Macrocircuit Imaging Ed Boyden Synthetic Neurobiology Laboratory - - PowerPoint PPT Presentation

Macrocircuit Imaging

Ed Boyden

Synthetic Neurobiology Laboratory Massachusetts Institute of Technology Invited lecture at Stanford in CS379C on April 10, 2013

Today’s outline

PRINCIPLES, USES, UNKNOWNS

EEG MEG PET NIR fMRI

Brain scanning

• What we will cover today

– Physical principles of brain signals – Intuitions for how to think about brain signals

• What we will not cover today

– Advanced signal processing – Software/algorithms

• Enough classes do this (i.e., half of Course 6)
• Takes forever
• Probably easier to read it and go through the math, than

to hear a lecture

Brain scanning

• Read directly from the brain itself

– EEG – MEG

• Watch something inside the brain emit

– PET

• Deliver energy and monitor changes induced by

brain

– NIR – fMRI

EEG

• Electroencephalography

– About as simple as it gets – Put electrode on scalp

• Sometimes after abrading the

surface

• Gold + conductive gel
• Silver/silver chloride

– Capacitive sensing – non-contact sensors – Instrumentation amplifier to subtract out small voltages

• Voltage levels: 100-900 µV

(lower with higher frequency)

– Much lower than in the brain

• Frequency band: 1-70 Hz

– Much lower than in the brain

Simple to build

• OpenEEG project

How many electrodes?

• Often use 1-2 dozen sites

Uses

• Rapid, effective, characterization of brain state

– Simple, low-bitrate brain-computer interfaces – Art – Videogames

• Few electrodes
• Dry electrodes

RE: Altered oscillations as pathological signatures

• Pathological states have

specific alterations, e.g. epilepsy

• Paroxysmal fast (6-10 Hz)
• scillations

– Tonic-clonic seizures

• Spike-wave discharges

(seconds long, 1-3 Hz

• scillation)

– Absence seizures

RE: Spatial structure of oscillations

• Local sleep

– Do a motor task à see enhanced slow waves after learning, in the motor area (Huber et al., 2004)

• ‘Dolphins sleep with half their brain’

Visual evoked potentials

• EEG: over occipital lobe, after a stimulus

presentation

– Easy to use to check for blindness

Dipoles hypothesized to be V1àV2àparietal Also a propagating wave? Voltage-imaging studies ongoing.

Time+space: role of ongoing activity

• State-dependence of sensory inputs

– The ongoing phase of the EEG determines the impact

• f a sensory input, for example
• Voltage-dye expt (Arieli et al., 1996)

– No behavioral work as to the significance of this, yet. But the EEG cycles could set detection thresholds, etc.

Averaged evoked (like a visual evoked potential) Initial Summed Actual

‘Whole brain’ connectivity

• TMS + EEG: send out a test pulse, see where it

goes

– TMS of motor cortex, in waking and sleeping individuals wearing high-density EEG electrode arrays (Massimini et al., 2005)

Invasive EEG

• Electrocorticogram

– Remove the skull and dura, but still use surface electrodes – Can get higher frequencies (100- 500 Hz) than EEG – BCI – much higher bitrate (Leuthardt et al., 2004)

MEG

• Magnetoencephalography

– Ampere’s law: current causes a magnetic field – Direct pickup of brain’s magnetic field – The hard part: sensing the magnetic field

• 10-14 T or less! (Earth = 5 x 10-5 T)
• Early days used regular copper coil, but now everyone uses

SQUIDs

• Need high magnetic permeability shielded room (nickel-iron

alloy (75% nickel, 15% iron, plus copper and molybdenum) )

MEG sensors

• SQUIDs

– Superconducting ring, interrupted by a thin insulator

• I = current through the

junction

• φ = phase of electron

wavefunctions across junction = 2π Φ/Φ0

• Ic = constant (critical

current)

• Φ = magnetic flux in

ring, Φ0 = flux quantum (2.07 × 10−15 T m2 ) Ι = Ιc * sin(π Φ/Φ0)

Uses

• Expensive, but in principle you can see deep

magnetic sources, without distortion by overlying tissues

– Magnetic fields pass through unaffected, unlike EEG – Decent spatial resolution – Very fast time resolution PROBLEM: Underconstrained inversion of the sensors into deep magnetic fields Lots of effort to invent assumptions that allow inversion

• f the sensors.

