In recent years there has been an emphasis on the use of increasingly high field strengths in functional MR imaging. While there are undoubtedly benefits to adoption of higher field strength instruments, quantitative data demonstrating their advantages is essential to justify the increased cost and complexity. Conversely, an understanding of the capabilities and limitations of lower field systems such as 1.5T is important to guide appropriate utilization of the large installed base of clinical 1.5T systems.
The overall objective of this paper is the characterization of sensitivity and specificity of gradient-echo BOLD functional MRI at 1.5T and 3T over a range of echo-times, and to compare their performance at each field strength. We study the sensitivity and specificity of BOLD using a controlled global stimulus (hypercapnia). In particular, we aim to identify the optimal echo-time (TE) at two different magnetic field strengths in areas of cortical gray matter and resolvable veins, in order to maximize specificity to cortex or maximize sensitivity (by avoiding the contribution of the vascular component). Because we compare the BOLD responses at different field strengths and across different scanning sessions, we introduce and define a new quantitative index of BOLD sensitivity, the Flow Relaxation Coefficient (FRC), using flow-responses induced by CO2 inhalation (i.e. direct Cerebral Blood Flow measurements) to control for inter-session variations and ensure that the manipulation is equivalent across imaging sessions. Given the broad availability of both 1.5T and 3T systems and the prevalence of gradient-echo BOLD fMRI at these field strengths, we have focused on these field strengths.
Since increases in susceptibility effects that occur with increased field strength affect both the functional responses 
and physiological noise 
, it is important to determine the net increase in the contrast-to-noise ratio (CNR) that occurs with field strength as it is this ratio that is the ultimate determinant of sensitivity in functional MRI experiments. The CNR increases in, for example, visual stimulation experiments have been compared for different field strengths 
, and have also been examined for motor activation 
. Poser and Norris 
have investigated the sensitivity of BOLD imaging at 7T by measuring and combining responses at different echo-times, while Olman et al. 
have compared the sensitivity of spin-echo BOLD imaging at 3T and 7T.
In addition to sensitivity
, a second important criterion in functional imaging is specificity
. In BOLD fMRI the major challenge in achieving specificity is the large amplitude of responses in venous blood vessels that drain activated tissue regions. Venous responses generally exceed parenchymal responses by an appreciable factor 
. Most of the above literature has promoted the notion that higher fields offer better specificity against macro-venous responses. Spin-echo BOLD responses are also a subject of considerable interest for improving specificity, especially at ultra high field strengths (e.g. 7T and above), 
. Uludag et al. 
have described a comprehensive model of susceptibility-based MRI contrast, from micro and macro vasculature, which explains important differences between spin- and gradient-echo acquisitions at different field strengths and echo-times, based on simulations. The present study contributes systematically acquired data to explore the significance of venous responses at 3T and compare against the 1.5T field strength system to investigate further, whether this effect has increased by the field strength or eliminated.
A limitation of using sensory stimuli to characterize BOLD contrast, as done in the above studies, is that there is uncertainty in establishing the “ground truth” of where the activation actually occurs. In this respect, the use of hypercapnia as a reference condition is that, being a global manipulation, there is no need to identify “activated” tissue regions through statistical methods. This avoids some of the circularity that may arise when small, noisy activation signals are characterized by sampling regions localized using statistical methods based on those same signals. In the case of hypercapnia, all parenchymal gray matter and associated veins are “activated” in the sense of undergoing increased blood flow and the precise location of the region of interest analyzed is not important as long as voxels can be categorized into appropriate compartments (vascular, parenchymal). Furthermore, the global activation allows use of larger regions of interest without risk of selection bias. Using conventional task-activated definitions of analysis regions may result in the over-representation of veins due to the fact that veins generally provide the highest CNR activation. With the whole cortex uniformly activated, regions free of large veins can be readily selected to estimate the parenchymal response amplitude.
Previous investigators have used carbon dioxide (CO2
) inhalation to study BOLD responses in humans. Bandettini et al
normalized BOLD activation images by maps of CO2
-induced BOLD signal change in an attempt to attenuate large responses associated with veins 
. Davis et al
, Hoge et al
, and others have used hypercapnic calibration methods to estimate changes in the cerebral metabolic rate of O2
, and Corfield and collaborators have investigated the additivity of neuronal and global increases in BOLD signal 
. Other researchers 
have used hyperoxia to produce BOLD signal increases which can be calibrated by using the change in end-tidal O2
to estimate the venous O2
saturation. Cohen et al. 
have also used CO2
-induced ASL signals to normalize BOLD responses to neuronal activation for the purpose of improving comparison of results acquired on different scanning systems.
In addition to the magnetic field strength, the TE of the pulse sequence will also play a role in determining both sensitivity and specificity. Differentiation of the signal equation for a T2*-weighted acquisition shows that the maximum effect will be observed when the TE is equal to the T2* value of the tissue compartment in question. Since the T2* value of large veins is considerably shorter than that of gray matter, the choice of TE can be expected to play a role in determining both sensitivity and specificity.
An additional challenge to comparing BOLD responses at different field strengths using flow-responses induced by CO2 inhalation is to ensure that the manipulation is equivalent across imaging sessions and systems. Although breathing a fixed concentration of inspired CO2 offers advantages as a repeatable reference condition, it is still possible that the actual change in arterial CO2 may vary between sessions, since changes in breathing rate will affect the degree of hypercapnia achieved. To control for possible differences in the Cerebral Blood Flow (CBF) response achieved during the sessions on the different scanners, we embedded Arterial Spin-Labeling (ASL) based flow measurements in the relevant BOLD acquisitions. This allowed us to compare BOLD reactivity by using the ratio of the percent change in BOLD signal per unit of percent change in CBF signal. This ratio is likely to be a more invariant reflection of the BOLD sensitivity for probing different tissue compartments and field strengths. By calibrating the stimulus using CBF measurements and determining the BOLD response as a function of TE, we calculated the change in R2* (ΔR2*) per unit fractional flow change (Flow Relaxation Coefficient, FRC) for both field strengths (1.5T and 3T) and each component: parenchymal and major veins.
Following the notation used by Hoge and colleagues 
, we review the contributions to the percent BOLD response per unit fractional CBF change in response to inhaled CO2
. The transverse relaxation rate R2*
is assumed to be the sum of the de-oxyhemoglobin (dHb
) contribution, R2*|dHb
, and a relaxation rate term due to other sources, R2*|other
Given that the relationship between R2*
and Cerebral Blood Volume (CBV
) can be expressed as:
is dependent on the field strength and the sample under study, [dHb]v
is the venous de-oxyhemoglobin concentration and β
is a constant defined to have values between 1 and 2, also depending on the field strength and venous blood volume fraction within a voxel. The change in the transverse relaxation rate, ΔR2*|dHb
is expressed by:
The fractional BOLD signal response as a function of TE can be expressed as:
This expression (Eq. 4) can be approximated for small changes using the following linearization:
By calibrating the BOLD response using direct CBF measurements, Eq. 5 becomes:
We can then determine the Flow Relaxation Coefficient (FRC) as the change in R2*|dHb
) per unit fractional flow change: