1. Strategies for Information Extraction and Artifact Reduction in Functional MRI Peter A. Bandettini, Ph.D Unit on Functional Imaging Methods & 3T Neuroimaging Core Facility Laboratory of Brain and Cognition National Institute of Mental Health

2. Categories of Questions Asked with fMRI Where? When? How much? --- How to get the brain to do what we want it to do in the context of an fMRI experiment? (limitations: limited time and signal to noise, motion, acoustic noise)

3. A Primary Challenge:...to make progressively more precise inferences using fMRI without making too many assumptions about non-neuronal physiologic factors.

4. Neuronal Activation Hemodynamics Measured fMRI Signal ?? Physiologic Factors

5. Physiologic Factors that Influence BOLD Contrast Blood oxygenation Blood volume Blood pressure Hematocrit Vessel size Coupling: Flow & CMRO 2

6. Blood Volume –Contrast agent injection and time series collection of T2* or T2 - weighted images BOLD –Time series collection of T2* or T2 - weighted images Perfusion –T1 weighting –Arterial spin labeling Contrast in Functional MRI

7. Resting Active

10. BOLD Contrast in the Detection of Neuronal Activity Cerebral Tissue Activation Local Vasodilation Increase in Cerebral Blood Flow and Volume Oxygen Delivery Exceeds Metabolic Need Increase in Capillary and Venous Blood Oxygenation Decrease in Deoxy-hemoglobin Deoxy-hemoglobin: paramagnetic Oxy-hemoglobin: diamagnetic Decrease in susceptibility-related intravoxel dephasing Increase in T2 and T2* Local Signal Increase in T2 and T2* - weighted sequences

11. The BOLD Signal Blood Oxygenation Level Dependent (BOLD) signal changes task

13. EPISTAR FAIR... Perfusion Time Series - -- - Perfusion / Flow Imaging

14. FAIREPISTARTI (ms) 200 400 600 800 1000 1200

15. Scanner and Hemodynamic Limits

16. Single Shot Imaging T2* decay EPI Readout Window ≈ 20 to 40 ms

17. Multishot Imaging T2* decay EPI Window 1 T2* decay EPI Window 2

18. Excitations1248 Matrix Size64 x 64128 x 128256 x 128 256 x 256 Multi Shot EPI

19. Partial k-space imaging T2* decay EPI Window

23. BOLD Perfusion RestActivation

24. Anatomy BOLD Perfusion

25. Venous inflow (Perf. No VN) Arterial inflow (BOLD TR < 500 ms) Hemodynamic Specificity

26. Volume BOLD Perfusion +-+- unique information baseline information multislice trivial highest C / N easy to implement multislice trivial non invasive highest temp. res. unique information control over ves. size baseline information non invasive invasive low C / N for func. complicated signal no baseline info. multislice non trivial lower temp. res. low C / N

27. 0200400600800100012001400 -10 -5 0 5 10 15 20 CBF (% increase) Time (seconds) 0200400600800100012001400 0 1 2 3 BOLD (% increase) Time (seconds) CMRO 2 -related BOLD signal deficit: N=12 hypercapnia visual stimulation CBFBOLD Simultaneous Perfusion and BOLD imaging during graded visual activation and hypercapnia Hoge, et al.

28. CBF-CMRO 2 coupling 01020304050 0 1 2 3 4 Perfusion (% increase) BOLD (% increase) 0 +10+20 -10 Characterizing Activation-induced CMRO 2 changes using calibration with hypercapnia Hoge, et al.

29. Computed CMRO 2 changes Subject 1 Subject 2 % 40 0 10 20 30 -10 -20 -30 -40 Hoge, et al.

30. fMRI signal change dynamics...

