Fast fMRI acquisition methods enable the straightforward removal of unaliased physiological noise, and moreover the monitoring of fast functional activation. Whole brain fMRI with TR as low as 100ms¹ and nominal isotropic resolution of 3mm has been achieved by single shot imaging with highly undersampled three-dimensional non-cartesian trajectories combined with multi-coil arrays. However, these methods suffer from off-resonance artifacts due to a long read-out, and undersampling artifacts which can be removed by regularized reconstruction. We investigated a segmented version of the spherical stack-of-spirals trajectory that retains highly efficient data sampling and signal recovery, but grants more flexibility in choice of TE and TR, and provides options for sampling of more data points per shot at the expense of temporal resolution.A spherical stack of spirals trajectory¹ with slow encoding in $$$k_z$$$ direction was rotated by $$$2\pi / N_i$$$ per shot about the $$$k_z$$$-axis to create $$$N_i$$$ interleaves, as shown schematically in fig. 1. The trajectories were designed for nominal isotropic resolutions of 3mm and 2.25mm using 3- and 4-fold segmentation, respectively, a field of view of 192x192x144mm and a TE of 29ms. TRs of each interleave were 65ms including gradient spoilers, adding up to total TRs of 195ms (3 mm) and 260ms (2.25 mm). Spiral spacing in z-direction was 1.5 times the Nyquist spacing, linearly increasing to 3 in the case of 2.25mm. The sampling density of the spiral interleaves ranged within [5,10] and [9,18] for 3 and 4 interleaves, respectively, depending on the distance to the k-space center. The flip angle was set to 18 degrees, the Ernst angle for the chosen TR. The images were acquired on a 3T PRISMA scanner (Siemens, Erlangen, Germany) with a 64 channel head/neck coil. Images were reconstructed iteratively using a conjugate gradient method, modeling the signal by a nuFFT with a grid size matching the nominal resolutions, without necessity of regularization using the parameters given above. Before reconstruction, the data of the time series was corrected for global off-resonances². The activation images were obtained by GLM of data acquired during a flickering checkerboard block-design experiment. tSNR values for the two acquisition types were obtained by voxel-wise division of mean value by standard deviation of the time series.Fig. 2 shows every 3^rd/4^th slice of the first frame of a 3mm/2.25mm acquisition on the top/bottom row, respectively. Fig. 3 shows the corresponding tSNR maps. Note that areas depicted in cyan still exhibit tSNR values of around 40, and not the whole range of values up to the maxima of 270/185 for 3mm/2.25mm, respectively, is shown. Fig. 4 depicts an exemplary time series of an activated voxel from the visual cortex at 3mm.As previously shown¹, a slow encoding direction pointing in the direction of an off-resonance gradient can reduce blurring and drop out of signal compared to other types of trajectories, but leads to signal shift. This effect can be observed in fig. 2 especially above the sinuses, and is dependent on the speed of the slowest encoding. As in the case of the used 2.25mm resolution trajectory the encoding in the k-space center is slower in the z-direction compared to the 3mm trajectory. Slightly increased distortion and signal drop-out is observed, which may be due to the increased number of interleaves in the presence of physiological noise. One additional shot and a voxel volume of a fraction of 0.42 compared to 3mm resolution leads to a tSNR ratio of approximately 0.7 for 2.25mm resolution. This shows that physiological noise is enhanced less than thermal noise when reducing voxel size, as described for non-segmented EPI trajectories³. This holds regardless of the higher segmentation of the trajectory.The benefit of segmented non-cartesian trajectories for fast fMRI has been shown. The interleaved version of the spherical stack-of-spirals trajectory allows a faster read out along the slow encoding direction and therefore shows reduced off-resonance artifacts in comparison to the longer single-shot version, while data acquisition becomes more flexible. Physiological artifacts need to be quantified. However, first results show that TR is low enough for 4-fold segmentation with good tSNR behaviour. Interleaved spirals might lead to blurring, which has to be investigated for this type of trajectory, but is likely to be reduced because of shorter read-outs and therefore decreased off-resonance. As we deal with a steady state sequence, necessity for variable flip angles is not of concern. In future work we will investigate other trajectories and segmentation patterns in order to further improve image quality.