Brain scanning

• Read directly from the brain itself

– EEG – MEG

• Watch something inside the brain emit

– PET – SPECT

• Deliver energy and monitor changes induced

by brain

– NIR – fMRI

Properties of the brain

Brain scanning

• Read directly from the brain itself

– EEG – MEG

• Watch something inside the brain emit

– PET – SPECT

• Deliver energy and monitor changes induced by

brain

– NIR – fMRI

General principles

• No need for magnetic fields like MRI (safe for implanted

patients)

• Functional: can measure glucose, dopamine binding, etc.
• Energy transmits through tissue. Image indicates

– Distribution of energy sources – Molecular environments or properties of tissue through which radiation propagates

• Contrast = signal vs. background, ΔI/I
• Signal-to-noise = signal vs. variation of signal
• Resolution = Δx
• Often calibrate these using “standard” signals or materials

– Current in a wire = “axon” – Water, blood, etc. = simulates flow in body

Tomography: building up a 3-D picture from lots of 2-D pictures

• Take pictures from all sides of an object
• For each 2-D picture, each point represents a line

integral

Same holds in polar coordinates

R

PET

• Positron Emission Tomography

– Radioactive isotope introduced into body

• Isotope: different numbers of neutrons in the

nucleus

• Unstable: called a radionuclide

– Not enough neutrons (usu 150% of # of protons)

– Decays, emitting positron

• weak force decay: p+ à n0 + β+ + νe
• Half life (t1/2 = ln(2) * τ, where decay is e-t/τ)

– carbon-11 20.3 minutes – oxygen-15 2.03 minutes – fluorine-18 109.8 minutes – bromine-75 98.0 minutes

ß About ideal!

PET

• Positron Emission Tomography

– Positron hits electron

• β+ + β− à γ + γ (0.511 MeV each)
• mean free path of the positrons in brain

tissue limits the resolution of PET scanning to about 4 mm

• Conservation of energy, linear

momentum, angular momentum

• Gamma rays emitted in opposite

directions, polarized in opposite directions

Detection of the gamma rays

• Two gamma rays emitted in
• pposite directions, with opposite

angular momentum

– Hits scintillator (e.g., bismuth germanate, BGO), which emits light, which goes to PMT, which causes current – 0th order: the pair of detectors

• Over time: intersection point of

these lines yields the target

• Lines project back to sources

– 1st order: the time-of-flight

• 1 ft = 1 ns, hard to do, but

modern machines do it

18FDG: 18fluoro-deoxyglucose

• Accumulated by metabolizing cells (esp if low

blood sugar – often patients fast)

• Can’t be metabolized until it decays (2 hr)
• When decays, 18F à 18O, and becomes normal

glucose (except with slightly higher molecular weight; extra neutron)

• Need a cyclotron to make FDG

– Must be within 2 hours of a cyclotron

Examples

• PET of depression

–

15O water – blood flow

measurement – (Mayberg et al., 2002)

Can use radionuclides attached to probes

• ther than FDG
• 11C-raclopride

– Binds to D2 receptor; antagonist – Easily displaced by dopamine

• Less binding = more dopamine

– Can see dopamine release when:

• Taking nicotine
• Changing a decision-making rule
• Cocaine abusers taking Ritalin (Volkow et al., 1997)

– Less dopamine release in abusers of cocaine; correlates with lower self-report of pleasure – More dopamine release when abusers see paraphernalia (e.g., pipes, etc.)

SPECT

• Single photon emission computed tomography

– 99mTc-HMPAO (hexamethylpropylene amine oxime)

• Blood contrast agent
• 99mTc: half-life of 6.01 hours, excited state of the

nucleus (m = ‘metastable’)

– Made from decay of 99Mo à 99mTc + β + ν (half-life, 66 hours; cheaper than 18F)

• Relaxes by releasing a gamma ray (140 keV), and

nucleons rearrange

SPECT

• Detecting the gamma ray: the

gamma camera

– Goes through collimator (sheets of lead) – Hits a detector crystal (sodium iodide), which scintillates (i.e., fluoresces) – Photons go through PMT, which amplifies signal via electron cascade – Electrons picked up and digitized

