Multi-Center Diffusion-Weighted MRI Validation for 0.35T MR-Linac

Abstract

Background

MR-linac (MRL)-based radiotherapy allows daily anatomical and physiological imaging for adaptive treatment. Apparent diffusion coefficient (ADC) derived from diffusion-weighted imaging (DWI) is a quantitative biomarker correlated with tumor cellularity and response. Verifying ADC repeatability and reproducibility across centers is key for multi-center physiological adaptive radiotherapy trials on low-field MRLs.

Methods

Echo-planar imaging (EPI)-based DWI was performed on a commercial phantom containing 13 0–50% polyvinylpyrrolidone (PVP) vials. Diffusion weighting was applied in three orthogonal directions using two b-value sets: [0,30,60,90,120,150,180,200,500,1000] and [0,500,1000] s/mm². Reproducibility was assessed across four institutions using brain and torso coils. Repeatability was evaluated with four scans at one site. To assess gantry angle effects, scans were repeated at the angle inducing highest distortion. Coefficient of variation (COV) and normalized deviation from reference ADC values were calculated.

Results

Using three b-values, COVs on a single MRL ranged 0.339-2.914% (brain coil) and 0.466-6.902% (torso coil); and with ten b-values, ranged 0.196–3.679% and 0.408–6.699%, respectively. Multi-site COVs ranged 0.329–3.413% (three b-values) and 0.499–2.420% (ten b-values) using brain coil, and 0.267–4.442% and 0.381–5.357% with torso coil. At zero gantry angle, normalized root mean squared error (nRMSE) ranged 2.857–3.538% (single MRL) and 3.015–3.952% (multi-center). Distortion increased nRMSE to 10.767% and 6.236% with three and ten b-values, respectively.

Significance

These findings demonstrate excellent repeatability and reproducibility of ADC values on low-field MRLs across coils and b-value sets, supporting its use as a reliable biomarker in multi-center adaptive radiotherapy trials.

INTRODUCTION

The MR-linac (MRL) offers a unique opportunity for daily imaging during a radiotherapy course and allows for treatment plan adjustments based on interfraction anatomical changes[1]. Unlike other linac imaging devices, MRL images provide excellent soft tissue contrast[2]. Additionally, various MRI pulse sequences can generate quantitative MRI (qMRI) maps which offer physiological information about the tumor microenvironment[3]. A key opportunity is to adapt treatment based on microenvironmental response during radiotherapy—a process known as physiologic adaptive radiation therapy (PART) [4].

The functional and biological information provided by MR images is primarily based on quantitative imaging biomarkers (QIBs) [5] which can also be acquired on MRLs. Common qMRI methods in oncology include dynamic contrast-enhanced MRI (DCE-MRI), dynamic susceptibility contrast MRI (DSC-MRI), and diffusion-weighted imaging (DWI) [6] which can all be integrated within multi-parametric MRI (mpMRI) techniques. mpMRI can play a significant role in the modern radiotherapy (RT) workflow, offering opportunities for individualized RT by providing information on treatment response and aiding in adaptive treatments [3]. mpMRI images can be integrated into the RT simulation process, either in an MR-only workflow or by fusing them with CT simulation images [7]. The apparent diffusion coefficient (ADC), derived from DWI images, provides information about tissue and tumor cellularity based on diffusivity of water molecules within intra- and intercellular spaces. ADC can differentiate malignant from benign lesions, grade tumors, assess treatment response, and identify residual or recurrent tumors and can be used in the brain, head and neck, chest, breast, hepatobiliary, pancreas, esophago-gastrointestinal, gynecological, prostate, urinary system, lymphatic system, bone and soft tissue tumors [8]. It can also be a component of adaptive radiation treatments [9, 10]. Furthermore, time-dependent diffusion imaging can estimate biomarkers such as the volume-to-surface area ratio of cells (V/S) and cell membrane permeability (κ) in tumors[11]. In a clinical trial, Naghavi et al. used radiomics analysis of ADC and DCE data from both diagnostic and low-field MRL images to identify areas of radioresistance within soft tissue sarcomas (STS), and improving treatment outcomes by optimizing the dose in these areas [10]. Given the significant role of mpMRI in RT, standardization and quantification of these techniques and the related QIBs is critically important, which was the motivation for our study to assess the repeatability and reproducibility of ADC on the low-field MRL.

Lutsik et al. recently demonstrated the feasibility of acquiring echo-planar imaging (EPI) DWI images on the 0.35T MRL and demonstrated that ADC measurements at 0.35T and 3T are comparable and can be tracked throughout the radiotherapy course in patients with glioblastoma[12]. As a precursor to future multi-center adaptive radiotherapy trials, this study aims to ensure that NIST-traceable phantom ADC measurements on the same machine are repeatable and furthermore that measurements across different institutions are consistent.

