Diffuse Optical Imaging of Brain Activation David Boas photon migration imaging lab http://www.nmr.mgh.harvard.edu/DOT/ Andy Siegel Maria Angela . Franceschini Juliette Selb Ted Huppert Danny Joseph Jon Stott Sol Diamond Rick Hoge Anna Custo Yiheng Zhang Gary Strangman Anders Dale 2 cm 1.5 EEG/MEG PET/SPECT 1 Diffuse Optical 0.5 mm Spatial Sensitivity (log mm) m Why Optical Imaging of the Brain? fMRI 0 -3 -2 millisecond -1 0 second 1 2 3 hour 4 day Temporal Sensitivity (log sec) But also NOVEL CONTRAST that spans all other methods! Intrinsic Optical Contrast Parameters Hemodynamic •Oxy- & DeoxyHemoglobin •Blood Volume •Blood Flow Neuronal - Electrical •Scattering Metabolic •Cytochrome oxidase •CMRO2 Outline • Basics of Diffuse Optical Imaging Forward and Inverse Problem • Examples of Hemodynamic and Fast Imaging • Improving spatial uniformity, CNR, depth sensitivity • Multi-modality – Optical and MRI but also MEG/EEG CMRO2 temporal and spatial correlation Synergy Penetrating Deep Tissues la ~ 100 mm ls ~ 1 mm Absorption Scattering Radiative Transport Equation ˆ ˆ ˆ ˆ ˆ 1 L r , ,t ˆ ˆ d ˆ S r, ˆ ,t L r, , t t L r , , t s L r , ,t f v t FLUENCE ˆ ˆ r , t dL r , , t ˆ ˆ ˆ J r , t dL r , , t FLUX ˆ 1 3 L r , , t r , t J r , t 4 4 DIFFUSION APPROXIMATION Photon Diffuse Equation 2 Time Domain: D v a r, t vSr, t t Frequency: 2 k 2 r v S r D v a i k D 2 Photon Migration in the Head Source Photon Migration in the Head (a) (b) layer a (cm-1) s’ (cm-1) scalp 0.191 6.6 skull 0.136 8.6 csf 0.026 0.1 gray matter 0.186 11.1 white matter 0.186 11.1 Comparison of X-Ray CT and DOT DOT Image Sensitivity Profile X-Ray Linear Image Reconstruction PERT rs , rd d 3 r INC rs , r r G r, rd y Ax { yi PERT rs ,i , rd ,i Ai , j INC rs ,i , r j G r j , rd ,i d 3r x j r j T 1 T ~ Least Square’s Solution: x ( A A) A y Under-determined, ill-posed, ill-conditioned REGULARIZATION Spectroscopic Determination of Oxy- & Deoxy-Hemoglobin a HbO ( )[ HbO 2 ] HbR ( )[ HbR ] 2 0 a (1 ) 1 (1 ) A(1 ) ( ) 0 ( ) A ( ) a 2 2 1 2 Φ1 Ax n W RA T ( ARA T C) 1 xˆ WΦ m [ HbO 2 ] AE [ HbR ] HbO2 (1 )A(1 ) HbR1 (1 )A(1 ) HbO2 ( ) A ( ) ( ) A ( ) HbR 2 HbR 2 2 HbO2 2 INSTRUMENTATION CW, RF, and Time Domain Advantage Disadvan. Time Frequency Time Impulse Response Expensive Complex Physicist Freq Phase & Amp Fast Cheaper RF elec. CW Cheapest Easy electronics Fast Amp only instrument development - CW I0 It Techen Inc., Milford, MA, http://www.nirsoptix.com instrument development - CW 32 laser diode sources (690 & 830nm) frequency encoded in 200Hz steps between 4.0kHz and 7.4kHz 32 parallel APD detectors detector’s output is digitized at ~40kHz on-line, individual source signals obtained off-line by infinite-impulse-response filters acquisition time per image (32x32 channels) can be as short as 10ms!!! Techen Inc., Milford, MA, http://www.nirsoptix.com instrument development - TD I0 t=0 It ~ns light source: mode-locked Ti:Sapphire laser (Spectra Physics MaiTai, <100fs pulse width) tunable from 750–850nm light detector: temporally gated image intensified CCD detector (LaVision) which simultaneously