Long exam times in clinical MRI negatively affect the costs for the provider and payer, flexibility in scheduling, and patient experience and (possibly) compliance. Reconstruction-based approaches to fast imaging with sparse sampling (e.g. partial Fourier sampling, parallel imaging, compressed sensing) nearly always decrease scan times at the expense of decreased inherent SNR, which is sometimes mitigated by nonlinear constraints on the image reconstruction. Our lab has been building the infrastructure for comprehensive, robust spiral MRI (1, 2) of the brain, focusing on optimizing acquisition methods with long sampling duration τ. For spiral scans this allows one to decrease scan times with an accompanying increase in image SNR (3), at the expense of sensitivity to field inhomogeneities, gradient fidelity, and T2*. We have worked to address these challenges with the near-term goal of matching resolution and contrast of conventional images, with equal or better image IQ and reduced scan time. The long-term (future) goal of a 5 minute high quality brain exam is discussed below. The details of our work have been published elsewhere(4-8); this submission is intended to give an overview of the general progress of this initiative.Our sequences use 2D spiral or 3D Distributed Spiral Trajectories (9) (Fig. 1), all designed to have blurring only in 2 directions. Currently all data are fully sampled, including Cartesian comparisons, reconstruction is linear with no image filtering. Images were collected on normal volunteers on a Philips Ingenia 3T scanner. K-space trajectories are corrected using system pre-emphasis and measured system delays only (no trajectory measurements used). Reconstruction uses standard gridding (10) and sampling density correction (11, 12). Off-resonance deblurring is achieved by solving the blurring matrix equation Ax=b using a Conjugate Gradient approach, where x contains the unknown water and fat images, b contains 2-3 images with different TE’s, and A is a blurring matrix, which is implemented by spatially-varying convolution with appropriate blurring kernels used in conjunction with a field map collected at the beginning of each exam with a fast scan (13). Currently sampling durations (τ) are on the order of 10msec. Care has been taken so that reconstruction infrastructure is shared across all types of scans, is fully automatic, and runs on the scanner. All spiral scans except FLAIR TSE (with chemical fat saturation) include multiple TE scans for fat/water separation, while none of the Cartesian scans include this feature.Figures 2-4 show comparative images of normal volunteers between spiral and Cartesian (conventional) approaches to gradient echo, spin echo, and time-of-flight MR Angiography scans. For spiral, the water-only images are shown. All data reflect fully sampled (R=1) data sets with no filtering and linear reconstruction. In general, spiral images have much less Gibbs ringing and flow-related artifacts than conventional images. Spiral images suffer most in the nasopharynx region and some areas around the skull, where accurate field maps can be difficult to generate. In the brain, the most difficult region for both spiral and Cartesian images is in the inferior frontal lobe where there is rapid change in B0; even here there is generally similar image quality. Figure 5 shows reconstruction times. Online reconstruction, including full deblurring, is < 1 sec/slice for full (e.g. 320²) matrices.Our long term goal is to achieve the fastest high quality scans possible, based on simple analyses with τ = 20ms to limit T2* decay effects. We calculate that - if all engineering challenges are met - high resolution brain scans (0.6 mm in-plane x 3mm thick contiguous slices) covering the whole brain with SNR > 20 can be obtained, using our current hardware (Philips Ingenia 3T), in roughly 30 seconds per scan (excluding DWI, and slightly longer for FLAIR). This would allow adoption of a 5 minute complete, high quality brain exam using spiral MRI. Scan times for both Cartesian and spiral images can be reduced using parallel imaging and other reduced-data approaches; we are currently developing and assessing post-Cartesian parallel imaging approaches for the next phase. Given the inherent speed of Spiral MR, we believe only modest parallel imaging (R < 3 at most, depending on the sequence) will be necessary to achieve the desired times (and stay above the SNR limit of 20). We also are developing enhanced deblurring methods that will allow us to robustly extend tau, and further improve SNR efficiency.