Truly reproducible uniform estimation of the ADC with multi-b diffusion data- Application in prostate diffusion imaging

Abstract

Purpose: The ADC is a well-established parameter for clinical diagnostic applications, but lacks reproducibility because it is also influenced by the choice diffusion weighting level. A framework is evaluated that is based on multi-b measurement over a wider range of diffusion-weighting levels and higher order tissue diffusion modeling with retrospective, fully reproducible ADC calculation.

Methods: Averaging effect from curve fitting for various model functions at 20 linearly spaced b-values was determined by means of simulations and theoretical calculations. Simulation and patient multi-b image data were used to compare the new approach for diffusion-weighted image and ADC map reconstruction with and without Rician bias correction to an active clinical trial protocol probing three non-zero b-values.

Results: Averaging effect at a certain b-value varies for model function and maximum b-value used. Images and ADC maps from the novel procedure are on-par with the clinical protocol. Higher order modeling and Rician bias correction is feasible, but comes at the cost of longer computation times.

Conclusions: Application of the new framework makes higher order modeling more feasible in a clinical setting while still providing patient images and reproducible ADC maps of adequate quality.

Keywords: ADC; b-value; diffusion; kurtosis; prostate.

FIGURE 1

Averaging effect for different higher order model functions in normal prostate (N) and prostate tumor (T) tissue at different $b_{\max }$ values (indicated on top of each plot). The symbols represent the simulation results obtained at SNR 20 and the curves the corresponding analytically derived functions. Vertical dashed lines indicate the clinically sampled b‐values of 1000 and 1500 s/mm². Diffusion‐weighted images obtained with a diffusion weighting around 1500 s/mm² are considered optimal to examine prostate lesions,

FIGURE 2

Diffusion‐weighted images of the prostate in a patient with a PI‐RADSv2.1 score of 4 and a radical prostatectomy specimen Gleason score of 3 + 3, with a tertiary Gleason score of 4. Images are shown for b‐values 0, 100, 1000, and 1500 s/mm² obtained with geometric averaging of three orthogonal encoding directions using the clinical protocol (top), using raw images from the 21 b‐value protocol without repetition (middle) and reconstructed from the 21 b‐value protocol with a kurtosis higher‐order model using OBSIDIAN (bottom). An example of blood vessels outside the prostate not visible in the reconstructed images because of exclusion of signal data measured at b = 0 is indicated in the corresponding clinical image (red arrow). Raw images of the 21 b‐value protocol obtained without averaging at b‐values of 1000 and 1500 s/mm² are considerably noisier than the corresponding clinical and reconstructed images. OBSIDIAN, optimized bias and signal inference in diffusion image analysis

FIGURE 3

Diffusion‐weighted images of same patient and section as shown in Figure 2. The images were reconstructed from the 21 b‐value scan with a kurtosis (top), a biexponential (middle) and a gamma distribution model function (bottom) for b‐values of 1000, 1500, and 2000 s/mm² (b‐value indicated at the top of each column) using OBSIDIAN. Note that the kurtosis images for b‐values of 1000 and 1500 s/mm² are identical to the reconstructed images shown in Figure 2 for the corresponding b‐values. Reconstructions at b = 2000 s/mm² are noisier for all model functions, but the reconstruction based on the kurtosis model function appears slightly noisier than the reconstructions based on the biexponential and the gamma distribution model function. OBSIDIAN, optimized bias and signal inference in diffusion image analysis

FIGURE 4

Diffusion‐weighted images of the prostate in a patient with PI‐RADSv2.1 score of 2. Images are shown for b‐values of 1000, 1500, and 2000 s/mm² (left to right) from the clinical scan and reconstructed from the 21 b‐value scan with a kurtosis model function for $b_{\max }$ values of 1500, 2000, and 2500 s/mm² (top to bottom) using OBSIDIAN. In the case of biexponential and gamma distribution model function the extrapolated images at b = 2000 s/mm² for a $b_{\max }$ value of 1500 s/mm² are substantially less noisy (Figures S2 and S3). OBSIDIAN, optimized bias and signal inference in diffusion image analysis

FIGURE 5

ADC maps resulting from various approaches for the same patient as in Figure 2. The ${\mathrm{ADC}}_K$ map was generated from the kurtosis model function fit using OBSIDIAN. The maps produced with the basic new method, irrespective of model function used, are nearly indistinguishable from the clinical ADC map, whereas the bias‐corrected New Method OBSIDIAN maps appear slightly brighter and slightly noisier in areas outside the prostate. The 2/21b ADC map is noise‐inflicted outside the prostate. The ${\mathrm{ADC}}_K$ is generally brighter and does also appear noisier outside the prostate. OBSIDIAN, optimized bias and signal inference in diffusion image analysis

FIGURE 6

Comparison of clinical ADC map and ADC maps that result with different $b_{\max }$ values (top to bottom) and modeling approaches (left to right) for the same patient as in Figure 4