Quantitative cardiac NMR imaging, and more particularly T1 mapping has become a popular modality to characterize myocardial tissue. Among the most reliable sequences, one can cite MOLLI¹, SASHA² and SAPPHIRE³ acquisition schemes. They are all based on the acquisition of several images (8 to 11) during a single breath-hold at different time-points after application of inversion and/or saturation pulses. The longitudinal relaxation time is computed from these data-points using exponential models. Depending on the method, the acquisition time of a single slice varies between 9 and 17 heart-beats. In this work, we developed and validated a radial variant of the MOLLI acquisition (raMOLLI) that allows to significantly decrease the acquisition time down to 5 heart beats, while keeping high precision on T1 estimation due to a large number of acquired data-points along the T1 relaxation recovery curve. Insensitivity of measured T1 values to heart rate was also demonstrated with this sequence.Experiments were performed at 3T (Prisma, Siemens Healthcare). Figure 1 depicts the raMOLLI sequence diagram: after non-selective inversion, 5 shots of 80 radial spokes were acquired at end of diastole with a balanced steady state free precession (bSSFP) readout. A golden angle of 111,246° was imposed between 2 successive spokes. Sequence parameters were: TR = 2.76ms, BW = 1000 Hz/pix, nominal flip angle (FA) = 35°. For each shot, 10 different images were reconstructed using a view sharing method⁴. A KWIC filter was calculated in order to select k-space data used for the reconstruction of each image (Figure 2-a). Undersampled data were then reconstructed using a compressed sensing algorithm with total variation regularization⁵ (Figure 2-b). Bloch simulations were then used to fit the temporal evolution of the signal and obtain a T1 map⁶ (Figure 2-c). A dictionary approach using 700 values for T1 (between 300 and 3000ms) and 40 values for FA (between 3 and 50°) was implemented in order to decrease the post-processing time. Six Gd-DOTA doped agar phantoms with a broad T1 distribution were used to validate the method. True T1 values were assessed using a standard inversion recovery turbo spin echo (IR-TSE) sequence with 15 inversion times (from 100 to 9000ms). Simulated heart rate was varied between 50 and 85bpm. Four healthy volunteers were investigated using the raMOLLI sequence and a patient suffering with glycogen storage disease type 3 was scanned before and 10 min after injection of a Gd-DOTA contrast agent. For comparison, standard 3-3-5 MOLLI was also acquired with the following parameters: TE/TR = 1.25/2.5ms, FA = 35°, TImin = 100ms with 80ms increment.Figure 3-a represents the T1 map obtained using the raMOLLI sequence on the phantom. Figure 3-b shows the relative difference between estimated T1 and the true T1 measured with the IR-TSE sequence, as a function of heart rate. For all tubes, relative differences were less than 3% and did not depend on the heart rate. Figure 4 shows the T1 maps acquired with the MOLLI and raMOLLI sequences on a healthy volunteer. In the myocardium, T1 values were slightly higher with the raMOLLI processing, due to the Bloch simulations based post-processing that takes into account the T2 contribution during the b-SSFP echo trains. On the four volunteers, the mean T1 estimated in the myocardium was 1391±44ms with the raMOLLI sequence and 1205±33ms with the MOLLI. Mean standard deviations within the myocardium were 42±7ms and 45±18ms with the raMOLLI and MOLLI respectively. Figure 5 represents the T1 map of the patient suffering from glycogen storage disease type 3 obtained after injection of contrast agent with the raMOLLI sequence. The regions with decreased T1 values (arrows) correlated well with regions of late gadolinium enhancement depicted in high resolution T1 weighted images.The raMOLLI sequence offers several advantages compared to the other cardiac T1 mapping sequences. First, acquisition time is substantially reduced: which can either improve patient comfort, or allows the acquisition of extra slices without increasing scan time. Then, we demonstrated that contrary to the MOLLI sequence, there was no effect of heart-rate on estimated T1 values. Furthermore, as the center of k-space is sampled at each radial spoke, view sharing techniques could be used to reconstruct a large number of images along the T1 recovery curve, providing a high robustness of post-processing. Finally, radial acquisitions are immune to back-folding artifact when the FOV become smaller than the imaged object: this could be used to increase spatial resolution of the T1 maps by decreasing FOV while keeping the same matrix sizes for acquisition.