Magnetic resonance fingerprinting (MRF) is a promising new approach for rapid quantitative imaging [1,2,3,4]. A gradient t-Blipped multi-slice acceleration (MSA) implementation for MRF has recently been proposed [5]. However, radiofrequency (RF) based phase encoding may work better. In this work we demonstrate an RF based MSA approach for radial sampled MRF experiments such as PnP-PTX [3,4] including a fully integrated online image reconstruction pipeline that creates both quantitative maps (T1, T2, PD and B₁⁺) and synthesized contrast weighted images (MP-RAGE, T1-TSE and T2-TSE).

Radial MSA works by cycling the relative excitation phase between the simultaneously acquired slices [6]. For example, when using a MSA factor of two, the sequence interleaves two different multi-band pulses. One pulse excites both slices with the same relative phase [0, 0] (blue lines in Fig. 1), the other flips the phase on the second slice [0, $$$\pi$$$] (yellow lines in Fig. 1). When a large number of alternating radial samples (spokes) are collected, the signal from the second slice is canceled out (Fig.1). This slice is revealed by negating the phase of every other spoke before regridding.

On the other hand, in a radial MRF acquisition, only a small number of spokes are available per time point [4]. In this situation, large gaps exist between neighboring spokes which result in significant slice aliasing (Fig. 2). One way to reduce the inter-spoke distance, without increasing the scan time, is to aggregate groups of subsequent time points into compressed samples [4]. However, to encode the T2 relaxation time robustly, the RF phase relation between subsequent excitations must follow a specific pattern. Some coherence pathways may not refocus if the phase evolution is disturbed [7], causing the T2 sensitivity of the fingerprint to deteriorate. Therefore, one has to tread carefully when inserting multi band RF pulses into a MRF sequence.

The PnP-PTX sequence was revised to avoid unwanted RF-spoiling effects by alternating the multi-band RF pulses between repetitions of the pulse train (Shots). The golden angle radial sampling scheme [8] was used to evenly distribute the spokes acquired with different RF pulse configurations (Fig 3). Using 8 shots and a compression factor of 15, a total of 120 spokes per compressed time point were collected. A more detailed description of the base sequence can be found in [5].

Phantom and brain images were acquired using a standard 20-channel head-neck receive coil on a clinical 3 Tesla system (Siemens Magnetom Skyra, Erlangen, Germany). PnP-PTX sequence parameters were: 240x240 matrix, 1.0x1.0mm² in-plane resolution, 5mm slice thickness, 8 shots. The total scan time was kept constant for each measurement (~5 min), but the number of slices was doubled when using an MSA factor of 2. The online reconstruction pipeline including dictionary matching and synthetic image generation was implemented in C++ and integrated into the scanner platform. The study was approved by our institutional review board (IRB), and written informed consent was obtained prior to the examination.

Figure 4A shows a slice through the multi-compartment phantom obtained without (left) and with MSA (right). Quantitative analysis, comparing the measured T1, T2, and PD values in each compartment, showed that the accuracy of the multi-parametric maps remains similar (Fig. 4B). In the brain, an MSA factor of two enables full brain coverage with significantly thinner slices in equivalent time (Fig. 5). The online reconstruction took ~20sec per slice utilizing only the default computational resources available on the scanner. All four quantitative maps (T1, T2, PD, B₁⁺) and three synthetic contrast weighted images (T1-TSE, T2-TSE, MP-RAGE) for each of the 24 slices (MSA = 2) were reconstructed on the scanner in ~8 min.Faster acquisition strategies are always desirable in MR since they allow either higher resolution, larger coverage, or a reduced scan time. In the context of MRF, MSA combined with an integrated online reconstruction process is particularly appealing because it makes it easier to explore the benefits of MRF in a translational research setting. The current reconstruction process is still slower than most traditional imaging approaches, however it provides the operator with feedback at the console and demonstrates the feasibility of an online MRF reconstruction pipeline using currently available scanner hardware. Further optimization of the code and hardware are expected to bring further advances in scan time reduction and improve the image reconstruction time.