INTRO TO BOLD FMRI 2014 M.S. Cohen all rights reserved mscohen@g.ucla.edu OUTLINE FRANZ JOSEPH GALL (1758-1828) MRI & Fast MRI Observations Models Statistical Detection
PAUL BROCA (1824-1880) WILLIAM JAMES (1890) We must suppose a very delicate adjustment whereby the circulation follows the needs of the cerebral activity. Blood very likely may rush to each region of the cortex according as it is most active, but of this we know nothing. BRAIN ACTIVATION LEADS TO: CBF Increased +ΔR1 CBV O Utilization 2 Increased +ΔR2 (C+) Increased slightly? Venous [O 2 ] Increased -ΔR2* BOLD Glucose Utilization Increased? Lactate R1=1/T1 R2=1/T2
SIGNAL LOSSES FROM SPIN DEPHASING NMR Signal (%) 100 80 60 Darkness Flashing Lights injection Inhomogeneous Magnetic Fields Within Voxels Result in Spin Dephasing and Signal Loss in Gradient Echo Sequences B 0 10 20 30 40 Time (seconds) Capillary Jack Belliveau 1959-2014 http://www.nmr.mgh.harvard.edu/in-memoriam-jack-belliveau mscohen@ucla.edu 2014 M.S. Cohen all rights reserved BOLD 2014 M.S. Cohen all rights reserved Gradients of several Gauss/cm may exist near deoxy-hb-filled capillaries. mscohen@ucla.edu FMRI explores intensity variations in MR signal Effect of blood CO2 level on BOLD contrast. (a) Coronal slice brain image showing BOLD contrast from a rat anesthetized with urethane. The gas inspired was O2. (b) The same brain but with 90% O2/10%CO2 as the gas inspired. BOLD contrast is greatly reduced. S Ogawa, et al., intensity variations reflect venous [O2] PNAS, 87(24):9868,1990 2014 M.S. Cohen all rights reserved mscohen@ucla.edu 2014 M.S. Cohen all rights reserved 12 mscohen@ucla.edu
WHY DOES VENOUS O 2 INCREASE? (1) [O 2 ] 98% 60% 65% WHY DOES VENOUS O 2 INCREASE? (2) [O 2 ] 98% 30% 65% 0% flow Artery Brain Vein 0% Artery Brain Vein flow 0% 50% 0% 50% Under normal conditions oxygen diffuses down its concentration gradient from the capillary to the brain parenchyma As the brain becomes more active, the oxygen consumption increases, increasing the transluminal oxygen gradient. 13 14 WHY DOES VENOUS O 2 INCREASE? (3) WHY DOES VENOUS O 2 INCREASE? (4) 98% 98% 30% 32% 30% 65% [O 2 ] [O 2 ] 0% Artery Brain Vein flow 0% Artery Brain Vein flow 0% 50% 0% 50% As oxygen flows across the capillary lumen it is depleted in the capillary and no further oxygen can be delivered The vascular system responds by increasing blood flow so that more oxygenated blood is available throughout the capillary 15
WHY DOES VENOUS O 2 INCREASE? (5) [O 2 ] Because the blood flow is increased more oxygenated blood passes into the venous end of the capillary 0% Artery Brain Vein flow 98% 0% 50% 30% 65% WHY DOES VENOUS O 2 INCREASE? (6) [O 2 ] Because the blood flow is increased more oxygenated blood passes into the venous end of the capillary 0% Artery Brain Vein flow 98% 0% 50% 60% 65% GRADIENT-RECALLED ECHO Ken Kwong Baseline 30 s 50 s INVERSION RECOVERY TE=42 TR=3000 TI = 1100 THICKNESS=10 OFF OFF 110 s 130 s 170 s ON ON OFF 190 s 250 s 270 s Ken Kwong OFF ON ON Seiji Ogawa Ken Kwong
A Chink in the Armor HEMIFIELD ALTERNATION right hemisphere left hemisphere Signal Intensity 0 50 100 150 200 250 300 Time (seconds) Cohen 2013 Mark Cohen, all rights reserved www.brainmapping.org 22 ACTIVATION WITH MOVING VISUAL STIMULI CONTRAST RESPONSE TEST 1.6% 6.3% 25% 78% 82% MT V1 MT / V5 V1 0 60 120 180 240 300 360 Time (seconds) From R. Tootell
MOTION SENSITIVITY TEST MT TRADITIONAL MRI ANALYSIS - MODEL DRIVEN Task Timing V1 Moving Stationary Moving Stationary Observed Signals 0 60 120 180 240 300 360 Time (seconds) From R. Tootell TRADITIONAL MRI ANALYSIS - MODEL DRIVEN Task Model Signal Model z=5 z=1.5 Hemodynamic Response Model CONVOLUTION OF IMPULSE RESPONSES WITH STIMULI % increase over baseline 40 30 20 10 0-10 -20-30 stim stim stim -40 0 20 40 60 80 100 120 Time (seconds) Actual Response Convolutio n Model
AMPLITUDE-WEIGHTED LINEAR ESTIMATE Rate: Signal and Estimate 1 0 52 104 208 104 52 208-1 -0.5 0 40 80 120 160 200 240 Time (seconds) 0.5 0 Residual Error TRADITIONAL MRI ANALYSIS - MODEL DRIVEN Task Model Signal Model z=5 z=1.5 Hemodynamic Response Model Observed noise = MRI instrument noise + Scanner instabilities + Thermal noise from subjects + Subject physiological fluctuations + Subject motion + Interference from devices + other subject factors White Noise Pink (1/f) Noise
Loudness 60 Hz b a a b a - Scott Joplin piano rags b - Classical Radio c - Rock Radio d - News and Talk Radio c c d Pitch ad R. F. Voss and J. Clarke, Nature 258 (1975) 317. Human EEG 1kΩ Well-Behaved Noise: Scanner Noise Subject Thermal Noise 1kΩ BW =16000 Hz, T = 300 K, R 1000Ω N = BW 4 1.38E 23 T R N BW 4 1.38E 23 300 1000 V 0.5µV
10 0 Amplitude 10 BADLY BEHAVED NOISE Interference from Devices Scanner Instabilities 10 0 10 10 0 thermal noise 10 1 10 2 Frequency 10 0 quantization noise Subject physiological fluctuations Subject motion Amplitude 10 Amplitude 10 other subject factors 10 10 0 10 1 10 2 Frequency 10 10 0 10 1 10 2 Frequency
SCANNER PROBLEMS MODEL-FREE MRI ANALYSIS 2 0-2 -4-6 0 6 12 INDEPENDENT COMPONENTS ANALYSIS (ICA) Time Location (space) Scan #k fmri Image Data Spatial ICA for fmri # ICs Location (space) # ICs Time IC Spatial Maps ICA EXPOSES FUNCTIONAL NETWORKS data are decomposed into a set of spatially-independent maps and a set of time courses. impose spatial indepedence http://www.fmrib.ox.ac.uk/fslcourse/lectures/melodic.pdf
BRAIN READING MACHINE LEARNING IN FMRI Postulate: All interesting behavioral, affective, mental or cognitive states are the expression of, or reflected in, neural activity Resulting maps are difficult to interpret. Haxby, et al., Science 293:2426