MATERIALS AND METHODS

Phantom

The diffusion phantom for DWI Standardization (CaliberMRI, Boulder, CO) was used in this study. It consists of thirteen 300 mL vials of aqueous polyvinylpyrrolidone (PVP) solutions at concentrations of 0%, 10%, 20%, 30%, 40%, and 50% w/w (three vials at 0% and two vials at each of the other concentrations). The reported ADC values of the vials measured at 3.0T and 22°C are 2.106 ± 0.045, 1.640 ± 0.036, 1.258 ± 0.028, 0.886 ± 0.021, 0.545 ± 0.014, and 0.0293 ± 0.009 x 10⁻³ mm²/s for each PVP concentration, respectively.

Imaging

Imaging was performed on four 0.35T MRLs (MRIdian Linac - ViewRay, Mountain View, CA) at different institutions. To investigate the effect of different coils on image quality and resulting ADC maps, images were acquired using both the 6-channel torso coil (clustered in two groups of three) and the MR high-resolution 6-channel brain coil (Figure 1). Diffusion-weighted images were acquired with an EPI sequence with TR/TE = 3200/120 ms, Flip Angle = 90°, NA = 16, Pixel size = 3 x 3 mm², BW = 1350 Hz/Px, slice thickness = 6 mm, FOV = 300 x 300 x 144 mm³. Images were acquired with diffusion weighting applied in three orthogonal directions with two sets of b-values: [0, 30, 60, 90, 120, 150, 180, 200, 500, 1000] s/mm² and [0, 500, 1000] s/mm² (Figure 2a–c). The total time for acquiring the images with the three b-values was 4 minutes and 58 seconds, and for the protocol with 10 b-values the acquisition time was 6 minutes and 3 seconds. Both sequences were part of the clinical/research protocols used for imaging human subjects, and no change to any of the parameters were made for performing the current phantom study. Measurements were repeated using both coils with the gantry angle set to zero. At each center, the gantry angle with the highest distortion effect was identified with the method explained in the following sections, and separate sets of DWI images were acquired using the brain coil at these angles. The couch and in-room lasers were turned off during image acquisition to eliminate RF signal interference.

Figure 1.

Phantom setup with (a) brain and (b) torso coils

Figure 2.

DWI of the diffusion phantom for b-values of (a) 0 mm²/s, (b) 500 mm²/s, and (c) 1000 mm²/s and (d) the ADC map generated using these images. The yellow contours indicate the regions where sampling of the voxels was performed.

ADC Calculation

ADC values were calculated using MATLAB (R2023a, MathWorks, Inc., Natick, MA, USA), based on the mono-exponential nature of Gaussian diffusion and by fitting the data to the following equation[13]:

$$ADC=\frac{1}{b}\ln (\frac{S_0}{S_b})$$ (1)

where S₀ is the signal acquired with no diffusion weighting and S_b is the diffusion-weighted signal acquired at each b-value. After calculating the ADC values, a circular ROI of consistent size, was drawn by a single person on each vial on the central slice of the phantom using ImageJ (National Institutes of Health, USA), and the median values were found for each vial (Figure 2d). The average of the median ADC values of the vials with the same PVP concentration was calculated and reported.

Temperature Correction

Prior to measurements at each center, the phantom was kept at room temperature to stabilize the temperature during measurements. At the end of all sequence acquisitions, the temperature of the fluid within the phantom was measured using a digital thermometer, and the value was recorded to apply corrections to the calculated ADC values. For each PVP concentration, the temperature dependence of the ADC values was calculated using values provided by the vendor.

The temperature dependence curves of the ADC values for the vials with six different PVP concentrations are shown in Figure 3 based on the values reported by the vendor. The vendor has provided the ADC values at 0°C, and for a range of temperatures from 16°C to 26°C in increments of 2°C. The slope of the linear fit to the data points was found to be 0.0555, 0.0123, 0.0186, 0.0271, 0.0346, and 0.0453 ×10⁻³ mm²/s per °C for the 0%, 10%, 20%, 30%, 40%, and 50% w/w PVP concentrations, respectively. Using the slope for each vial along with the deviation of the phantom temperature from 22°C, the ADC values measured at each center were corrected to represent the values at 22°C.

Figure 3.

Temperature dependence of ADC values for the six PVP concentrations within the diffusion phantom as reported by the phantom vendor.

Considering that ADC values measured at 0°C are considered standard references due to the reproducibility and stability of the temperature, we did a single time-point measurement at one site by bringing down the phantom temperature to close to zero degrees using a mixture of water and ice. The DWI scans were done using the brain coil and both sets of b-values. In addition, during this session we did a single measurement by shifting the phantom 10 cm vertically from the iso center, to assess the changes in the estimated ADC values.