• 1-3 gamma cameras circle the

patient

– As they rotate, record 2-D snapshots

• f gamma-ray intensity

– Backproject to form 3-D image

SPECT Images of Common Neurological and Psychiatric Disorders

Right Sided Stroke Alzheimer’s Disease pervasive hypoperfusion Head Trauma to left PFC - severe aggression problems/violence Depression increased limbic activity (left) and decreased prefrontal and temporal lobe activity

http://brighamrad.harvard.edu/education/online/BrainSPECT/Main_Slide_Show/Main_SS.html

PET and SPECT vs. other methods

• PET and SPECT measures absolute levels of

blood flow

• PET and SPECT can be used without large

magnetic field (compatible with brain implants)

• PET and SPECT can be used with functional

indicators (e.g., dopamine binding, etc.)

Brain scanning

• Read directly from the brain itself

– EEG – MEG

• Watch something inside the brain emit

– PET – SPECT

• Deliver energy and monitor changes induced

by brain

– NIR – fMRI

NIR spectroscopy

• Near-infrared photons can pass through the skull,

and through the brain

– 800 nm – measure blood volume – 700 nm – measures deoxy Hb (vs HbO2) – 900 nm – measures HbO2 (vs deoxy Hb)

NIR

• Oxygenation of blood goes up (‘BOLD,’

accompanies neural activity – explored in the last part of this talk)

– 700 nm light à absorbed less à optical path lengths increase – Using two wavelengths, one above 800 nm and one below 800 nm, can give more data

Diffuse optical tomography: rough calculations

• Beer Lambert Law

– OD = optical density, attentuation of light – µa=absorbance, <Lhead>= length photon goes – G = scattering loss (not measurable) – Assume that with brain use, for any given path, only µa changes – that is, no change in scattering occurs – Also asssume that all changes are small

Full models

• Radiative transport equation

– Photons into a point equal to all the photons scattering in, minus all the photons scattering out, minus those that scatter in and are absorbed – Very complex; can approximate as a diffusion equation when scattering events are numerous

• A photon going in scatters quite a bit in the brain:

hundreds of times before it comes out again

– Scattering coefficient: 2.2 mm-1 for gray matter, 10 mm-1 for white matter

• Assume isotropic scattering

– Absorption coefficient: 0.036 mm-1 for gray matter, 0.014 mm-1 for white matter

Can model photon diffusion with Monte Carlo methods

• In principle, doesn’t

depend on as many assumptions

• Inject in a billion photons,

simulate the actual scattering and absorption

– Nice for visualization – Photon can travel 5-10 times the actual distance (e.g., 10 cm distance becomes 100 cm actual distance) – Light can go a meter in the brain!

Then solve the inverse problem

• Can’t just backproject exactly like we did for PET/SPECT: VASTLY

UNDERCONSTRAINED PROBLEM

– The fact that a photon may scatter 100 times, means that assumptions must be made, in order to derive the pattern of brain regions from the

• ptical readout data.

– EXAMPLE

• In general: create a forward model

– Model light transport, with voxels of varying tissue composition

• Simplify to a linear model, perhaps one that can be solved by matrix

multiplication: input x M = output

– Assume that whatever changes are fairly small; okay for differential measurements

• Then solve for the inverse model through inversion of this matrix

– Often have some cost functions – Some people beginning to use ‘anatomical priors’ based on MRI data – greatly helps constrain the analysis

Time resolved measurements

• Speed of light: 30 cm ~ 1 ft is 1 ns

– 100 cm = 3 ns

• Can measure the time of flight using advanced optical

techniques

– Inject picosecond pulse of light into brain – Measure when the photon comes out

• Streak photography: photons in a series are converted to

electrons, then subject to a swept voltage that steers them to different locations. Then, can just look at the temporal properties of the pulse.

• Time-to-amplitude converter: count the arrival times of single

photons very precisely (in a digital way), using microchannel- plate PMTs, comparing through-the-brain, to directly-from-the- source

• Can reject background illumination easily, since only

tightly correlated photons matter

Frequency domain: phase measurement

• Pulse a laser at ~100 MHz (10 ns

interval), and then detect the phase shift during brain usage (Chance et al., 1998; Chen et al., 2000)

• Might see a few degrees of phase

shift

• Resolution < 0.1o which

corresponds to 0.7 ps or 0.2 mm pathlength error

• Cheap. Gets less information than

pure time detection

• Potentially faster too (don’t have to

sit there recording photon arrival times)

– But, less sensitive

Beyond blood flow: fast signals?

