1. Biophysics of BOLD and common image artefacts Rhodri Cusack Overview Biophysics of BOLD signal -Neural > Hemodynamic > MRI Complications at each level - and what to do about them

2. Stimulus to signal Arthurs & Boniface (2002)

4. What aspect of neuronal activity determines BOLD response?? LFP (Logothetis et al, 2001) –Simultaneous BOLD FMRI & electrophysiological recording –Measured Local Field Potentials (pre-synaptic, input) and Multi Unit Activity (spiking, output) –Both provided reasonable fit to BOLD activity, but LFPs better Energy use/oxygen consumption (Hoge et al, 1999) –Used MRI to measure CBF & oxygenation –Found linear relationship Neurotransmitter (Attwell & Iadecola, 2002) –Assessment of energy use by different processes Spiking very energy expensive Most used by postsynaptic currents & action potentials –Argue hemodynamic response is driven by neurotransmitter signalling and not local energy use Glutamate/GABA?

6. Hemodynamic response (HDR) Increase in volume –Increase in volume throughout system, from arterioles to capillaries, venules & veins –Fastest and greatest response in arterioles (Vanzetta, Hildenshen & Grinwald, 2005) Large increase in flow Oxygenation –Initial de-oxygenation in capillaries (Vanzetta et al, 2005) –Then, flow increase leads to a increase in oxygenation relative to the baseline state (i.e., OEF decreases, Ogawa et al, 1990; Bandettini et al, 1997)

7. Influences on spatial distribution of HDR Spatial characteristics influenced by –Neurovascular coupling (Malonek & Grinvald, 1995) –Vascular plumbing & structure Intracortical vessels Pial network Larger vessels –Size of region activated (Turner, 2002) Larger regions may give signal from veins further away

8. Influences on temporal profile of HDR Temporal characteristics influenced by –By neurovascular coupling in arterioles/capillaries –Flow times Function of vessel size Blood velocity proportional to radius – much of delay in small vessels –Mixing due to laminar flow within vessels (de Zwart, 2005) Should include diffusion (Parrot) –Other effects (e.g., vessel size dependence of post-stimulus adaptation – Mandeville et al, 1999; Yacoub et al, 2006)

9. The Balloon Model Buxton, Uludağ, Dubowitz, Liu (2004) Similar elements incorporated into SPM’s DCM forward model

11. Hemodynamic-MRI coupling HDR affects MRI signal through several mechanisms (Hoogenraad et al, 2001) –Extravascular: reduced field gradients around venules & veins Reduced deoxygenation causes magnetic property (susceptibility) of blood to become more similar to surrounding tissue Reduced field gradients in GM & CSF –Intravascular change in T2* –Reduced phase mismatch between signal from inside and outside venules & veins –Change in blood volume Affected by parameters of MR acqusition –Field strength (higher field – more sensitivity, particularly to smaller vessels/capillaries) (Haacke, 1994; Yacoub et al, 2003) –Gradient echo vs. spin echo – latter insensitive to large vessels – less signal, but more spatially specific (Lee et al, 1999; Zhao et al, 2006)

14. Non-linearities in the BOLD signal measured linear Signal 0.25 s0.5 s1 s2 s20 s time (s) Stimulus timing BOLD Response Brief stimuli produce larger responses than expected Degree of non-linearity varies across space Neural in origin Slide adapted from Bandettini; Birn, Saad & Bandettini (2001)

17. Vascular artefacts Vascular structure can affect signal at many levels Vascular signal –Larger in magnitude –More noisy Cusack et al (in preparation) Group level –Large variability in venous structure reduces chance of bias –May be more important in case studies or MVPA/ functional localiser studies

19. Fields should be flat MRI scanners apply a strong magnetic field –3 Tesla at the WBIC Ideally, field should be homogeneous –Easy to arrange when it is empty, but ruined as soon as a head is put in Different materials act to strengthen (para) or weaken (dia) magnetic fields With Shaihan Malik

20. 00.27-0.27 Field strength relative to 3 Tesla (parts per million)

21. Tackling inhomogeneneity artefacts Distortion –Optimise acquisition –Acquire fieldmaps & undistort –Distortion by movement interaction Put movement parameters into model Use Realign & unwarp (Andersson, 2001) Dropout –Optimise acquisition parameters (e.g., TE, slice orientation, voxel size) –Z-shimming –Use spin echo (Schwarzbauer) –Passive shimming (Wilson, Jezzard and colleagues; Cusack et al, 2004) –Choose the right subjects

22. Fieldmap undistortion: evaluation by eye StructuralEPI mean Undistorted EPI mean StructuralEPI mean Undistorted EPI mean Cusack, Brett & Osswald (2004)

23. Quantitative evaluation - shape match to structural scan - Measure similarity in shapes between structural scan and EPIs before and after undistortion - Use mutual information statistic Cusack, Brett & Osswald (2004)

24. Overlap across subjects Cusack, Brett & Osswald (2004)

25. Undistortion on data from Siemens Trio Shimming: apply corrective field gradients –Much better on modern machines –Higher order shims optimised automatically –Shims only optimised for volume being acquired in EPI Acquire smaller volumes Undistortion most important for regions near inhomogeneities

26. Comparative studies Watch out when comparing to other models –Field inhomogeneities vary dramatically Human vs. sphinx vs. supine

27. Passive shimming EPI BOLD Wilson & colleagues, FMRIB, Oxford; Cusack et al (2005)

28. Optimising parameters to reduce dropout From Rik Henson

29. Summary BOLD FMRI involves a complicated set of couplings –Be careful when interpreting effects, or comparing FMRI with other imaging modalities It can fail in many ways –Optimise acquisition and analysis –Perform proper quality control