Identifying the Gantry Angle with the Highest Distortion

During gantry angle rotation, motion of the linear-accelerator components perturbs the static magnetic field (B0), producing a gantry angle dependent center frequency offset (Δf) that affects single-shot EPI-DWI. Because a uniform Δf generates an in-plane shift along the phase-encode direction, we selected the gantry angle with the largest Δf as a surrogate metric for the gantry angle with the worst-case distortion. To measure Δf, a Siemens 24cm DSV (Diameter of Spherical Volume) phantom was placed at isocenter and allowed to settle for approximately 10 minutes. A short, single voxel F₀ (center frequency) was acquired for gantry angles from 0 to 345 degrees in increments of 15 degrees (24 measurements), with a 30 second delay after each rotation and before acquisition to allow the field to settle. The resulting values were then resampled to increments of 5 degrees from 0 to 355 degrees (72 measurements) using a distance weighted linear interpolation. The gantry angle with the largest Δf from 0 to 355 degrees was designated as the gantry angle with the worst-case distortion.

Repeatability and Reproducibility

To explore the reproducibility of ADC values, DWI scanning was performed on four MRIdian systems across different centers using the same phantom for all measurements. All imaging parameters of the DWI sequence were kept consistent across all scans. If a center did not have the dedicated brain coil, a coil from another center was used. The existing torso coils at each center were used for all other scans. Repeatability of the measurements was investigated by acquiring four sets of scans over two weeks at a single center with two scans performed on the same day (Scanning was performed over four separate sessions). The phantom was kept at room temperature inside the vault during this period to ensure homogeneous temperature across the volume and all vials. To assess the repeatability and reproducibility of the repeated measurements, the coefficient of variation (COV) [14] was calculated as the ratio of the standard deviation of the measurements to their mean for each PVP concentration. Also, for each of the 13 vials in the diffusion phantom, repeatability was evaluated using the Repeatability Coefficient (RC), defined as 2.77 times the within-subject standard deviation (SD), following the QIBA DWI Profile recommendations [15]. Based on QIBA guidelines, for phantom measurements done during the same session (Short-term RC) the upper threshold is 0.015 × 10⁻³ mm²/s and for those done on multiple sessions (Long-term RC) this threshold is 0.065 × 10⁻³ mm²/s. These thresholds help determine if changes in ADC values in a phantom reliably represent a true change, indicating consistency in scanner performance over time. While RC is defined for measurements done under identical conditions, for comparison purposes, we used the same definition to calculate the reproducibility coefficient (RDC). Additionally, the deviation of the mean ADC value for each PVP concentration from their corresponding reference values was calculated as the difference normalized by the reference value and represented as a percentage. The COV and deviation were calculated for each group in the reproducibility and repeatability studies and for each set of measurements by each coil. To estimate the overall performance of the measurements for each setup, the normalized mean squared error (nRMSE)was calculated as below:

$$nRMSE=100\times \sqrt{\frac{1}{n}{\sum }_{i=1}^n{\left(\frac{a_i^m-a_i^r}{a_i^r}\right)}^2}$$ (2)

Where $a_i^m$ is the measured ADC value, $a_i^r$ is the reference value reported by the vendor, and n = 6 is the total number of PVP concentrations in the phantom. The nRMSE calculated with this formalism is reported as percentage.

RESULTS

Repeatability Results of ADC Measurements

Table 1 shows the ADC values for the repeatability study measurements using the DWI datasets acquired with three b-values, and using the brain and torso coils with the gantry angle set to zero degrees. The highest COV of the ADC values in each set was observed for the 50 %w/w PVP concentration vials, with vial #13 having the highest values (2.914% and 6.902% for the brain coil and torso coil, respectively). These vials also have the lowest ADC values relative to the other vials. The average COV for the measurements across all the vials acquired with the brain coil was lower than those acquired with the torso coil (1.014% ± 0.820% vs. 1.939% ± 1.930%). The mean normalized deviation of the estimated ADC values from the mean ADC values reported by the vendor (expressed in %) is also tabulated in this table and is generally higher for the vials with lower PVP concentration. These range from −2.215% to 7.468% for the measurements done with the brain coil and −1.633% to 8.482% with the torso coil. The nRMSE across all the vials in each group for the measurements done with the brain and torso coils were 3.415% and 3.516%, respectively.

Table 1.

ADC values of vials with different PVP concentrations measured on a single MRL using brain and torso coils, to estimate the repeatability of these values using the DWI sequence with three b-values. The coefficient of variation (COV), repeatability coefficient (RC,) and deviation from the vendor-reported mean values are tabulated for each vial. All measurements were done with the gantry angle set to zero degrees.

Vial #	PVP Concentration (%w/w)	Reported ADC value at 22°C (10−3 mm2/s)	Brain Coil				Torso Coil
			Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RC	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RC
1	0	2.106±0.045	2.116 ± 0.015	0.453	0.715	0.042	2.098 ±0.024	−0.363	1.147	0.067
2			2.133 ± 0.014	1.277	0.652	0.039	2.098 ±0.025	−0.373	1.186	0.069
3			2.150 ± 0.021	2.108	0.991	0.059	2.127 ±0.010	0.999	0.466	0.027
4	10	1.640±0.036	1.669 ± 0.008	1.785	0.507	0.023	1.673 ±0.019	2.016	1.114	0.052
5			1.676 ± 0.010	2.209	0.607	0.028	1.666 ±0.022	1.581	1.323	0.061
6	20	1.258±0.028	1.352 ± 0.005	7.468	0.339	0.013	1.365 ±0.006	8.482	0.461	0.017
7			1.256 ± 0.004	−0.151	0.353	0.012	1.261 ±0.008	0.240	0.612	0.021
8	30	0.886±0.021	0.866 ± 0.003	−2.215	0.368	0.009	0.872 ±0.008	−1.633	0.954	0.023
9			0.881 ± 0.005	−0.546	0.563	0.014	0.895 ±0.009	1.054	1.045	0.026
10	40	0.545±0.014	0.566 ± 0.008	3.895	1.414	0.022	0.568 ±0.009	4.249	1.643	0.026
11			0.567 ± 0.006	4.026	1.065	0.017	0.580 ±0.017	6.513	2.844	0.046
12	50	0.293±0.009	0.306 ± 0.008	4.553	2.693	0.023	0.303 ±0.017	3.565	5.507	0.046
13			0.307± 0.009	4.913	2.914	0.025	0.300 ±0.021	2.443	6.902	0.057

In Figure 4–a and c, the mean measured ADC values for each PVP concentration using three b-values are plotted against the mean reference values with the error bars showing the standard deviation of the measured values. The plotted trend lines show overall excellent agreement of the ADC values with the reference values with R² of 0.9993 and 0.9989 for the brain and torso coils, respectively. The ADC values estimated using ten b-values and from the four measurements done on a single MRL are tabulated in Table 2 and plotted in Figure 4 (b) and (d). The average COV of the estimated ADC values were slightly lower than the measurements done with 3 b-values (0.914% ± 1.024%, 1.599% ± 1.659% for the brain and torso coils, respectively), and the overall agreement with the reference values based on the nRMSE was slightly better for both the brain and torso coils (2.857% and 2.866%, respectively). The estimated RC for all vials was less than 0.065 ×10⁻³ mm²/s for all the repeatability measurements, except for the two vials with the lowest PVP concentration, when measured with the Torso coil and with three b-values (RC = 0.069 and 0.067), and one vial with PVP concentration of 10 %w/w, when measured with the 10 b-values and the Torso coil (RC = 0.070).

Figure 4.

Mean ADC values of the PVP concentrations measured four times on a single MRL with the brain (a & b) and Torso (c & d) coils. Measurements were performed using the three (a & c), and ten (b & d) b-values with the gantry angle set to zero degrees (G0). Error bars show the standard deviation of measurements.

Table 2.

ADC values of vials with different PVP concentrations measured on a single MRL using brain and torso coils, to estimate the repeatability of these values using the DWI sequence with ten b-values. The coefficient of variation (COV), repeatability coefficient (RC,) and deviation from the vendor-reported mean values are tabulated for each vial. All measurements were done with the gantry angle set to zero degrees.

Vial #	PVP Concentration (%w/w)	Reported ADC value at 22°C (10−3 mm2/s)	Brain Coil				Torso Coil
			Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RC	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RC
1	0	2.106±0.045	2.111 ± 0.004	0.246	0.196	0.011	2.077 ± 0.012	−1.377	0.565	0.032
2			2.139 ± 0.006	1.585	0.287	0.017	2.083 ± 0.008	−1.114	0.408	0.024
3			2.151 ± 0.016	2.118	0.726	0.043	2.128 ± 0.025	1.037	1.186	0.070
4	10	1.640±0.036	1.655 ± 0.005	0.888	0.274	0.013	1.648 ± 0.015	0.517	0.888	0.041
5			1.671 ± 0.005	1.876	0.291	0.013	1.653 ± 0.013	0.765	0.782	0.036
6	20	1.258±0.028	1.346 ± 0.003	6.979	0.200	0.007	1.344 ± 0.015	6.839	1.084	0.040
7			1.258 ± 0.006	0.017	0.469	0.016	1.255 ± 0.009	−0.204	0.692	0.024
8	30	0.886±0.021	0.869 ± 0.005	−1.909	0.550	0.013	0.864 ± 0.009	−2.438	0.989	0.024
9			0.875 ± 0.002	−1.189	0.176	0.004	0.886 ± 0.006	0.015	0.715	0.018
10	40	0.545±0.014	0.571 ± 0.010	4.793	1.786	0.028	0.563 ± 0.009	3.339	1.646	0.026
11			0.563 ± 0.005	3.217	0.889	0.014	0.572 ± 0.009	4.868	1.617	0.026
12	50	0.293±0.009	0.294 0.011	0.215	3.679	0.030	0.291 0.020	−0.569	6.699	0.054
13			0.301 0.007	2.755	2.362	0.020	0.304 0.011	3.709	3.523	0.030