• UIUC group sees ‘early’ signals at 100 ms

(Gratton et al., 1997, 2000, 2001)

– A direct change in neural absorption, during the firing

• f a spike? Water flowing in through sodium

channels? Some fast metabolic change? – No other group has reported this?

Resolution is not great – cubic centimeters

• Paper this week: how one might be able to

reconstruct the signal better by looking at non- photonic signatures of the IR trajectory

– Can use IR illumination to image down to micron- scale!

Results

• Usually people just use fMRI
• Can be used for non-invasive brain-machine

interface

Can be used on infants

• Can image after birth,

continuously

– Thin skull – Can’t put in scanner

• Infants’ response to

painful stimuli

• Infants’ response to

language (Pena et al., 2003)

Hemorrhage detection

• Inexpensive handheld version exists now

– Detects pooling of blood in the brain

MRI

• Magnetic Resonance Imaging

– Can have a whole course on MRI itself

Basic idea of NMR

• Use nuclei as magnetometers
• Nuclei with an odd number of nucleons have

non-zero spin s

– Angular momentum P = γh/(2π) (s(s+1))1/2

• From now on assume s = ½, like it is for a

proton – Magnetic moment of the nucleus proportional to the total angular momentum: µ = γ P, where γ = gyromagnetic ratio (e.g., 42.58 MHz/T for 1H) – Align the nuclei in a magnetic field Bz along the z-axis, and you get µz = γh/(2π) mI, where mI =

• 1/2 or 1/2.
• µz = ±γh/(4π)
• Energy of spin is E = -µ · B = ± γhBz /(4π),

splits into two levels

• Boltzmann distribution applies: at room

temperature, spins will be in both the high and low energy states, but slightly more in the low energy state (1 part in a million!)

Tip a spin

• If a magnetic moment is tipped out of

alignment with B, then it will experience a torque µ x B, equal to the change in angular momentum P

– dP/dt = µ x B

• Can derive this (8.02) by integrating torque

= r x F over a loop of current, using Biot- Savart’s law for F = qv x B = I x B – But, remember µ = γ P – Therefore, dµ/dt = µ x (γB) – Solve: µ precesses (rotates around) Bz, like a top rotating around, at the frequency ω = γ|B|

• Larmor frequency
• Magnetic fields can cause spins to rotate about

the field axis

• Thus, we can rotate a spin away from Bz by

applying a field in the xy plane

– Skip the derivation of this – you get the idea – Now we have all the ingredients for using a proton as a magnetometer

Spins relax

• So,

– Tip a spin – Watch it precess: emits RF at Larmor frequency as it precesses – It will relax:

• Variation in local magnetic field (e.g.,

spin-spin interactions) causes frequency to vary

– ω = γ|B| depends on |B| – Dephasing causes signal to decay: T2 – Tip a spin 180 degrees, it’ll precess back

• Loss of energy to thermal noise and other

aspects of the “lattice”

– Like friction, irreversible – Returns to thermal equilibrium (e.g., alignment with bias field), loss of information occurs – Exponential decay: T1

The Gradient Equation

• Let B vary with spatial position along an axis (y-

axis)

• Larmor frequency is product of position, with

the gradient slope at that point (here constant)

• Now, frequency differs along the y-axis (and

more generally, for short durations, the angle or phase of the spin)

• If you tip spins now with a specific RF pulse,

you’ll selectively excite just one slice of spins, which will then precess, generating a signal

– Get a 1-D image!

• Can induce gradient along any axis, or with any

steepness, giving specific spins phase

y

Into 3-D

• If G varies continuously in time, can continuously tune the

phase of spins, and therefore the NMR signal

– Suppose at position r, there is signal amplitude (e.g., spin density) A(r) – For a given point, the accumulated phase at a given time is then proportional to ∫G · r dt = (∫G dt) · r ≡ K(t) · r

• The time-varying signal from a point is just sinusoidal:

S(r,t) = A(r) * e-2πi K(t)· r

• Integrate over the volume to get your total signal, and you

see that is just the signal you’ve measured is the Fourier transform of your desired density, in this “K-space”

– Then, your goal becomes, like the CT scan, to sample as many such points as you can, and then invert TONS OF EFFORT ON PULSE SEQUENCES to speed up this

• effort. New ones come along every year or so.