33. -0.5 0 0.5 1 1.5 2 1520253035 Time (sec) 1000 msec 100 msec 34 msec

35. BOLD response is nonlinear Short duration stimuli produce larger responses than expected Observed responseLinear response 010203040010203040 1000 ms 2000 ms 250 ms 500 ms 20 s

36. Source of the Nonlinearity Neuronal Hemodynamic –Miller et al. 1998 – Flow is linear, BOLD is nonlinear –Friston et al. 2000 – hemodynamics can explain nonlinearity If nonlinearity is hemodynamic in origin, a measure of this nonlinearity will reflect any spatial variation of the vasculature

37. Methods 16 s Stimulus Duration (SD) SD = 250 ms SD = 500 ms SD = 1000 ms SD = 2000 ms … … … … 20 s Blocked Trial Visual SD = 500 ms SD = 1000 ms SD = 2000 ms SD = 4000 ms Motor …

38. Observed Responses 250 ms500 ms1000 ms2000 ms 500 ms1000 ms2000 ms4000 ms measuredideal (linear) visual stimulation motor task

39. Compute nonlinearity (for each voxel) Amplitude of Response Area under response / Stimulus Duration t  Fit ideal (linear) to response Output Area / Input Area

40. Nonlinearity 012345 2 4 6 Stimulus Duration f (SD) 012345 2 4 6 Stimulus Duration f (SD) 012345 1 2 3 4 5 Stimulus Duration 012345 2 4 6 Magnitude Area Output / input linear VisualMotor

41. Results – visual task Nonlinearity Magnitude Latency

42. Results – motor task Nonlinearity Magnitude Latency

43. Results – visual task 012345 -2 2 4 6 8 Stimulus Duration f (SD) 012345 -2 2 4 6 8 Stimulus Duration f (SD) 010203040 010203040

44. Results – motor task 010203040 010203040 2 4 6 8 012345 Stimulus Duration f (SD) 2 4 6 8 012345 Stimulus Duration f (SD)

45. Reproducibility Nonlinearity 2 Nonlinearity 1 Nonlinearity 2 Nonlinearity 1 Visual taskMotor task Experiment 1Experiment 2Experiment 1Experiment 2

47. Latency Magnitude + 2 sec - 2 sec

48. Regions of Interest Used for Hemi-Field Experiment Right Hemisphere Left Hemisphere

49. 10 2030 Time (seconds) 9.0 seconds 15 seconds 500 msec

50. -2.4 -1.6 -0.8 0 0.8 1.6 2.4 3.2 0102030 Percent MR Signal Strength Time (seconds) Average of 6 runs Standard Deviations Shown Hemi-field with 500 msec asynchrony

51. = + 2.5 s - 2.5 s 0 s 500 ms Right Hemifield Left Hemifield -

52. = + 2.5 s - 2.5 s 0 s 250 ms Right Hemifield Left Hemifield -

53. What is “in” the data? Neuronal Activation 1. Task Related & Time Locked 2. Task Related & Not Time Locked 3. Not Task Related Hemodynamics MRI Signal Changes Cardiac Pulsation Respiration Motion

54. Recognize? Edge effects Shorter signal change latencies Unusually high signal changes External measuring devices Correct? Image registration algorithms Orthogonalize to motion-related function (cardiac, respiration, movement) Navigator echo for k-space alignment (for multishot techniques) Re-do scan Bypass? Paradigm timing strategies.. Gating (with T1-correction) Suppress? Flatten image contrast Physical restraint Averaging, smoothing

55. Power Spectra Image Respiration map 0 0.25 0.5 Hz 0.25 Hz Breathing at 1.5T 3ms 26ms 49ms 3ms

56. 26ms 49ms 0 0.68 (aliased) 0.5 Hz 0.68 Hz Cardiac rate at 3T Power Spectra Image Cardiac map

57. Temporal vs. Spatial SNR- 3T 26ms49ms SPIRAL EPI 27ms50ms 26ms49ms 27ms50ms

58. Auditory Activation by the Scanner... Another artifact...

59. Prior EPI Gradients ( 20 sec ) Image Acquisition ( 20 to 30 sec) Average Time SeriesDifference Time Series TCTCTCT -= TC C

60. Time (sec) 0 1 2 3 4 5 6 7 a. b. c.

62. a.

63. b.