Reproducibility Results of ADC Measurements

The ADC values measured across the four centers with three b-values are tabulated in Table 3. Compared to the measurements done at the single center (repeatability study) and when the gantry angle was set to zero degrees, the overall average COV across all the vials was slightly higher for both the brain and torso coils (1.125% ± 0.870% and 1.743% ± 0.955%, respectively) with the highest COV belonging to the one of the 50 %w/w PVP vials measured with both the brain and torso coils at 3.413% and 4.442%, respectively. The deviation of the measured ADC values from the reference values was also higher than the repeatability study and ranged from −2.308% to 8.742% (nRMSE = 3.705%) and −3.738% to 8.092% (nRMSE = 3.952 %) for the brain and torso coils, respectively. In Figure 5(a) and (c), the mean of the measured ADC values for each PVP concentration using three b-values are plotted against the reference values.

Table 3.

ADC values of vials with different PVP concentrations measured on four MRLs using brain and torso coils to estimate the reproducibility of these values using the DWI sequence with three b-values. The coefficient of variation (COV), reproducibility coefficient (RDC,) and deviation from the vendor-reported mean values are tabulated for each vial.

Vial #	PVP Concentration (%w/w)	Reported ADC value at 22°C (10−3 mm2/s)	Brain Coil				Torso Coil
			Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RDC	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RDC
1	0	2.106±0.045	2.129 ± 0.022	1.108	1.016	0.060	2.110 ± 0.031	0.175	1.462	0.085
2			2.124 ± 0.008	0.855	0.395	0.023	2.057 ± 0.027	−2.343	1.294	0.074
3			2.125 ± 0.020	0.901	0.919	0.054	2.066 ± 0.034	−1.899	1.669	0.096
4	10	1.640±0.036	1.646 ± 0.005	0.393	0.329	0.015	1.629 ± 0.004	−0.658	0.267	0.012
5			1.674 ± 0.009	2.066	0.531	0.025	1.657 ± 0.023	1.038	1.383	0.063
6	20	1.258±0.028	1.351 ± 0.005	7.354	0.389	0.015	1.337 ± 0.016	6.279	1.205	0.045
7			1.268 ± 0.010	0.790	0.805	0.028	1.236 ± 0.015	−1.729	1.223	0.042
8	30	0.886±0.021	0.866 ± 0.010	−2.308	1.184	0.028	0.853 ± 0.020	−3.738	2.306	0.054
9			0.879 ± 0.003	−0.817	0.340	0.008	0.870 ± 0.013	−1.797	1.473	0.036
10	40	0.545±0.014	0.566 ± 0.013	3.832	2.366	0.037	0.566 ± 0.015	3.883	2.697	0.042
11			0.593 ± 0.009	8.742	1.487	0.024	0.573 ± 0.008	5.072	1.397	0.022
12	50	0.293±0.009	0.297 ± 0.010	1.263	3.413	0.028	0.308 ± 0.014	5.147	4.442	0.038
13			0.305 ± 0.004	4.226	1.451	0.012	0.317 ± 0.006	8.092	1.840	0.016

Figure 5.

Mean ADC values of the PVP concentrations measured across four MRLs at multiple centers using the brain (a & b) and torso (c & d) coils and with the gantry angle set to zero (G0). The bottom row (e & f) shows the measurements done with the gantry angle set to the “worst” (Worst G. Angle), or the gantry angle with the highest distortion at each center. Measurements were done using DWI sequences with three (a, c & e), and ten (b, d, & f) b-values at G0. Error bars represent the standard deviation of measurements.

In Table 4, the results of the reproducibility study using ten b-values are tabulated. Compared to the measurements with three b-values, the overall COV for the ADC values across all vials was lower when ten b-values were used, with 1.144% ± 0.573% and 1.655% ± 1.190% for the brain and torso coils, respectively. The highest deviation from the reference values was for vials #6 (20 %w/w PVP) at 7.354 (Brain coil) and vial #13 (50 %w/w PVP) at 6.874 (Torso Coil) and the nRMSE was 3.015% and 3.790% for the brain and torso coils, respectively. The RDC for most vials in the reproducibility study were less than 0.065 ×10⁻³ mm²/s, except for three vials with 0 %w/w and 10 %w/w PVP concentration (RDC = 0.096, 0.074, 0.085 ×10⁻³ mm²/s) when three b-values and the torso coil were used, and for vial #1 (0 %w/w PVP, RC = 0.068 ×10⁻³ mm²/s, Brain coil) and vial #3 (10 %w/w PVP, RC = 0.091 ×10⁻³ mm²/s, Torso coil). These measurements are plotted in Figure 5(b) and (d).

Table 4.

ADC values of vials with different PVP concentrations measured on four MRLs using brain and torso coils to estimate the reproducibility of these values acquired using the DWI sequence with ten b-values. The coefficient of variation (COV), reproducibility coefficient (RDC,) and deviation from the vendor-reported mean values are tabulated for each vial. All measurements were done with the gantry angle set to zero degrees.

