Capturing Brain Dynamics: a combined neuroscience and engineering approach Adrian KC Lee, ScD ! Department of Speech & Hearing Sciences Institute for Learning and Brain Sciences University of Washington ! June 30, 2015 [LABS]N Speech & Hearing ! Sciences Laboratory for Auditory Brain Sciences & Neuroengineering Listen up! There may be a price! Human can do this effortlessly, computers cannot (for now). Can I order your Huskies cocktail? Do you like this! small batch! Did you order! order your rye? Manhattan with! bourbon or rye? Artwork: Michelle Drews That’ll be! $9.50 Rye. Wanna try? Alex Katz, The Cocktail Party, 1965. Licensed by VAGA, New York, NY Cocktail party problem (Cherry, ’53) Neuroengineering Listeners attend to “objects” Let’s play “password” (Listen to the male voice) What’s the password? What did you miss out on? Neuroengineering Top-down selective attention Let’s play “password” (Listen to the male voice) What’s the password? What did you miss out on? Neuroengineering Neurons Neurons ‣ Building blocks of the nervous system # neurons # synapses Sponge 0 0 Roundworm 302 5,000 Mouse 75,000,000 Cat 1,000,000,000 Human http://en.wikipedia.org/wiki/List_of_animals_by_number_of_neurons ~10 ~10 85,000,000,000 ~10 Neuroengineering Neuron morphology Dendrites Soma Axon Myelin sheath Terminal Button http://animatlab.com/Help/Documentation/NeuralNetworkEditor/NeuralSimulationPlugins/FiringRateNeuralPlugin/NeuronBasics/tabid/117/Default.aspx Neuroengineering Types of neurons (structural) ‣ Bipolar • 2 processes: 1 axon & 1 dendrite - seen in special sensory neurons for olfaction and vision ‣ Pseudo-unipolar • 1 process - most sensory neurons ‣ Multipolar • multiple processes - most common http://droualb.faculty.mjc.edu/Course%20Materials/Physiology%20101/Chapter%20Notes/Fall %202011/chapter_7%20Fall%202011.htm Neuroengineering Types of neurons (functional) ‣ Functional classification • Afferent - neurons transmitting sensory information from sensory receptors to the CNS • Interneurons - perform complex integrative and analytical functions. Most neurons are interneurons • Efferent - neurons transmitting commands from the CNS to sensory organs. Neuroengineering Action potential / Myelination ‣ Action Potential YouTube video • http://www.youtube.com/watch?v=ifD1YG07fB8 ‣ Myelination • several-layer wrapping around axon • prevents ion flow across membrane • Nodes of Ranvier: regions of exposed axon, action potentials regenerate here Node of Ranvier http://www.nature.com/nrn/journal/v4/n12/fig_tab/nrn1253_F1.html Neuroengineering Graded Potentials Neuroengineering Graded vs. Action Potentials ‣ Graded potentials • Variable magnitude • +ve or -ve - excit. or inhib. • Passive propagation • Spatial and temporal summation ‣ Action potentials • All-or-none - Threshold - Post-AP refractory period • Active propagation • Saltatory propagation (in myelinated axon) Neuroengineering Brain Structures Brain Structures ‣ Cerebrum (or cortex) http://www.youtube.com/watch?v=h5f56Ynb01E • largest part of the human brain • 4 lobes: frontal, parietal, occipital, temporal ‣ Cerebellum • “little brain” ‣ Limbic system • contains thalamus, hypothalamus, amygdala and hippocampus ‣ Brainstem • made of midbrain, pons and medulla http://www.withthebraininmind.org/buildingbrains/unit_3/unit_3_03.php Neuroengineering Lobes of the brain Frontal lobe Parietal lobe Occipital lobe Temporal lobe Neuroengineering Brodmann area Primary Somatosnsory Primary area (3, 1, 2) motor (4) ‣ Cerebral cortex defined by Wernicke’s cytoarchitecture Broca’s area area (39,40) • based on structure and (44, 45) organization of cells • first published by German anatomist Korbinian Broadmann in 1909 • published maps in humans, monkeys and other species. Auditory cortex (41, 42) Neuroengineering Gray / White matter ‣ Gray matter • made up of cell bodies ‣ White matter • made up of nerve fiber / mylinated axons Neuroengineering Methodology Neuroanatomy ‣ Gross anatomy