Accurate, repeatable, and geometrically precise diffusion-weighted imaging on a 0.35 T magnetic resonance imaging-guided linear accelerator

Highlights

• •A novel echo planar imaging-based imaging protocol allows for accurate and repeatable ADC quantification on a 0.35 T MR-linac.
• •ADC maps produced on a 0.35 T MR-linac are shown to have submillimeter geometric precision and to be significantly more geometrically precise than ADC maps produced on a 3.0 T diagnostic scanner.
• •ADC maps acquired in two sarcoma patients exhibit similar image features whether acquired on a 0.35 T MR-linac or a 3.0 T diagnostic scanner.

Abstract

Background and purpose

Diffusion weighted imaging (DWI) allows for the interrogation of tissue cellularity, which is a surrogate for cellular proliferation. Previous attempts to incorporate DWI into the workflow of a 0.35 T MR-linac (MRL) have lacked quantitative accuracy. In this study, accuracy, repeatability, and geometric precision of apparent diffusion coefficient (ADC) maps produced using an echo planar imaging (EPI)-based DWI protocol on the MRL system is illustrated, and in vivo potential for longitudinal patient imaging is demonstrated.

Materials and methods

Accuracy and repeatability were assessed by measuring ADC values in a diffusion phantom at three timepoints and comparing to reference ADC values. System-dependent geometric distortion was quantified by measuring the distance between 93 pairs of phantom features on ADC maps acquired on a 0.35 T MRL and a 3.0 T diagnostic scanner and comparing to spatially precise CT images. Additionally, for five sarcoma patients receiving radiotherapy on the MRL, same-day in vivo ADC maps were acquired on both systems, one of which at multiple timepoints.

Results

Phantom ADC quantification was accurate on the 0.35 T MRL with significant discrepancies only seen at high ADC. Average geometric distortions were 0.35 (±0.02) mm and 0.85 (±0.02) mm in the central slice and 0.66 (±0.04) mm and 2.14 (±0.07) mm at 5.4 cm off-center for the MRL and diagnostic system, respectively. In the sarcoma patients, a mean pretreatment ADC of 910x10^-6 (±100x10^-6) mm²/s was measured on the MRL.

Conclusions

The acquisition of accurate, repeatable, and geometrically precise ADC maps is possible at 0.35 T with an EPI approach.

1. Introduction

Online magnetic resonance imaging-guided radiotherapy (MRgRT) has experienced widespread clinical implementation with two commercially-available fully-integrated MRgRT systems [1], [2]. Its primary advantage over other online imaging systems is superior soft-tissue contrast [3] allowing visualization of interfractional/intrafractional changes in size/shape of tumors and surrounding critical structures [4]. This facilitates online adaptive radiotherapy incorporating updated anatomy and daily plan reoptimization [5]. Although MRgRT-driven adaptation based on morphological changes has proven successful [6], MRI’s functional imaging capabilities to interrogate tumor physiology in vivo [7], [8], [9] have not been fully exploited [10].

One physiological parameter accessible by MRI is tumor cellularity [11], which quantifies cellular density within a tumor [12] and is a surrogate for cellular proliferation [13]. Diffusion weighted imaging (DWI), which measures water molecule Brownian motion [14], [15] by applying diffusion-sensitizing gradient magnetic fields and observing signal reduction caused by molecular random thermal fluctuations [16], can measure cellularity. Varying diffusion weighting (via b-values [17]) and modeling exponential signal reduction provides an apparent diffusion coefficient (ADC) [18]. ADC is an important imaging biomarker [19], [20], as tumor ADC increases correlate with radiotherapeutic response [21], [22], [23], [24], [25] and can be prognostic [26], [27], [28], [29] across many disease sites, often preceding morphological signals [30]. Thus, ADC maps have potential in adaptive radiotherapy [31], by allowing identification of cellular subpopulations with restricted diffusion and, hence, increased cellularity [32], for which dose escalation and/or biologically-guided plan adaptation may be clinically advantageous.