Vial #			Brain Coil				Torso Coil
	PVP Concentration (%w/w)	Reported ADC value at 22°C (10−3 mm2/s)	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RDC	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RDC
1	0	2.106±0.045	2.127 ± 0.025	1.013	1.158	0.068	2.112 ± 0.023	0.264	1.099	0.064
2			2.144 ± 0.016	1.807	0.765	0.045	2.041 ± 0.020	−3.105	0.999	0.056
3			2.120 ± 0.029	0.682	1.379	0.081	2.045 ± 0.033	−2.878	1.613	0.091
4	10	1.640±0.036	1.650 ± 0.009	0.599	0.549	0.025	1.629 ± 0.006	−0.641	0.381	0.017
5			1.662 ± 0.008	1.365	0.499	0.023	1.646 ± 0.016	0.361	0.951	0.043
6	20	1.258±0.028	1.351 ± 0.011	7.379	0.806	0.030	1.324 ± 0.017	5.282	1.303	0.048
7			1.266 ± 0.010	0.623	0.769	0.027	1.228 ± 0.022	−2.389	1.784	0.061
8	30	0.886±0.021	0.874 ± 0.005	−1.406	0.616	0.015	0.849 ± 0.006	−4.151	0.746	0.018
9			0.892 ± 0.010	0.698	1.089	0.027	0.878 ± 0.017	−0.904	1.992	0.048
10	40	0.545±0.014	0.567 ± 0.011	4.045	1.855	0.029	0.579 ± 0.014	6.291	2.402	0.039
11			0.563 ± 0.006	3.393	1.040	0.016	0.555 ± 0.007	1.925	1.209	0.019
12	50	0.293±0.009	0.300 ± 0.007	2.225	2.420	0.020	0.308 ± 0.016	5.098	5.357	0.046
13			0.306 ± 0.006	4.589	1.927	0.016	0.313 ± 0.005	6.874	1.674	0.015

In Table 5, the ADC values of the measurements performed with the gantry angle with the highest distortion in DWI images are tabulated. As expected, the calculated ADC values had both COV and percent deviations higher than the measurements done with the gantry set to zero degrees. When using three b values, the COV across all vials was 3.219 % ± 2.191 % with nRMSE of 10.767%. The highest COV was for the vials with 50 %w/w PVP concentration at 7.937% (vial #13), and the highest deviation was 25.732% (Vial #12). When using ten b-values, these measurements also showed worse results compared to the same DWI configuration with the gantry angle set to zero, resulting in higher deviation from the reference values, as high as 12.327% (for the 50 %w/w PVP) and with an nRMSE of 6.236% and a higher mean COV of 5.013% ± 4.458% across all vials. When using this setting, the RDC of some of the vials exceeded the 0.065 ×10⁻³ mm²/s threshold.

Table 5.

ADC values of vials with different PVP concentrations measured on four MRLs using the brain coil to estimate the reproducibility of these values acquired using the DWI sequence with ten b-values, and the gantry angle set to that with the highest distortion. The coefficient of variation (COV), reproducibility coefficient (RDC,) and deviation from the vendor-reported mean values are tabulated for each vial.

Vial #	PVP Concentration (%w/w)	Reported ADC value at 22°C (10−3 mm2/s)	Three b-values (Acquired with Worst Gantry Angle)				Ten b-values (Acquired with Worst Gantry Angle)
			Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RDC	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	COV (%)	RDC
1	0	2.106±0.045	2.129 ± 0.033	1.110	1.555	0.088	2.112 ± 0.061	0.276	2.896	0.163
2			2.133 ± 0.016	1.273	0.765	0.043	2.141 ± 0.041	1.652	1.914	0.109
3			2.143 ± 0.037	1.756	1.722	0.098	2.151 ± 0.037	2.120	1.710	0.098
4	10	1.640±0.036	1.683 ± 0.021	2.627	1.226	0.055	1.677 ± 0.034	2.252	2.005	0.089
5			1.710 ± 0.039	4.275	2.281	0.104	1.720 ± 0.033	4.883	1.890	0.086
6	20	1.258±0.028	1.411 ± 0.024	12.182	1.715	0.064	1.397 ± 0.020	11.031	1.450	0.054
7			1.288 ± 0.039	2.424	3.035	0.104	1.286 ± 0.029	2.247	2.269	0.078
8	30	0.886±0.021	0.907 ± 0.016	2.320	1.791	0.043	0.912 ± 0.040	2.937	4.383	0.106
9			0.920 ± 0.034	3.852	3.711	0.091	0.895 ± 0.034	0.994	3.828	0.091
10	40	0.545±0.014	0.613 ± 0.040	12.547	6.473	0.106	0.569 ± 0.034	4.448	5.947	0.090
11			0.594 ± 0.020	9.060	3.410	0.054	0.573 ± 0.046	5.115	8.102	0.123
12	50	0.293±0.009	0.368 ± 0.023	25.732	6.225	0.061	0.327 ± 0.053	11.624	16.189	0.141
13			0.352 ± 0.028	19.995	7.937	0.074	0.329 ± 0.041	12.327	12.580	0.110

Measurements at zero degrees Celsius and Off-Center Position

Table 6 summarizes the ADC measurements done at zero degrees Celsius and the deviation from the values reported by the phantom vendor. With three b-values, the deviation from the reference values ranged from −4.536% to 8.639%. When 10 b-values were used the deviation ranged from −4.468% to 9.442%. Generally, the deviation of the lower ADC values was higher in both setups. After shifting the phantom along the Y-axis by 10 cm, the deviation from reference values increased to as high as −28.636%.