of the brain Suzana Herculano-Houzel, 2009 Neuroengineering ‣ Golgi • whole neurons Staining ‣ Immuno • attach dye to specific chemical markers ‣ Nissl • cell bodies http://en.wikipedia.org/wiki/ File:Anaplastic_astrocytoma_-_gfap__very_high_mag.jpg http://www.conncad.com/gallery/spines_synapses.html http://vanat.cvm.umn.edu/neurHistAtls/pages/neuron6.html Neuroengineering Magnetic Resonance Imaging (MRI) ‣ Non-invasive technique • Limited spatial resolution ‣ Adjust signal contrast to differentiate tissues, nuclei based on different magnetic properties ‣ Measures brain morphometry Neuroengineering Determining connectivity ‣ Non-invasive techniques using MRI • Diffusion tensor imaging • Diffusion spectrum The Connectum Project http://upload.wikimedia.org/wikipedia/commons/d/d2/Illus_dti.gif Neuroengineering Functional anatomy Lomber and Malhotra, 2008 http://philosophicaldisquisitions.blogspot.com/2012/05/ hauskeller-on-enhancement-for-common.html ‣ Lesion studies • chronic - patient studies, surgical lesions • reversible - transcranial magnetic stimulation; cooling Neuroengineering Functional anatomy ‣ Functional neuroimaging • non-invasive techniques - most common: fMRI, EEG, MEG - other techniques: PET, NIRS, ASL: ★ ★ ★ PET = Positron emission tomography NIRS = Near-infrared spectroscopy ASL = Arterial Spin Labeling Neuroengineering Functional MRI Measuring BOLD response ‣ Blood-oxygen-level-dependent contrast (BOLD) • Discovered in the early 1990’s - Belliveau et al (HMS) - Ogawa et al (AT&T) • Oxygenated blood is diamagnetic - minimal effects on magnetic field • Deoxygenated blood is paramagnetic - clearly measurable additive magnetic field Neuroengineering Temporal dynamic of BOLD response ‣ BOLD hemodynamic response • Sampling rate at 0.5Hz is adequate • Repetition Time (TR) - Depends on SNR Spatial coverage (# of slices) Long TR maximizes SNR Short TR maximizes fMRI stats BOLD signal amplitude - M/EEG sampling rate ~1kHz ~20s Time © 2004 Sinauer Associates, Inc. Neuroengineering fMRI “Event-related designs” ‣ Assumes BOLD signal to be Linear Time-Invariant Estimate individual BOLD signal: 1. Parametrically (based on a canonical waveform, e.g., gamma functions) II. Non-parametrically (finite impulse response estimation) Block designs -- good for detection Event-related -- good for estimating shape of HRF Neuroengineering M/EEG Right-Hand Rule Neuroengineering Action Potentials Current configurations m Q m B Synapse Action currents Postsynaptic currents Courtesy of Matti Hämäläinen, Aug 2010 Time behavior ‣ Electric signals propagate within the brain along axons as a series of action potentials 100 mV 10 mV (AP) • 1 ms Modeled as “quadrupolar” 10 ms source Action potential Postsynaptic potential • Mediated by Na and K voltage-dependent ionic conductances http://www.youtube.com/watch?v=ifD1YG07fB8 http://www.youtube.com/watch?v=LT3VKAr4roo Neuroengineering Postsynaptic Potentials Current configurations m Q m B Synapse Action currents Postsynaptic currents Time behavior ‣ Postsynaptic potential (PSP) • In synapses, chemical transmitter 100 mVchange ion 10 mV permeabilites of postsynaptic membrane 1 ms 10 ms • Can adequately be Action potential Postsynaptic potential described as a single current dipole oriented along the dendrite Courtesy of Matti Hämäläinen, Aug 2010 Neuroengineering AP vs. PSP Current configurations Time behavior Current Source m Q m B Synapse Approx. 