Feasibility of DWI on a 0.35 T tri-cobalt MRgRT system has been demonstrated [33], [34]. This was illustrated in soft-tissue sarcoma where DWI signal predicted tumor histology [35]. Additionally, DWI combined with deep-learning predicted sarcoma radiotherapeutic response [36]. DWI on a 0.35 T radiotherapy system became more challenging when the tri-cobalt system was replaced by a gantry-mounted linear accelerator (linac) [37]. Eddy currents from the gantry lead to geometric distortion and/or artifacts [38] and can be particularly problematic in echo planar imaging (EPI)-based DWI [39]. It was recently shown that ADC quantification and geometric accuracy on a 0.35 MRI-guided linac (MRL) depended on gantry angle and was markedly inferior to ADC quantification on higher-field diagnostic systems [40]. Moreover, since lower magnetic field has reduced signal-to-noise ratio (SNR) [41], sensitivity concerns, particularly at higher b-values, called into question the feasibility of reliable ADC mapping since noisy data affects accuracy/repeatability of the exponential fit [42].

In this work, ADC maps of a National Institute of Standards and Technology (NIST)-traceable diffusion phantom are acquired on a 0.35 T MRL using single-shot EPI-based [43] diffusion. These are tested for quantitative accuracy over a range of clinically-relevant ADC values and repeatability over multiple measurements, with results compared to those obtained on a 3.0 T diagnostic scanner. Additionally, geometric precision is interrogated via comparison to spatially precise computed tomography (CT) scans of the same phantom. The in vivo longitudinal potential of this protocol is illustrated in five patients with soft-tissue sarcoma treated on the MRL and receiving same-day diagnostic DWI scans at one or more timepoints in the patients’ treatments.

2. Materials and Methods

2.1. Phantom preparation

A NIST-traceable diffusion phantom (CaliberMRI, Boulder, CO) was imaged. This phantom is a water-filled sphere of radius 9.7 cm and contains thirteen 30-mL cylindrical (2.9 cm inner diameter, 5.0 cm height) vials with varying concentration of polyvinylpyrrolidone (PVP) from 0 to 50 % in 10 % increments. Varying PVP concentration systematically varies ADC. A schematic illustrating the vial PVP concentrations and arrangement within the phantom is in Fig. 1A. With known temperature, NIST-traceable reference ADC can be deduced for comparison to measured ADC. Three phantom measurements were performed on a 0.35 T MRL on different days over three months (constant temperature 22.0° C). Three additional phantom measurements were provided by the phantom vendor for a 3.0 T diagnostic scanner on different days (constant temperature 19.5° C).

Fig. 1.

The accuracy and repeatability of ADC quantification performed on a 0.35 T MRL is depicted and compared to the performance of a 3.0 T diagnostic scanner. 1.A shows a schematic for the diffusion phantom used which contains thirteen vials of differing concentrations of PVP. 1.B demonstrates the measured values of ADC in each vial over three separate measurements (gray) compared to the reference ADC (cyan) in the relevant vial. Here, ADC quantification is accurate and repeatable with significant discrepancies (as denoted by a star) only seen in vials approaching free diffusion where the SNR is low. Similarly, the accuracy and repeatability of ADC mapping on a typical 3.0 T diagnostic scanner is shown in 1.C. Standard deviations of the measured values reflect the variation in ADC over each voxel in a given vial. Standard deviations of the reference values were provided by the phantom manufacturer. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.2. Image acquisition and ADC calculation

Images were acquired on a 0.35 T ViewRay (Oakwood Village OH, USA) MRIdian MRL with gantry at 0° and couch electronics disabled to minimize deleterious RF noise/interference. A pair of 6-channel phased-array receive surface body coils [44] were used for all phantom and patient imaging. A multi-slice EPI diffusion sequence was applied (matrix size = 100x100x21, FOV = 350x350x190 mm³, 6 mm slice thickness, 3 mm slice gap, TR = 3200 ms, TE = 120 ms, BW = 1352 Hz, α = 90°, 6 averages). GRAPPA-based parallel imaging [45] with an acceleration factor of two was applied to accelerate acquisition and reduce geometric distortions by shortening the echo train [46]. Acquisition took 4.32 min. Phase encoding was anterior-posterior (assuming head-first supine). The b-values were 0, 200, 300, 500, and 800 s/mm² with diffusion weighting applied along three principal directions. For a given diffusion direction, voxel ADC was calculated via exponential fit (MATLAB vR2021a, MathWorks, Inc, Natwick MA, USA) as in Equation (1).

