To explore spatial variations in temporal signal-to-noise ratio (tSNR) and echo time (TE_local) and develop the most reliable metric for BOLD sensitivity estimation.Estimating BS is useful for fMRI studies in selecting the sequence (simultaneous-multi-slice, 3D or slice-by-slice EPI) and optimizing parameters^1–3, but also in the assessment of the reliability of the functional results (considering false negatives and positives)⁴. BS is commonly estimated using two different metrics: 1) the product of the local effective echo time and the image signal (TE_local∙S)^2,5 or 2) tSNR^3,6. We explore the pitfalls of each metric, suggest TE_local∙tSNR as a more reliable measure of BS and investigate all three approaches at 7T in healthy volunteers and tumor patients, where prominent regional variations are expected.Gradients in B₀, induced by tissue-air susceptibility differences, introduce intravoxel dephasing and signal loss, but also local modifications of the effective echo time through shifts in the k-space signal⁷. If TE_local is reduced by the gradients, image signal (S) increases, but BS decreases, as BOLD contrast has less time to develop. In such regions BS metric 2 (tSNR) overestimates BS. If, on the other hand, the TE_local is increased, signal is reduced, but BOLD contrast increases. Here tSNR underestimates BS unless TE_local is so long that the signal is not acquired. At this point, signal, tSNR and BS suddenly fall to zero (type 2 signal loss). Metric 1 (TE_local∙S) considers both signal and TE_local changes, but does not identify dynamic physiological and technical noise sources, which are included by substituting S by tSNR. We, thus, suggest using a third metric: TE_local∙tSNR as a more reliable BS measure.Measurements were carried out with a 7T Siemens scanner and a 32-channel head coil. Field maps (multi-echo GE, TEs=[5,10,16]ms, TR=400ms, 1.7x1.7x3.0mm³, bandwidth=540Hz/pix) and a 5 min long resting state scan (EPI, TE=22ms, TR=2000ms, 1.7x1.7x3.0mm³, bandwidth=1447Hz/pix) were acquired for 3 volunteers and 11 tumor patients. Field maps were calculated using the Hermitian inner product⁸. TE_local maps were derived from field gradient according to Ref.⁹ and normalized to TE=22ms. Voxels with type 2 signal loss (TE>47ms) were marked white in TE_local maps. EPI runs were coregistered to GE scans, distortion corrected, and mean signal S and tSNR were calculated.

Fig. 1 shows normalized TE_local and two BS maps calculated according to metric 1 and 2 for a representative healthy subject. BS metric 1 shows high values in CSF and gray matter and low values in white matter (Fig.1c), whereas metric 2 shows rather the opposite: higher values in homogeneous white matter (Fig.1d). This does not, however, include variations due to changes in TE_local (Fig.1b). Since metric 1 does not consider temporal variations in signal, as tSNR does, it was omitted from further consideration.

Variations in TE_local and the differences in BS metric 2 and 3 representative for healthy volunteers are presented in Fig. 2. In a ventral slice (number 3) tSNR values are highest (about 70) close to sinuses and ear canals (Fig.2e, see arrows), where TE_local is strongly reduced (Fig.2d, by about 50%). The reduction in BS due to short TE is apparent in TE_local∙tSNR maps (reducing BS values from 70 to 35 in Fig.2f at the arrow positions). TE_local is also reduced by up to 40% in the basal ganglia (Fig.2d, slice 12, see arrows). Ventrally, close to the ear canals and dorsally in frontal lobes the TE_local increases strongly leading to type 2 signal loss (Fig.2d and f, white regions).

Results in the four tumor patients with the strongest gradients in B₀ are presented in Fig. 3. In all cases large regions with strongly reduced TE_local (by about 50%) but relatively high tSNR were found in the proximity of pathologies (see arrows). For patient 1 and 2 this occurred close to primary motor cortex, for patient 3 near Broca’s area and patient 4 near auditory cortex. In BS metric 3 the tSNR values are readjusted accordingly (Fig.2f).

We have shown that tSNR maps may indicate sufficient BS to detect activation but that BS may be dramatically reduced by changes to TE_local. This effect was prominent at 7T in patients with large oedema, postoperative cavities close to vital brain functions (motor, speech, auditory). Neglecting changes to TE_local can lead to false negative results in clinical fMRI being hard to detect, because underlying EPI and tSNR indicate that there is adequate signal. We therefore advise that both, TE_local and tSNR, be considered in the assessment of bold sensitivity.