image light from up to 315 detector fibers the laser delivers light to up to 150 different source positions, with a dual-axis galvanometer that switches between any two fiber outputs in ~1ms the gated image intensifier acts as a fast (200–1000ps) shutter passing only a “window” of the emerging light back to the CCD ART, Montreal Canada, http://www.art.ca probe development experimental setup grass stimulator CW4 optical probe motion sensors strain gauge belt pulse oximeter on a toe Data Analysis Tools 2 1 1 3 5 5 4 17 9 18 10 7 6 11 3 4 6 8 10 2 8 19 20 21 9 24 12 22 12 26 13 29 14 30 7 14 15 31 16 32 11 23 25 27 13 28 HOMER filters raw data stimuli onsets probe geometry block average of the active stimuli Data analysis - Homer© HbO and HbR time traces Imaging average of multiple trials hemodynamic evoked response of the sensorimotor cortex during active and passive tasks passive stimuli give a weaker and smaller activation passive stimuli are more controlled and less prone to motion artifacts the study demonstrates the capability of DOI to detect the hemodynamic evoked responses to voluntary and non-voluntary stimuli in the sensorimotor cortex M. A. Franceschini et al., Psychophysiology 40, 548 (2003) Motor-Sensory Stimuli electrical median nerve finger tactile oxy maps finger opposition 2 9 14 4 15 6 10 11 5 6 1 3 1 detectors sources (690 & 830 nm) 8 12 7 8 5 1.9 cm x 2 16 -0.80 0.80 0.0 HbO (M) -0.50 0.0 HbO (M) 0.50 -0.40 0.0 Hb (M) 0.40 -0.25 0.0 Hb (M) 0.25 -0.25 0.0 HbO (M) 0.25 0.0 Hb (M) 0.12 7 3 cm 3 4 deoxy maps 13 Franceschini et al, Psychophysiology, 40:548 (2003). -0.12 Motor-Sensory Stimuli hemoglobin changes (M) right hand finger opposition 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 right hand tactile 0.2 0.2 0.1 0.1 (a) 0 0 -0.1 -0.2 left hand finger opposition 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -10 right hand electrical 0.3 0.3 -0.1 left hand tactile 0.2 left hand electrical 0.2 0.1 0.1 (b) 0 0 -0.1 0 10 20 time (s) 30 40 -0.2 -10 0 10 20 time (s) HbO contralateral HbO ipsilateral 30 40 -0.1 -10 0 10 20 time (s) Hb contralateral Hb ipsilateral 30 40 FAST SCATTERING SIGNAL noninvasive optical measurement of direct neural activation (time scale of milliseconds) How? M.A. Franceschini and D.A. Boas, NeuroImage, 2004 event-related optical signal G. Gratton et al. (1995) Shades of gray matter: non-invasive optical images of human brain responses during visual stimulation. Psychophysiology 32, 505–509 Event-related optical signal (EROS) recorded from medial occipital cortical area during a visual stimulation paradigm. (b) EROS effect. Filled circles refer to the EROS activity (average of three subjects) recorded from predicted locations for each quadrant stimulation condition, and open circles to the EROS activity recorded from the same locations when the opposite quadrant is stimulated. we want to measure the fast signal with CW MonteCarlo simulation of light propagation through the head absorbing perturbation scattering perturbation logarithmic scale blue = decrease in intensity