Apparent diffusion coefficient as a quantitative biomarker for prostate cancer treatment response on a 1.5 Tesla magnetic resonance-linear accelerator: Impact of image registration and acquisition type

Abstract

Background and purpose: Diffusion-weighted magnetic resonance imaging (DW-MRI) is a quantitative biomarker for cancer detection and treatment monitoring. On magnetic resonance-linear accelerator (MR-Linac) systems, diffusion-weighted echo planar imaging (DW-EPI) suffers from geometric distortion, reducing the repeatability of apparent diffusion coefficient (ADC) measurements. This study evaluated the effect of low-distortion split acquisition of fast spin-echo signal (SPLICE) sequences, and of image registration on the repeatability coefficient (RC) of ADC.

Materials and methods: ADC bias, repeatability, signal-to-noise ratio (SNR) and geometric fidelity were measured in a diffusion phantom using three DW-EPI and two DW-SPLICE protocols. ADC short-term and long-term RCs were measured in healthy volunteers. In patients, the registration of DW-EPI to unweighted images (b0) was tested for its effect on RC in gross tumour volume (GTV) and non-tumour prostate (NT-P), and for its ability to detect significant ADC changes.

Results: Phantom experiments showed strong linear correlation with ground-truth ADC (R² > 0.99). Among EPI protocols, DW-EPI-AP offered the best balance of high SNR and low RC, while Z-direction encoded DW-EPI was the most variable. Both DW-SPLICE variants exhibited reduced distortion compared with EPI but poorer repeatability. In volunteers, long-term RCs (8.0-33.7 %) varied more than short-term RCs (8.9-15.4 %). In patients, registration improved RCs (GTV: 28.0 → 25.1 %; NT-P: 19.6 → 12.6 %) and improved detection of significant ADC change in patients (GTV: 0/6 → 1/6; NT-P: 2/6 → 5/6).

Conclusion: RC and accuracy of DW-EPI agrees with published literature and improves after registration. DW-SPLICE shows lower geometric distortion but would require further optimization and validation to improve repeatability.

Keywords: Apparent diffusion coefficient; Diffusion weighted echo planar imaging; Geometric distortion correction; MR-Linac; Repeatability coefficient.

Graphical abstract

Fig. 1

Linear fits of measured versus reference ADC values for each sequence are shown in 10⁻³ mm²/s, with R², slope (proportional bias; ideal = 1), and intercept (additive bias; ideal = 0) reported alongside ±95 % CIs. All sequences exhibit strong linearity (R²). DW-EPI and DW-EPI-AP show no detectable bias. DW-EPI-Z demonstrates proportional but not additive bias. DW-SPLICE performs the worst, exhibiting both additive and proportional biases. EPI-b500 refers to ADC calculated the b-values of 0 and 500 s/mm², unlike the others where ADC was derived using all the b-values.

Fig. 2

Distances between markers in the phantom measured on various DW images and 3D T2W images, which served as distortion-free reference. All DW-EPI sequences caused significant distortion compared to the reference for all the marker distances, DW-SPLICE-LH showed reduced distortion, while DW-SPLICE showed no significant distortion compared to the T2. Values are represented as mean ± SD; * indicates p < 0.05; ** indicates p < 0.01.

Fig. 3

Delta ADC measured in the prostate of three healthy volunteers for different DW sequences (colour-coded). Hollow markers denote short-term repeatability (within the same session). Filled markers denote long-term repeatability between sessions A and B (one week apart) across DW sequences for the three volunteers. Average delta ADC and ±1 SD are shown. The bar graphs show ADC values across sequences. The EPI-Z sequence exhibits significantly higher ADCs compared to all other sequences. DW-SPLICE-LH shows significantly higher ADCs than DW-EPI and DW-SPLICE, between which no significant difference is observed.

Fig. 4

Registration results for a slice containing the GTV anatomy from patient P11. The non-deformed b0 image (fixed) is shown along with deformation fields in the left–right (LR) and anterior–posterior (AP) directions for each b-value and diffusion encoding direction, overlaid on the corresponding diffusion image (moving images). Numbers indicate the average deformation (mm) within the prostate contour. For each diffusion-weighting direction, the scanner’s three physical axes (X, Y, Z) contribute: Dir 1 = [−0.67, 0.33, −0.67], Dir 2 = [0.33, −0.67, −0.67] and Dir 3 = [−0.67, −0.67, 0.33]. Globally, deformation is higher for higher b-values. Deformations for diffusion direction 2 follows an opposite pattern to that of the deformations in diffusion encoding 1 and 3.

Fig. 5

Mean ADC for the GTV, non-tumour prostate (NT-P) and whole prostate (WP) in the 14 scans from the 7 fractions used to calculate the RC. Registration reduced the ADC in the whole prostate. The ADC in the GTV is significantly lower than for NT-P for both the registered (Reg) and the non-registered (non-Reg) workflows.

Fig. 6

ADC measurements for six patients in the GTV and non-tumour prostate (NT-P) from the five-fraction cohort, comparing non-registered (black) and registered (red) data. Measurements were taken at two different fractions (F), roughly one week apart. Dashed lines and hollow circles represent NT-P while bold lines and filled circles represent GTV. Horizontal lines represent 95% confidence limits for detecting a true change. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)