Numerous studies of the diffusion of water in brain tissue and other biological systems have documented a non-monoexponential behavior of the MR signal as a function of the b-value at fixed diffusion times. High-quality multi-b Diffusion Weighted Images (DWI) are required to guarantee the accurate outcomes of the non-monoexponential models, like bi-exponential [1], stretched-exponential [2], and statistical model [3]. Getting high-quality multi-b DWIs, especially with high b- values, in the clinically acceptable scan time has remained a challenge. Increasing the number of acquisitions (NSA) will improve the signal to noise ratio (SNR), however, it will expend the scan time as well. In this study, instead of increasing the NSA, we developed a novel methodology to improve the quality of the multi-b DWIs using the intrinsic multi-exponential pattern. It was evaluated on a healthy brain and compared with the mono-exponential model. In addition, its potential value of improving the robustness of IVIM was also evaluated. The study was performed using a 3.0-T system (Ingenia, Philips Medical Systems, Best, the Netherlands) equipped with a dedicated 32-channel phased-array head coil. Diffusion-weighted images were acquired using the following parameters: 20 axial slices, FOV 240 x 240 mm, matrix 128 x 128, thickness/gap 5/1.5 mm, TE/TR 125/3314 ms. Eleven b-values included 0, 25, 50, 100, 200, 400, 800, 1200, 2000, 3000, 4000 s/mm2 were measured with three mutually orthogonal diffusion encoded directions plus two acquisitions. Generally, the diffusion-attenuated MR signal can be expressed as a multi-exponential function given by (1): S(b) = Σy_iexp(-bD_i) (1), where S is DWI signal intensity, b is the b-value, y is the desired signal intensity with specific diffusion coefficient D and n is the number of diffusion components. All the calculations and programs were done in Matlab R2012b. First, y was obtained by solving the Eq. (1) using nonnegative least square algorithm [4] with n=100 and D exponentially equispaced from 0.001 to 100 μm2/ms. Then, the new computed DWI (cDWI) with any specified b-value could be calculated using the Eq. (1) with the known y. For comparison, the mono-exponential (n=1) was also performed, and one ROI with 8*8 pixels was drawn in the white matter (see the red square in Fig. 3) to evaluate the errors of the fitting results. In addition, we compared the calculated IVIM parameters using the raw DWIs and the multi-exponential cDWIs as well.Fig.1 shows the raw DWIs (the 1st row), the cDWIs using multi-exponetial model (the 2nd row), and mono-exponential model (the 3rd row). Obviously, the multi-exponetial model restored the contrasts of the raw DWIs, while the mono-exponential model altered the contrasts at some b-values (0, 800, 1200, 2000). The fitting results using these two different models in the white matter are shown in Fig. 2. The result of the multi-exponential model (χ2 = 68.7) was much better than that of the mono-exponential model (χ2 = 8161.9). The SNR of cDWI with high b-value was higher than the raw DWI (see Fig. 3) at b-value 3000 s/mm2. The result of the IVIM parameter, perfusion fraction (F), was shown in Fig. 4. The values of F from the ROI (see the red square in Fig. 3) were 0.3108±0.0379 (raw DWIs) and 0.3067±0.0246 (multi-exponential cDWIs) respectively. In this preliminary study, the proposed multi-exponential cDWI method presented a good fitting result, and the image quality was obviously improved at high b-values (Fig. 3). Although the mono-exponential model could also increase the SNR at high b-values, it altered the contrasts of DWIs at some b-values in the meantime. In Fig. 4, the standard deviation in the ROI was decreased after multi-exponential model was applied, and the image looks smoother on the right side. The other two IVIM parameters, diffusion coefficient and pseudodiffusion coefficient, had the similar results (not listed here). Other non-monoexponential models need to be evaluated in the future work. In conclusion, the multi-exponential method can improve the image quality of the multi-b DWIs and may become an effective preprocessing way for the non-monoexponential models.