Table 6.

ADC values of vials with different PVP concentrations measured on a single MRL using the brain coil at 0°C using the DWI sequence with three and ten b-values, and the gantry angle set to zero degrees, with the center of the phantom placed at isocenter. The measurements with 10 b-values were repeated after shifting the phantom by 10 cm vertically from the isocenter.

Vial #	PVP Concentration (%w/w)	Reported ADC value at 0°C (10−3 mm2/s)	Three b-values (At Isocenter)		Ten b-values (At Isocenter)		Ten b-values (Off Isocenter)
			Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	Measured ADC values (10−3 mm2/s)	Mean Deviation (%)	Measured ADC values (10−3 mm2/s	Mean Deviation (%)
1	0	1.109±0.025	1.119	0.897	1.117	0.703	1.136	2.434
2			1.110	0.115	1.097	−1.047	1.044	−5.846
3			1.103	−0.579	1.098	−1.027	1.003	−9.597
4	10	0.817±0.019	0.817	−0.001	0.807	−1.267	0.782	−4.237
5			0.836	2.278	0.824	0.830	0.826	1.115
6	20	0.579±0.015	0.629	8.639	0.634	9.442	0.602	3.968
7			0.605	4.494	0.598	3.241	0.558	−3.561
8	30	0.380±0.011	0.377	−0.884	0.369	−2.842	0.314	−17.291
9			0.393	3.361	0.385	1.352	0.329	−13.389
10	40	0.220±0.007	0.227	3.100	0.215	−2.059	0.187	−15.000
11			0.226	2.645	0.213	−3.307	0.192	−12.727
12	50	0.110±0.005	0.106	−3.627	0.105	−4.468	0.074	−33.182
13			0.105	−4.536	0.106	−3.559	0.079	−28.636

DISCUSSION

In this study, we conducted a repeatability and reproducibility analysis of ADC values measured on a phantom using four different 0.35T MRLs and two different coils with an EPI-based DWI sequence. Overall, both the COV of the ADC values for individual vials and the deviation from vendor-reported ADC values showed excellent results validating repeated and multi-institutional ADC measurements for clinical trials involving patients treated on the MRL.

We estimated ADC values using two sets of DWI images: one with three b-values and the other with ten b-values. The set with b-values is part of an Intravoxel Incoherent Motion (IVIM) protocol, implemented to estimate perfusion in live tissue[16]. The images acquired with b-values lower than 200 s/mm² can be used to estimate the pseudo-diffusion coefficient (D*)[17]. The difference between these two sets lies in the seven image volumes acquired with b-values between 30 s/mm² and 200 s/mm². For patient DW images, the diffusion signal can be modeled by a two-compartment model, as described by Le Bihan et al. [18], using a bi-exponential equation to estimate both the pseudo-diffusion coefficient (D*) and the diffusion coefficient (D). D* accounts for diffusion signal loss due to intravascular flow in tissue which is more apparent in DW images acquired with b-values <200 s/mm²[17]. Since there is no flow in the phantom, the pseudo-diffusion component is essentially non-existent, so we fit both datasets to the mono-exponential diffusion equation to estimate D (or ADC). Overall, the deviation of ADC values from vendor-reported values was slightly less when using ten b-values which is expected due to the larger number of data points used for calculation of the ADC value, and the higher signal to noise ratio of the images with lower b-values existing in the dataset with ten b-values. However, the nRMSE of measurements with the gantry angle set to zero was less than 4% in all the repeatability and reproducibility measurements for all data sets and for both coils used. Considering these observations, and to simplify the post-processing workflow and reduce imaging time, acquiring DW images on patients with three b-values can be more practical as it enhances patient comfort and reduces overall machine/scanning time while still providing reliable and reproducible ADC values.

