Inhomogeneous Magnetization Transfer (ihMT) is a novel MR contrast mechanism that has shown improved specificity for myelinated tissue as compared to conventional MT [1,2]. Recently, fundamental developments have led to theoretical modeling of the ihMT effect [3]. Such a model is a precious tool for improved understanding of the underlying mechanisms and to guide experimental optimization. Here we report on the use of forward modeling of ihMT within the framework of a steady-state ihMT gradient echo (GRE) sequence designed for rapid whole brain imaging [4-5]. Numerical simulations and experimental investigations are performed to optimize the ihMT sensitivity for various TRs, power levels and ihMT pulse-train duration.Model: The ihMT theory accounts for the dipolar-interaction associated with motion-restricted macromolecules in addition to the usual Zeeman interaction modeled in conventional MT [3]. The equations describing the time evolution of the free- and bound-pool Zeeman magnetizations (Mf and Mb) as well as the inverse spin temperature (β) associated with the dipolar order were numerically integrated in Matlab (The MathWorks Inc., Natick, MA, USA) as a function of tissue properties (e.g. dipolar relaxation time T_1D, and corresponding fraction f) for varying sequence parameters: TR, ihMT pulse-train duration and B_1RMS (over TR). Simulation of the ihMT ratio (ihMTR) and signal-to-noise ratio (SNR) obtained at the White Matter (WM) Ernst angle with a steady-state interleaved ihMT/spoiled GRE acquisitions were performed considering piecewise definition of elementary events (e.g. saturation, relaxation). Sequence Development: Following previous implementation of a 3D ihMT-GRE sequence based on a pulsed dual-frequency saturation scheme [5] (Δf=±7kHz, 0.5ms pulsewidth, 1ms pulse repetition time), refinements were made to improve flexibility and allow more efficient use of available RF power [6]: 1/ square-spiral center-out k-space trajectory; 2/ ihMT partial Fourier (PF) preparation capability (i.e. ihMT pulses only played at the center of the k-space); 3/ readout segmentation allowing for rapid scan time with long TRs (i.e. N k-space lines acquired after one single ihMT preparation module). MRI: Experiments were performed at both 1.5T and 3T (Siemens, Erlangen, Germany) on healthy volunteers. Two studies were considered: I/ varying TRs at constant average power for a reference B_1RMS=4.1μT (~80-90% SAR level under normal supervision mode for fully prepared k-space) and enhanced B_1RMS=5.4μT (using 50% ihMT PF, identical SAR level); II/ varying TRs for 6ms and 12ms ihMT pulse-train duration. Readout FAs were adjusted in proportion to the Ernst’ angle rule for WM tissue. Readout segmentation was used when compatible with timing parameters, with proper adjustment of readout FA to maintain constant Mf excitation within TR. Other readout parameters were identical to [5], except (2.5mm)³ isotropic resolution. Processing: A fully automatic pipeline [5] consisting of volume registration using SPM [7], automatic brain segmentation (WM, cortical Grey Matter – cGM) using Freesurfer [8], and WM tracts extraction (JHU atlas [9]), was used to measure average ihMTRs and SNR (mean_ROI/SD_noise) over selected brain areas.Simulations of the TR dependency of ihMTR and SNR indicated that a strong sensitivity boost may be obtained when using relatively long TRs as compared to TR=19ms, i.e. when concentrating the RF energy by using less frequent, more intense ihMT pulses (Figure 1). This was clearly evidenced experimentally (Figure 2) with ihMTR enhancement of 25-70% (depending on brain area and B_1RMS). Full simulation (Figure 1) and experimental datasets (Figure 3) both indicate that optimal TR settings tend to promote sparse and intense RF energy deposition. Interestingly, increasing average power for long TRs seems counter-productive, indicating inefficient usage of RF energy and opening new perspectives for SAR reduction, especially at high field. A fair agreement was observed between theory and experiment although the model tends to predict higher ihMTRs. Theoretical ihMTR enhancements of 70-110% (depending on B_1RMS and ihMT pulse-train duration) were predicted for WM (T_1D≈6ms, f≈0.65, adapted from [3]) when optimizing TR as compared to TR=19ms. Finally, in contrast with model prediction, experiments showed that the 12ms ihMT pulse train seems at least as efficient as the 6ms one. Overall this suggests that model refinements may be needed to improve fidelity with actual data and improve model predictions.Improvement of the ihMT sensitivity was enabled on the basis of the recently proposed theoretical model. An efficient RF energy deposition scheme has been demonstrated for relatively long TRs, leading to enhanced ihMT effect. Using improved settings, ihMTRs as high as 15-17% and 10-12% were measured in WM at 1.5T (Figure 4) and 3T (Figure 5), respectively. This work opens new perspectives for patient studies at clinical field strength and ihMT implementation at higher field strength (e.g. 7T).