red= increase in intensity scattering perturbation 0 5 I/I0 (%) -5 0 0 -0.05 -0.005 -0.1 -0.01 -0.15 -0.015 -0.2 -0.02 -1 0 1 2 3 4 5 pixels 1 pixel = 2 mm a scattering perturbation in the cortex that causes a 0.1% change in intensity will cause a phase shift of 0.008 deg (@ a modulation frequency of 100 MHz) phase shift (deg) intensity vs. phase filtering out the heart 1 0 -1 -2 0 50 2 I/I0 (%) I/I0 (%) 2 100 time (sec) 150 1 0 -1 -2 50 7099 90 110 130 70 80 9090 110 120 130 60 70 80 110 120 130 140 90 94 96 98 102 104 106 108 150 110 98 80 90 101 110 102 120 93 95 97 99 100 103 105 107 96 92 98 100101 102 104 time (sec) 200 Filtering out the heart (b) power spectra relative signal change (a) time (sec) raw data heart fit frequency (Hz) subtraction of the two signals Apparent Fast Signal I/I0 (%) 0.02 raw data odd filtered 0.01 0.00 -0.01 even filtered heart filtered data -0.02 -0.1 0 0.1 0.2 time (sec) 0.3 0.4 criteria to assess the fast signal electrical median nerve stimulation ~4.0 Hz 0.06 0.06 0.04 0.04 830 nm 0.02 I/I0 (%) I/I0 (%) finger tapping self-paced~3.2 Hz 0.00 -0.02 -0.04 0 I/I0 (%) 0.1 0.2 time (sec) 0.3 0.4 -0.02 -0.1 tactile stimulation ~3.9 Hz 0.03 stim. on 0.1 0.2 time (sec) 0.3 0.4 0.020 0.015 odd stim. 0.01 0 electrical median nerve stimulation ~3.7 Hz all the stim. 0.02 0.00 ipsilateral hand 0.010 0.005 0.000 -0.005 -0.01 -0.02 -0.010 even stim. -0.03 contralateral hand -0.015 -0.020 -0.04 -0.1 0.00 -0.06 I/I0 (%) 0.04 0.02 -0.04 690 nm -0.06 -0.1 stim. off 0 0.1 0.2 time (sec) 0.3 0.4 -0.1 0 0.1 0.2 time (sec) 0.3 0.4 0.04 0.04 0.02 0.02 830 nm 0.00 I/I0 (%) I/I0 (%) fast signal during finger tapping -0.02 -0.04 690 nm -0.06 0.00 -0.02 -0.04 -0.06 stim. on -0.08 -0.08 -0.1 stim. off 0 0.1 0.2 0.3 -0.1 0.4 0 time (sec) 0.04 0.02 odd stim. -0.02 -0.04 -0.06 -0.1 0 0.1 0.2 time (sec) 0.4 0.3 ipsilateral hand 0.00 -0.02 -0.04 -0.06 even stim. -0.08 0.3 0.04 I/I0 (%) I/I0 (%) 0.00 0.2 time (sec) all the stim. 0.02 0.1 contralateral hand -0.08 0.4 -0.1 0 0.1 0.2 time (sec) 0.3 0.4 5 lhtap localization tactile stimulation contral. 0 25 50 75 100 125 0 0.1 150 175 200 225 ms RH LH contral. -0.1 I/I0 @ 830nm (%) 5 pas Scattering or Absorption Contrast absorbing perturbation scattering perturbation (c) 16 15 6 14 4 13 2 12 8 11 7 10 6 7 5 3 1 1.9cm 8 3cm (b) 4 3 2 logarithmic scale 1 9 5 detectors sources (690 & 830nm) overlapping measurements to improve fast signal detection overlapping measurements increase spatial uniformity improving detection rate!!! 1st neighbor overlapping scattering inclusion 1.5cm deep with 1st neighbor measurements spatial uniformity is poor and the detection of the fast signal strongly depends on the position of the scattering change relative to sources and detectors Current Imaging Issues • Spatial uniformity – More dense, overlapping measurements • Physiological signal clutter – cardiac, respiration, blood pressure PCA, ICA, physiological modeling • Depth penetration and sensitivity signal processing to filter physiological clutter from brain activation signals our raw optical data show ~1Hz