In a study by Rabe et al. the repeatability of EPI-based DWI measurements on the low-field MRL was quantified in human volunteers and on a phantom [19]. In their study, they measured ADC using two EPI sequences, one with an emphasis on high spatial resolution and the other with high signal to noise ratio (SNR). While both sequences resulted in high repeatability of ADC measurements in the phantom and volunteers’ brains, the high SNR sequence resulted in better repeatability and more accurate ADC values. The focus of this study was on repeatability rather than accuracy, and the ADC values measured in the phantom were lower than the reported values. One reason for this deviation was the lower phantom temperature relative to the standard temperature during the experiments. In our study, we mitigated the effect of temperature by finding the correction factors for each PVP concentration based on deviation of the phantom temperature from 22°C at each measurement and applying them to the calculated ADC values. This approach made the comparison of ADC values at each center and across repeated measurements independent of temperature differences at the time of measurements. The RC values that we estimated for the repeated measurements in our study are indicators of the reliability of these measurements on the low field MRL for the majority of the PVP vials for their ability to detect changes in the ADC values in the human subjects. When using the head coil, the RC for all the vials were below the threshold recommended by QIBA, and when using the torso coil, only one and two vials did not meet this criteria when using three and ten b-values, respectively.

The reference ADC values that we used in our study were the ones reported by the vendor and were measured at 3T. The alignment of our measurements with these reference values confirms previous DWI studies which show that with appropriate imaging parameters, the difference in estimated ADC values across different scanners and field strengths is reasonably small [20]. For example, DWI studies have shown no significant difference in ADC values in breast tissue across images acquired at 1.5T and 3.0T[21]. Another study by Weygand et al. compared ADC values in sarcoma tumors based on EPI DWI measurements on the 0.35T MRL and found them analogous to ADC values acquired at higher fields [22]. These findings also align with the study by Lutsik et al. showing that even though phantom images were acquired on a low-field MRI, the calculated ADC values matched those reported at higher field strengths. Also, EPI-based phantom ADC measurements by Wallimann et al. on the 0.35T MR-Linac were within 3 % of those measured by the 3 T scanner [23]. These results suggest that ADC maps of patients scanned on diagnostic MRI are comparable to those acquired on the low-field MRL and can be used and compared across different platforms.

Different studies have reported ADC values in normal brain tissue ranging from 0.62 × 10⁻³ mm²/s to 0.89 × 10⁻³ mm²/s[24], and approximately 0.8 × 10⁻³ mm²/s to 1.6 × 10⁻³ mm²/s across different types and grades of adult brain tumors[25], and 0.1 × 10⁻³ mm²/s to 2.4 × 10⁻³ mm²/s in pediatric brain tumors[26]. These values are comparable to the range of ADC values of the vials in the phantom used in our study. The excellent reproducibility and repeatability of our phantom measurements on the low-field MRL, along with the agreement of estimated ADC values with reference values, suggest the reliability of DWI measurements on human subjects using the low-field MRL.

It is important to note that the gantry position can significantly affect DWI images and estimated ADC values. Lewis et al. conducted a comprehensive evaluation of ADC values and geometric distortion of EPI images on the low-field MRL across the full range of gantry angles and found significant differences in geometric distortion and ADC values at different angles. They measured a shift of the isocenter ranging from 0.25 mm to 1.04 mm across the full range of gantry rotation [27]. They recommended acquiring DWI images at a gantry angle of zero degrees to achieve the lowest geometric distortion. However, they stated that this might be system-dependent, and the optimal angle should be assessed for each system. In a similar study, Nardini et al. identified the gantry angle of zero degrees to be the optimal angle to acquire artifact-free DWI images [28], although in their study they used a turbo spin echo sequence. Therefore, for any human study, it is crucial to ensure the gantry angle is set to the optimal position before image acquisition to reduce geometric distortion and to obtain more reliable ADC measurements.

One limitation of our work was that we did not do an independent assessment of the optimal gantry angle at each center to achieve minimum distortion for the MRLs used; however, based on the reported measurements [27, 28], setting the gantry angle to zero can ensure minimum or low distortion. Another limitation of our study was the relatively low number of centers included in the repeatability study. This was partly due to our intention to use the same exact phantom in all measurements to exclude intrinsic variability of ADC values across different phantoms, even from the same vendor. Logistical issues of transporting the phantom to distant centers limited the study to those within reasonable driving distance. We should also note that one of the vials (Vial #6) showed a consistent deviation from the vendor reported value across all the measurements by ~7%, which increased the nRMSE for all the measurements. This was likely due to a defect in the consistency of the vial contents; however, despite the deviation from the reported value, the measurements showed excellent repeatability and reproducibility for the ADC value of this vial.

CONCLUSION

Our study demonstrated excellent repeatability and reproducibility for ADC values measured on a phantom using an EPI-based DWI sequence on the low-field MRL. These results were consistent when using different numbers of b-values in the DWI sequence and when using either the brain coil or the torso coil for image acquisition. To ensure measurement reliability, it is crucial to set the gantry angle to zero before image acquisition. These results suggest the reliability of ADC values estimated using EPI-based DWI measurements on the low-field MRL for patients at different time points and at different centers, supporting their inclusion as QIBs in clinical trials.

Highlights.

• ADC values acquired on the low field MRLs show excellent reproducibility

• ADC values acquired on the low field MRLs show excellent repeatability

• It is crucial to set the gantry angle to zero before image acquisition

• Images acquired with both Brain and Torso coils provide reliable ADC values