65. How to deal with Scanner Noise? Clustered volume acquisition Talavage et al. Silent sequences

66. Neuronal Activation Input Strategies 1. Block Design 2. Frequency Encoding 3. Phase Encoding 4. Single Event 5. Orthogonal Block Design 6. Free behavior Design.

67. DeYoe et al.

68. Neuronal Activation Input Strategies 1. Block Design 2. Frequency Encoding 3. Phase Encoding 4. Single Event 5. Orthogonal Block Design 6. Free behavior Design.

70. 0.08 Hz 0.05 Hz spectral density c.c. > 0.5 with spectra

71. Neuronal Activation Input Strategies 1. Block Design 2. Frequency Encoding 3. Phase Encoding 4. Single Event 5. Orthogonal Block Design 6. Free behavior Design.

74. Neuronal Activation Input Strategies 1. Block Design 2. Frequency Encoding 3. Phase Encoding 4. Single Event 5. Orthogonal Block Design 6. Free behavior Design.

77. 234 5 678 9 101112 13 Overt Word Production

78. Tongue Movement Jaw Clenching

79. Neuronal Activation Input Strategies 1. Block Design 2. Frequency Encoding 3. Phase Encoding 4. Single Event 5. Orthogonal Block Design 6. Free behavior Design.

80. Example of a Set of Orthogonal Contrasts for Multiple Regression

81. Neuronal Activation Input Strategies 1. Block Design 2. Frequency Encoding 3. Phase Encoding 4. Single Event 5. Orthogonal Block Design 6. Free behavior Design.

82. Free Behavior Design Use a continuous measure as a reference function: Task performance Skin Conductance Heart, respiration rate.. Eye position EEG

83. Amygdala Sympathetic Nervous System The Skin Conductance Response (SCR) Ventromedial PFC Hypothalamus Resistance change across two electrodes induced by changes in sweating. Sweat Gland Orbitofrontal Cortex

84. Skin conductance data collection Equipment: UFI BioDerm Model 2701 Skin Conductance Meter, Slic-8000 8-channel A>D converter, Slic Software for Windows. Time from stim. to T 1/2 : ~6 to 10 s 4000mv range 0.1  S = to 500 mV amplitude 0.05  S threshold level: 250 mV 1  S = 1 mmhos Boucsein, Wolfram (1992). Electrodermal Activity. Plenum Press, NY Venables, Peter, (1991). Autonomic Activity ANYAS 620:191-207. 1 - 3 sec 1 - 3 sec 8 - 14 sec

85. Activity correlated with SCR changes

86. Brain activity correlated with SCR during “Rest”

87. Pulse Sequence Subject Scanner StimulusResponse Image Reconstruction Image Registration/ Motion Detection Time Series Analyses Compute & Display Neuropsychological Stuff Investigator! Processing Stream with Real Time fMRI

88. Reasons for real time fMRI Make sure you have good data before subject leaves the scanner Repeat bad imaging runs Provide feedback to subject (e.g., “stop nodding your head!”) Most important when dealing with patient populations –when FMRI is used for pre-surgical planning –when patients in study are hard to come by

89. Further Reasons.. Adjustment of stimulus level to reach a desired response magnitude –½ of peak response –mapping of voxel stimulus-response curves Carrying out experiment until some statistically significant result has been reached –Must pay attention to statisticians for this! –e.g., do tasks until language lateralization has been established S R

90. Things to Look At (à la AFNI ) Graphing voxel time series data Displaying EP images from time series Control Panel

91. FIM overlaid on SPGR, in Talairach coords Multislice layouts

92. Estimatedsubjectmovementparameters

93. The End < 1 s to render Blocked trials: 20 s on/20 s off 8 blocks Color shows through brain Correlation > 0.45 1 Blocks: 12345678

94. Functional Imaging Methods / 3T Group Staff Scientists: Sean Marrett Jerzy Bodurka Post Docs: Rasmus Birn Patrick Bellgowan Ziad Saad Clinical Fellow: James Patterson Graduate Student: Natalia Petridou Summer Students: Hannah Chang Courtney Kemps July 7, 2000