Diffusion-weighted imaging (DWI) characterizes the random microscopic motion of molecules and enables assessment of tissue microstructure. The fast attenuation of signal at low b values (0–100 s/mm2) has been shown to capture the perfusion information within the capillary network called Intra-voxel Incoherent Motion (IVIM) [1]. Using quantitative analysis of IVIM effect, both diffusion and perfusion component of tissue can be assessed separately; bi-exponential model has been widely used for the same [2]. Perfusion information available with IVIM is not very reliable for clinical interpretation. We implemented regularized parametric estimation for IVIM using two penalty function: a) Total Variation and b) Huber function with bi-exponential model in patients with Ewing sarcoma.

IVIM dataset from four patients (M:F=3:1, Age=25.3± 8.8 years), with Ewing sarcoma were acquired under the Institutional Review Board approved protocol. The acquisition were performed using 1.5T Philips Achieva MRI scanner with Spin Echo Planar imaging (SP-EPI) sequence with TE=66msec,TR= 1782msec, 5mm slice thickness and 144x144 matrix size. The DW images were acquired at 11 b-values (0, 10, 20, 30, 40, 50, 80, 100, 200, 400, 800 s/mm2). For one patient DWI images were acquired again after 2 cycles of chemotherapy. Thus in total five IVIM datasets were processed. Apparent diffusion coefficient (ADC) was estimated using mono-exponential fit for b≥ 200s/mm2. The IVIM datasets were analysed using four methods: i) Bi-Exponential (BE) model, ii) BE model with Total Variation (TV) penalty [3] (BE+TV), iii) BE model with Huber Penalty function [4] (BE+HPF) and iv) freeware Osirix. Three analysis methods, BE, BE+TV and BE+HPF, were implemented in an in-house built analysis toolbox implemented in MATLAB. An iterative optimization was performed using a nonlinear least square fitting algorithm along with the penalty functions.

Due to the nature of the data, the BE solution using simple optimization as well as that using Osirix package, might be highly noisy. In order to obtain physiologically meaningful solution we include gradient based penalties such as Total Variations [3] and Huber function [4] in our optimization model. Coefficient of determination (R2) was calculated to measure the goodness of fit for the four methods, using the in-vivo IVIM data and the best fitted signal with the model. R2 varies between 0 to1; values close to 0 being a poor and close to 1 a good model fit.

Figure 1 shows the R2 values calculated in-vivo using four methods. BE+TV and BE+HPF both methods had the highest R2=0.9± 0.02. One representative slice (b=800s/mm2) from a dataset is shown in Figure 2a and Figgure 2b shows the mean fitted signals by four methods to the tumour ROI. BE + TV and BE + HPF both outperformed other two methods in data fitting. Estimated parametric maps for Diffusion coefficient (D), Perfusion coefficient (D*), Perfusion fraction (f) for one representative patient are shown in Figure 3 BE+TV and BE+HPF, both the methods produced better and clinically interpretable images compared to other BE and Osirix freeware algorithm. Figure 4 shows the ADC, D, D*, f image using BE+TV method for both baseline and follow-up (chemotherapy). Both ADC and D in tumour demonstrated an increase in follow-up images. D* showed no change whereas perfusion fraction (f), demonstrated an increase in perfusion in tumour ROI after chemotherapy.IVIM analysis using Total Variation and Huber Penalty function methods showed a better fit to the IVIM dataset of Ewing sarcoma. Parametric maps (D, D* and f) estimated using BE+TV and BE+TV had less image noise and were clinically interpretable as compared to standard BE model or with freeware Osirix. IVIM analysis using Total Variation or Huber Penalty function methodology showed qualitatively and quantitatively better estimation of both perfusion and diffusion component in Ewing sarcoma. Addition of optimization functions like Total Variation and Huber in the IVIM analysis methodology help to reduces non-physiological spatial inhomogeneity, thus could benefit analysis with clinically interpretable IVIM parameters. These methods could be similarly applied for IVIM analysis of other type of tumours.