The influence of b-values, noise levels, range of parameters, and ROI characteristics on the bi-exponential IVIM model and its fitting methods

We aimed to verify the influence of b-value sequences, noise levels, range of diffusion and perfusion parameters, placement, and dimension of region-of-interest (ROI) on the method performance for bi-exponential intravoxel incoherent motion MRI (IVIM-MRI) signal fitting.

We defined four b-value sequences (b1, b2, b3, b4), seven SNR values, and three structures [f; D; D*]s of varying parameters to create voxel-wise IVIM bi-exponential signals. We calculated the performance of six different fitting methods with normalized Euclidean distance Deu between simulated and estimated IVIM parameters. We performed Kruskal–Wallis and multiple-comparison tests to differentiate results statistically. Afterwards, we used the best method/b-sequence combination to assess, with relative errors (RE) and standard deviations (SD), the placement and dimension effects of four distinctly dimensioned square ROIs on estimations based on an image of three simulated tissues. We also evaluate the effect of noise on ROI-based estimation by selecting a 42 × 46 pixel region of each tissue, so that this region did not involve the set’s background.

The combination Levenberg-Marquadt/b2 yielded the best performance during the voxel-wise analysis; in most cases, it had no statistical difference to Trust-Reflective-Region/b2 (P > 0.05). Segmented methods performed worse than the non-segmented ones. ROI placement, rather than its dimension, enhanced partial volume effects that deteriorate estimations, and bigger ROI dimensions mitigated noise drawbacks. D* estimations had the highest variabilities in both voxel and image simulations.

Non-segmented nonlinear methods may provide good estimations (Deu < 0.5) with sequences similar to b2 and SNR > 35. The distribution of low b-values in the sequence is crucial to estimate reliable parameters, mainly concerning D* estimations. The kind of tissue must be taken into account when choosing b-value sequences. We must consider noise conditions if we want good estimations, but the use of ROIs may mitigate noise effects. Moreover, ROI satisfactory dimensioning and placement are vital to avoid partial volume effects.