Numerical simulation dramatically improves the understanding and development of new MR imaging methods. Along with the increased complexity of new imaging technique, pulse sequences and tissue quantification methods, there are increasing demands for developing dedicated software to accommodate essential simulation. One challenging simulation task involves simulating the realistic signals for tissue systems with complex multi-compartmental, multi-exchange nature. In the previous work, we have developed simulation software named as MRiLab for performing fast 3D MRI simulation by leveraging the power of modern GPU parallel computing techniques (1). In this work, we proposed an improved version of MRiLab and engaged our efforts for incorporating a generalized exchange tissue model to facilitate flexible and realistic MR signal simulation. This abstract was aimed to demonstrate the feasibility of using this new version of MRiLab for investigating multiple quantitative MR experiments that remain challenging for numerical simulation.1) Generalized Tissue Model: The generalized model consists of N_F free proton pools F all inter-connected by the magnetization exchange (MT) pathways, and N_B bound proton pools B exchanging with the free proton pools (Figure 1). The free proton pools are reserved to model compartments with measurable transverse magnetization (e.g., water, fat, solute proton exchange compounds), while the bound proton pools are utilized to model semi-solid tissue macromolecular content non-visible on standard MRI (e.g, myelin, muscle fibers, collagen). Responses of free and bound proton pools are simulated using the finite differential Bloch-McConnell equations in the rotating frame (2) and MT saturation formalism (3), respectively. 2) Software Implementation: The simulation core code for calculating the mathematical equation was implemented in C language for high simulation performance. Compute Unified Device Architecture (CUDA) and Open Multi-Processing (OpenMP) were used for GPU and multi-core CPU parallel simulation computation. A high interactive user interface is developed with Matlab Graphical User Interface Developing Environment for designing convenient simulation workflow. The current version of MRiLab is composed of a main simulation panel, accessory function panels, discrete Bloch-equation solving kernel, image reconstruction and analysis toolkit. 3) Experiment Evaluation: Three different tissue models were used to evaluate MRiLab comparing results of modeling with previously described results and with actual phantom imaging results. Model 1 is a standard two-pool MT model widely used for quantitative description of MT effect (3). Model 2 extends Model 1 by an exchanging free proton pool representing OH- hydroxyl groups used to model glycosaminoglycan CEST (gagCEST) imaging (4). Finally, Model 3 extends Model 1 by an additional free proton pool to model MT effect in the presence of fat tissue compartment.Figure 2 show results of simulation of MT acquisition protocol at 3T by using Model 1, followed by estimation of macromolecule proton fraction (MPF) by standard cross relaxation imaging (CRI) (5) and recently proposed modified CRI (mCRI) (6). Higher resolution digital model, underlying flip angle map (corresponding to the 8 channel Biot-Savart transmission) and corresponding MPF image is shown in Figure 2a. Figure 2b shows an example of simulated image and FA estimated from MRiLab-simulated actual flip angle (AFI) (7) images. Note a good agreement between nominal and AFI estimated flip angle maps. The estimated MPF maps in Figure 2c show a good agreement between the true values and the values obtained from mCRI method. Also notice in Figure 2d the bias estimation from CRI corresponds well with findings in in-vivo human brain imaging (6). Figure 3 shows MRiLab-simulated gagCEST Z-Spectra and gagCEST asymmetry plots by using Model 2. Figure 3a shows the gagCEST Z-spectrum between -4.0ppm and 4.0ppm. Figure 3b shows the gagCEST asymmetry curve (4,8) from 0ppm to 4.0ppm with increase of effect at around 1.0ppm reflecting the theoretical presence of hydroxyl proton CEST effect on water signal. Figure 4 shows measured and MRiLab-simulated MT ratio (MTR) images and corresponding ROI-averaged MTR values in agar/fat phantoms for several fat fractions and echo times. MTR for fat-free images does not show variations with echo time both in simulated and experimental images, while for non-zero fat fraction MTR demonstrates echo time dependence due to varying fat/water signal interference. Note MTR variations in each individual echo image despite the fact that same 2% agar solution was used in all objects. The numerical simulation theoretically predicted the MTR oscillation pattern at different echo time with different fat concentration which is validated by the experimental data.In this abstract, we demonstrated MRiLab, a generalized tissue exchange model based fast MRI numerical simulation software, is feasible and flexible for simulating accurate and realistic MR signal in multiple types of numerical MR experiments.