Session Item

Using diffusion-weighted MR imaging (DWI) on an MR-Linac could enable new ways of early response monitoring or adapting treatment during radiotherapy. We investigated the feasibility of DWI on the MRIdian 0.35T MR-Linac (ViewRay) by analyzing the impact of different receiver coils and imaging settings on apparent diffusion coefficient (ADC) values in a phantom.

We used a spin echo, single shot echo planar imaging DWI sequence with the settings: TR=3200ms, TE=120ms, flip angle=90°, receiver bandwidth=1352Hz/Px, voxel size 3x3x6mm^3, 20 slices without gap, linac gantry angle 330°, three orthogonal diffusion directions and five b values (0, 200, 300, 500, 800 s/mm^2 ).

Furthermore, we varied the number of signal averages (NSA) (6 or 24 per b image), the coil used (head & neck coil (HNC) or prototype head coil (PHC)) and the application of prescan normalization (PN).

For each coil, a noise shot (RF transmit turned off) was acquired with NSA 6 and with PN.

We calculated the ADC values with an in-house python script using a voxel-wise mono-exponential fit. The different diffusion directions were combined using the geometric average of the intensities.

We analyzed three different variations of noise correction: No correction for noise, a correction for uniform Rician background noise (Dietrich et al., 2001) and a correction of the same form but using the non-uniform noise shot.

We investigated a custom-built phantom (HQ imaging) containing four vials with calibrated ADC values (400, 1000, 1600 and 2020 μm^2/s), in each of which a homogeneous region of interest (ROI) was manually selected. The determined ADC values were corrected per phantom manual from the measured temperature of 20.5°C to the calibration temperature of 20°C.

The images with NSA 6, with the HNC and using PN showed a strong inhomogeneity of noise across the image, which remained roughly constant for different b values. The same pattern could also be observed on the noise shot, but to a lesser degree on images with the PHC and not at all without PN (Figure 1).



Figure 2 shows the determined ADC values in each ROI. Increasing the NSA leads to smaller interquartile ranges, but almost unchanged median. Correcting for noise brings the median ADC closer to the true value in most cases. The best accuracy of the median was seen with the PHC without PN and with the uniform noise correction, where relative difference of the median to the true value ranged from -2.3% to 0.3% for the different vials.

The accuracy of the ADC estimates is satisfactory (error <3%) but requires special attention to noise contamination. The latter depends on the setting used, including the choice of the coil, the elements combination and the noise correction strategy. A higher NSA can increase the precision but has little benefit on accuracy of ADC values. Potential errors in the calibration of the phantom or the temperature correction were not considered for this analysis.