cardiac pulsation, ~0.1-0.3 Hz blood pressure and respiration fluctuations optical signal OD blood pressure respiration time (sec) cross-correlation of the optical data with respiration blood pressure cross-correlation coefficient heart beat time (sec) a random optode in the head spatial-temporal covariance 2 1 1 3 5 5 OD 3 4 9 18 10 7 19 20 21 6 11 17 4 6 8 10 2 8 9 24 12 22 12 26 13 29 14 30 7 14 15 31 16 11 32 time (sec) 23 25 27 13 28 cross-correlation of the optical data with heart beat -0.1 0 ×0.5 respiration 0.1 -0.2 0 ×1 blood pressure 0.2 10 frames per second -0.7 0 0.7 ×5 we want to use these maps to spatially filter systemic signal clutter spatial-temporal map of brain activation OD @ 830 nm during finger tapping: 10 sec stim and 20 sec rest 2 1 1 3 5 5 2 4 10 17 10 20 24 12 13 29 30 7 30 14 15 31 16 32 11 22 12 26 14 0 -5 19 21 9 7 9 18 7 6 11 3 4 6 8 8 raw data with temporal filtering – data band pass filtered between 0.02 and 0.8 Hz 23 25 27 13 28 PCA using baseline spatio-temporal info we used a principle component analysis on baseline data to identify the major component of spatial covariance and projected out this spatial component on the stimulus data raw data raw data PCA filter PCA cross-correlation optical signal and physiology vs. PCA maps I principal component II principal component CC with blood pressure CC with respiration Blood Pressure CC~0.5 Respiration CC~0.1 Systemic physiology has strong fluctuations in the scalp Our CW optical measurements are most sensitive to the scalp Can we reduce sensitivity to the scalp? Increasing Depth Penetration and Depth Resolution CW 1 ns 2 ns Time Domain measurements will give us: 1) Better depth resolution 2) Better depth penetration (CNR vs CBR) TD measurement of brain activity 3 cm 3 cm 7 Hz acquisition 2 cm CNR 19 to 25 1 cm CBR 8.9 to 6.8 Time (ns) DEPTH DISCRIMINATION WILL FURTHER INCREASE CBR! 0ns 0.5ns 1ns 1.5ns Time (s) 2ns 2.5ns 3ns CW Multi-Modality Imaging Why? – combine complimentary information Why optical? – unique functional information but inferior structural information Why combine DOT and MRI / X-Ray / US? 1. Benefit from MRI/X-Ray spatial and Optical spectroscopy - Structure and Function 2. Cerebral Metabolic Rate of Oxygen (CMRO2), tumor oxygen consumption Clinical utility of combine DOT and MRI / X-Ray / US? 1. Stroke: MRI (T1,DWI,PWI,BOLD) – DOT adds quant hemo and CMRO2 2. Neuro-Degenerative Brain Diseases – DOT brings more functional info 3. Breast – DOT adds functional overlay to X-Ray structural screening & diagnosis 4. Breast – with MRI could constrain DOT and determine tumor oxygen consumption Calculation of CMRO2 CMRO 2 CBF SaO 2 SvO 2 CBF OEF SaO 2 Assume SaO2=1 CMRO 2 CBF HbR HbT 1 1 1 1 R T CMRO HbR o HbTo 2,O CBFO ASL [HbO2]V BOLD [HbO2]A O2 Optical 1 NIRS and fMRI of CMRO2 Hoge, et al., Presented at the meeting of Human Brain Mapping 2003., submitted Simultaneous DOT and fMRI 785nm 830nm BOLD A D Strangman, et al. NeuroImage 17: 719-731. 