Now how do you see the brain?

• Blood flow: hemoglobin has different magnetic

properties with vs. w/o oxygen

– With oxygen: diamagnetic – Without oxygen: paramagnetic

• BOLD: perfusion of oxygenated blood decreases

tissue susceptibility, more closely matching it to the rest of tissue

– Less inhomogeneity à less dephasing à T2 lengthens à BOLD increase, for a T2-weighted image

Blood flow

• BOLD signal correlates more with local field

potential (e.g., dendritic input) than with local spiking

– Maybe even most correlated with gamma frequency activity? – Lots of activity going on here – People usually assume BOLD is linear – can just deconvolve blood flow changes back into neural activity, but blood saturates, and there are many nonlinearities

• May even depend on brain region? Not known yet

Examples

• Tons

– Already discussed the amygdala/cingulate imaging paper – Brain has been imaged on many many different tasks. Empathy, pain, Democrat/Republican, you name it.

MRI

• Tons of developments – one of the most

interesting areas of research

– New pulse sequences: extracting more and more information – Contrast agents – this week: a paper, on sensing calcium – SQUIDs as MRI sensors – Sitting-up MRI machines – Higher and higher magnetic fields: more resolution

Can you see the magnetic field of a firing neuron?

Papers for this week

READINGS Truong TK, Song AW. Finding neuroelectric activity under magnetic-field

• scillations (NAMO) with magnetic resonance

imaging in vivo. Proc Natl Acad Sci U S A. 2006 Aug 15;103(33): 12598-601. http://www.pnas.org/cgi/content/full/103/33/12598 Wang X, Pang Y, Ku G, Xie X, Stoica G, Wang LV. Noninvasive laser-induced photoacoustic tomography for structural and functional in vivo imaging of the brain. Nat Biotechnol. 2003 Jul;21(7):803-6. Epub 2003 Jun 15. http://www.nature.com/nbt/journal/v21/n7/abs/ nbt839.html Atanasijevic T, Shusteff M, Fam P, Jasanoff A Calcium-sensitive MRI contrast agents based on superparamagnetic iron oxide nanoparticles and calmodulin. Proc Natl Acad Sci U S A. 2006 Oct 3;103(40): 14707-12. Epub 2006 Sep 26. http://www.pnas.org/cgi/content/full/103/40/14707

Supplementary Readings Logothetis NK, Pauls J, Augath M, Trinath T, Oeltermann A. Neurophysiological investigation of the basis of the fMRI signal.

• Nature. 2001 Jul 12;412(6843):150-7.

http://www.nature.com/nature/journal/v412/n6843/abs/412150a0.html Ogawa S, Tank DW, Menon R, Ellermann JM, Kim SG, Merkle H, Ugurbil K. Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proc Natl Acad Sci U S A. 1992 Jul 1;89(13):5951-5. http://www.pnas.org/cgi/content/abstract/89/13/5951

• P. Tass, M. G. Rosenblum, J. Weule, J. Kurths, A. Pikovsky, J. Volkmann, A.

Schnitzler, and H.-J. Freund Detection of n:m Phase Locking from Noisy Data: Application to Magnetoencephalography

• Phys. Rev. Lett. 81, 3291 - 3294 (1998)

http://prola.aps.org/abstract/PRL/v81/i15/p3291_1 Parkes LM, de Lange FP, Fries P, Toni I, Norris DG. Inability to directly detect magnetic field changes associated with neuronal activity. Magn Reson Med. 2007 Feb;57(2):411-6. http://www3.interscience.wiley.com/cgi-bin/abstract/114099069/ABSTRACT Chance B, Nioka S, Zhao Z. A wearable brain imager. IEEE Eng Med Biol Mag. 2007 Jul-Aug;26(4):30-7. http://ieeexplore.ieee.org/iel5/51/4268305/04272295.pdf Chance B, Zhuang Z, UnAh C, Alter C, Lipton L. Cognition-activated low-frequency modulation of light absorption in human brain. Proc Natl Acad Sci U S A. 1993 Apr 15;90(8):3770-4. http://www.pnas.org/cgi/reprint/90/8/3770

Background

• Torque on a current-carrying loop