100 mV AP Quadrupolar 1 ms 10 ms Action potential PSP Field “falls off” rate 10 mV 1/r Postsynaptic potential Dipolar 1/r Action currents Postsynaptic currents Courtesy of Matti Hämäläinen, Aug 2010 Neuroengineering Time-varying electrical currents Current configurations Time behavior ‣ Action potential • fast time course (~ms) ‣ Postsynaptic potential 100 mV 10 mV m Q ms) • slower time course (~10 m 1 ms 10 ms ‣ Temporal summation of B Synapse currents is more effective for Action potential Postsynaptic potential synaptic currents Action currents Postsynaptic currents M/EEG signals are ‣ Therefore, largely due to synaptic current flow Courtesy of Matti Hämäläinen, Aug 2010 Neuroengineering Physics of M/EEG signals EEG V Jp Primary currents Courtesy of Matti Hämäläinen, Aug 2010 Neuroengineering Physics of M/EEG signals EEG V Jp ‣ The primary current is related to the postsynaptic activity ‣ The primary current generates a potential distribution (EEG) Primary currents and the associated volume currents ‣ The primary and volume currents together also create a magnetic field (MEG) Courtesy of Matti Hämäläinen, Aug 2010 Neuroengineering = Physics of M/EEG signals Conductivity profile is irrelevant B=0 B=0 No magnetic fieldradial from radial currents No magnetic field from currents in the sphere model Primary currents cortex EEG MEG = 0 EEG x 0 V Jp “MEG sees less, but sees more clearly.” - David Cohen, MEG inventor MEG x 0 EEG x 0 current sources Courtesy of Matti Hämäläinen, Aug 2010 Neuroengineering MEG vs. EEG field patterns Neuroengineering Maxwell Equations ‣ Equations describing the fundamental relationship between electricity and magnetism Quasi-static E = Electric Field approximation B = Magnetic Field 0 J = Total current density ∂B ρ = Total charge density ∇×E=− ∂t ε0 = Permittivity of free space ∇⋅B = 0 0 µ0 = Permeability of free space ∇⋅= Divergence operator ∂E + ( ∇ × B = µ0 * J + ε 0 ∇× = Curl operator - ρ ∇⋅E = ε0 ) ∂t , Neuroengineering Neuronal Synchronization Neural synchronization ‣ A red Ferrari sports car drives by you’re perceiving: • color -- red • stimulus category -- car • motion -- moving fast ‣ But information processed in different subregions of the brain • What binds these information? - phase synchronization supports neural communication - “communication through coherence” Neuroengineering Neural synchronization ‣ In animal experiments • extracellular action potentials - correlations between spikes in two regions - coupling between spikes in one region and local field potentials in the same or a different region (spike-field coherence) ‣ M/EEG • Phase relation between 2 regions - phase synchronization • Important to note: - no intrinsic relationship between power effects and phase synchronization on a larger spatial scale Neuroengineering Phase synchronization Phase synchronization phase lag = 0 phase lag ≠ 0 No phase synchronization Neuroengineering Cross-frequency coupling phase-amplitude coupling phase-phase coupling Neuroengineering fMRI vs M/EEG fMRI vs. EEG and MEG ‣ E/MEG = Electro-/ Magneto-encephalography • electric potential / magnetic field measured on / near the scalp - directly measures synaptic (postsynaptic) current flow - EEG and MEG provide complementary information ★ simultaneous E/MEG with additional MRI information can increase spatial resolution - temporal resolution: ~ ms Neuroengineering Trade-offs of fMRI and M/EEG Imaging Modality Temporal resolution Physiological Signals fMRI ~ seconds Hemodynamic ~ milliseconds Post-synaptic potential M/EEG ‣ Trade-offs: • fMRI spatial resolution (~ 1mm) < M/EEG (~ 1cm) • ms-resolution is important for auditory research • M/EEG can reveal oscillatory activities Salmelin, Parkkon, MEG: an introduction to methods Neuroengineering Choosing the right methodology Neuroengineering Source Imaging Forward Modeling Neuroengineering Inverse Estimate Neuroengineering Towards solving the Inverse Problem ‣ Mapping MEG signal onto the cortex Forward solution M/EEG y = Gq + ε { } E εε T = C Inverse estimate ( q̂ = arg min y − Gq q Equivalent currents 2 C Observed data Noise covariance + f ( q) ) Minimum L2-norm: |q| 2 Minimum Current: |q| Penalty term Neuroengineering Neuroengineering Next generation hearing-aid design? Artwork: Michelle Drews Sound Sound Brain Amp Amp Noise Reduction Brain Speech enhancement Feedforward only Sound Amp Brain with feedback Eric Larson Research scientist Majid Mirbagheri Postdoc fellow Mark Wronkiewicz Grad student (Neuro) Neuroengineering