ultrasound imaging of heterogeneous scattering from a peripheral artery with surrounding connective tissue and fat in addition to the intervening muscle tissue. Multiband analysis of the local spatial autocovariance reveals significant differences in the tissue organization in addition to reverberation components.
  • Thumbnail Image
    Item
    Sample 360 video for the analysis of plant movement
    (2018-09-12) Susko, Alexander, Q; susko004@umn.edu; Susko, Alexander, Q; University of Minnesota Oat Breeding and Genetics Lab
    Violent movement of cereal crop stems can lead to failure under high winds. Known as lodging, this phenomenon is particularly severe in cereal crops such as oat, barley, and wheat, and contributes to yield and economic losses. Quantifying the movement of cereal crops under field wind stress could aid in the breeding and selecting of lodging resistant cereals. We present a method to quantify the wave like movement of cereal crop rows in a high throughput fashion under field wind conditions. By analyzing pre-defined regions of hemispherical 4K resolution video, we obtain a time varying color signal of wind induced stem and canopy movement. Bandpass filtering is applied to the color signals to filter out changes in lighting due to sunlight changes, enabling comparisons across different lighting conditions. Peaks are then identified in the signal, and the distance in frames to the next peak as well as the absolute area under the curve between peaks is recorded. The distributions of distances to adjacent peaks (expressed as frequencies) are recorded and the area within a defined frequency bin is summed to get an approximation of the frequency and amount movement. We applied this method to analyze the wind induced movement of 16 cereal cultivars planted in a randomized complete block design on 5 different windy days. We detected significant differences in the mean frequency and amplitude within 0.2 Hz frequency bins among 16 cereal cultivars, with mean frequencies ranging between 1.24 and 1.53 Hz. This method quantifies the frequency and amplitude of movement in cereal varieties at high throughput in the field, and shows promise for characterizing the physiological basis for differences in cereal movement and lodging resistance.
  • Thumbnail Image
    Item
    Spatial and Temporal Identification of EUV Flares in Solar Active Regions
    (2022-12-15) Mahaffey, Theo M
    An automatic method developed to perform flare-finding on Level 1 EUV data from SDO/AIA is described in detail. The data are spatially binned, preprocessed, and compared against a Gaussian white noise background. The methods of Berghmans and Clette (1998) provide the basis for the automated detection algorithm; pertinent adjustments were made to address the high spatial resolution provided by AIA. Preliminary results for a short observation of NOAA active region 12712 from 16:00-16:45 UTC on 2018/05/29 are presented. Future work may include investigation of the link between EUV active regions and periodicities found in various observables, including radio emissions and magnetic field. Accurate determination of the flare frequency distribution for small flares is also a promising application of this method.
  • Thumbnail Image
    Item
    Tensor Methods for Signal Reconstruction and Network Embedding
    (2020-10) Kanatsoulis, Charilaos
    Over the past few years, the avalanche of data along with advances in methodological and algorithmic design have triggered an increased interest in machine learning (ML) and signal processing (SP) research. How do we fuse and complete multi-dimensional signals? What is a concise and informative representation of entities in multi-dimensional networks? How do we develop efficient lightweight algorithms that handle very large data? These are important questions that have risen on the top of the scientific and engineering agenda of ML and SP communities. A plethora of methods has been proposed to answer such questions. While neural networks are the current trend and powerful non-linear data-driven tools, there exist principled alternatives, such as multi-linear tensor methods, that are also effective and oftentimes significantly outperform neural network approaches. In the era of data deluge, multi-dimensional data, also known as tensors, are ubiquitous in a number of engineering tasks and data analytics. Tensors can model various types of data in high-impact domains. Images, for example, are space-space-spectrum cubes that can be naturally represented as tensors. Different types of networks as knowledge graphs and networks with attributed nodes are also tailored to tensor modeling. On the other hand, tensor decompositions have proven essential tools in understanding, analyzing and processing multi-dimensional data. They offer a flexible analytical framework with solid foundations, as well as efficient algorithms that effectively handle multi-dimensional data. This thesis aims to answer the aforementioned questions by exploiting tensor modeling and decomposition tools. The objective is to propose elegant and effective solutions to a number of challenging machine learning and signal processing problems. In particular three main research thrusts are investigated: i) Hyperspectral super-resolution; ii) Tensor sampling and reconstruction; and iii) Network representation learning. For each of the thrusts, this thesis offers an efficient framework that is supported by theoretical analysis, algorithmic foundations and thorough experimental investigation.