2002. temporal comparison BOLD & DOT 12.5 1.0 HbO HbO HbT HbT [HbR] [HbR] 0.100 10 [HbT] [HbT] BOLD BOLD 8 10.0 0.080 6 7.5 0.060 0.6 5.0 0.040 BOLD 0.4 2.5 4 2 0.020 BOLD 0.0 0.2 0 0.000 HbR HbR -2.5 0.0 0 1 2 3 4 5 5 66 -0.020 -2 77 88 99 time (sec) Time (sec) Time(sec) 10 11 11 12 12 13 13 14 14 15 15 10 BOLD (%) 0.8 BOLD (%) Normalized Change (AU) (au) change normalized (M) hemoglobin Concentration( M) hemodynamic evoked response measured with DOT and fMRI during event-related finger tapping[HbO] stimulation [HbO] temporal comparison BOLD & DOT change (au) normalized Normalized Change (AU) hemodynamic evoked response measured with DOT and fMRI during event-related finger tapping stimulation Normalized 1/-BOLD (AU) HbT 0.8 0.6 BOLD 0.4 0.2 0.0 0 HbR:(1/BOLD) 1 2 3 4 Hb R6 5 7 8 time (sec) 9 10 11 12 13 14 15 HbT:(1/BOLD) 1.2 Time(sec) 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.2 HbO 1.0 1.0 Normalized 1/BOLD (AU) 1.2 [HbO] [HbR] [HbT] BOLD 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 Normalized -HbR (AU) 0.8 1.0 1.2 -0.2 -0.2 0.0 0.2 0.4 0.6 0.8 Normalized -HbT (AU) 1.0 1.2 temporal comparison ASL & DOT HbR:ASL 1.2 HbT:ASL 1.0 1.0 0.8 0.6 0.6 0.4 ASL 0.2 0.0 -0.2 Normalized ASL (AU) Normalized ASL (AU) 0.8 0.4 0.2 0.0 -0.2 -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 -0.8 -0.8 1.2 -0.6 -0.4 Normalized HbR (AU) 1.2 HbR:(1/BOLD) 0.2 0.4 0.6 0.8 1.0 HbT:(1/BOLD) 1.2 1.0 0.8 BOLD 0.6 0.4 0.2 0.0 Normalized 1/BOLD (AU) Normalized 1/-BOLD (AU) 0.0 Normalized HbT (AU) 1.0 -0.2 -0.2 -0.2 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 Normalized -HbR (AU) 0.8 1.0 1.2 -0.2 -0.2 0.0 0.2 0.4 0.6 0.8 Normalized -HbT (AU) 1.0 1.2 spatial correlation DOT-fMRI 5 8 4 7 3 6 2 1 4 3 2 BOLD image projected on the surface of the head 8 3 6 5 2 4 7 1 6 3 4 3 2 2 1 5 2 1 8 2 4 7 6 5 4 3 3 2 1 2 1 DOT MRI Tissue Segmentation to guide DOT Tissue Parameter Estimation Using Multi-Spectral 3-D Protocol o FA = 3 o FA = 5 MRI Forward Model (FLASH/SSC): o FA = 30 1 e TR / T1 TE / T2 I (TR, TE, , T1 , T2 , P) P sin ( ) e TR / T1 1 cos e Dale and Fischl Inferring Tissue Optical Properties Infer Homogeneous Medium • Bayesian inference • Measurement and model noise • 8 Temporal Point Spread Functions Infer Segmented Medium Barnett et al, Applied Optics, 42:3095 (2003). cortically constrained DOT of brain activation localization of true simulated activation relative to optodes 1st neighbor planar reconstruction 1st & 2nd neighbor planar reconstruction coronal slice of the 1st neighbor reconstruction 1st & 2nd neighbor anatomical MRI segmented with anatomical cortical reconstruction with human head constrain anatomical cortical constrain Summary • NIRS Imaging of Brain Function is rapidly evolving to improve spatial resolution and depth penetration • Multi-modality integration enables quantitative imaging and estimate of CMRO2 • Explore Neuro-Metabolic-Vascular relationship • Neuroscience and Clinical Applications photon migration imaging lab http://www.nmr.mgh.harvard.edu/DOT/ Andy Siegel Maria Angela Franceschini Juliette Selb Ted Huppert Danny Joseph Jon Stott Sol Diamond Rick Hoge Anna Custo Yiheng Zhang Gary Strangman Anders Dale