$$S\left(b\right)=S_0{e}^{-bD}$$ (1)

S(b) is a voxel signal intensity at b-value b. S₀ is voxel signal intensity at b-value b = 0. D is ADC value. This was applied along each principal direction and average ADC value over each direction in a voxel gave mean diffusivity.

In vivo diffusion images were also acquired using a 3.0 T MAGNETOM Vida (Siemens Healthineers, Erlangen, Germany) diagnostic scanner for comparison to MRL images. A multi-slice EPI diffusion sequence was used (matrix size = 120x120x52, FOV = 280x280x190 mm³, 3.0 mm slice thickness, 0.75 mm slice gap, TR = 12600 ms, TE = 63 ms, BW = 1603 Hz, α = 90°, 2 averages, 2.23-minute acquisition). Two-fold GRAPPA was applied. This clinical protocol utilizes b-values of 100 and 1000 s/mm² with diffusion weighting applied along a single direction (as opposed to all three principal directions). Equation (1) was applied unidirectionally to calculate ADC.

2.3. ADC accuracy and repeatability measurement in phantom

Three repeated scans were acquired on both the 0.35 T MRL and a 3.0 T diagnostic scanner. ADC was determined in each voxel using Equation (1), as described above. Mean ADC for a cylindrical (2.1 cm diameter, 2.7 cm height) region of interest (ROI) centered within each vial was calculated. Mean and standard deviation (σ) of mean vial ADC over repeat experiments were calculated and compared to known NIST-traceable values (provided for given temperature with associated uncertainty by vendor). A two-tailed t-test (MATLAB) between measured vial ADC values over the three experiments and corresponding reference ADC values was used. A measurement was deemed significantly different from expected value when p < 0.05.

The repeatability coefficient (RC) assessing agreement between repeated measurements, was quantified as [47]

$$\mathit{RC}=2.77\sigma$$ (2)

RC was calculated for each vial and averaged across all thirteen vials (mean RC) for both 0.35 T and 3.0 T.

2.4. Geometric distortion analysis

Phantom CT images were acquired on a Philips (Amsterdam, Netherlands) Brilliance 64 scanner in axial-mode (matrix size = 512x512x160, FOV = 240x240x270 mm³, 1.25 mm slice thickness) and taken as geometric ground truth for comparing phantom ADC maps produced from diffusion images acquired on both the 0.35 T MRL and the 3.0 T diagnostic scanner. Nineteen structures were separately contoured by three independent observers (RayStation v11A, RaySearch Laboratories, Stockholm, Sweden) on CT and both ADC maps: thirteen vials in the central slice (Fig. 2A-C) and six plastic screws 5.4 cm from the central slice (Fig. 2D-F). The distances between the geometric center for 78 vial pairs (central slice) and fifteen screw pairs (5.4 cm off-center) were determined. System-dependent geometric distortion was quantified as the difference in distance between a given pair of structures measured from an ADC map relative to the CT image [48]. Mean geometric distortion for each observer was calculated by averaging over all structures in each slice. A one-tailed t-test (MATLAB) determined geometric distortion statistical significance (0.35 T less than 3.0 T geometric distortion if p < 0.05)), both in the central and 5.4 cm off-center slices. The slice 5.4 cm off-center was chosen for its discernible phantom features. Mean geometric distortion standard deviation taken over the three observers provided a measure of uncertainty for this geometric distortion quantification.

Fig. 2.

The geometric accuracy of ADC maps produced on a 0.35 T MRL and 3.0 T diagnostic scanner are compared. The distance between 93 pairs of phantom structures were measured on each system and compared to their distances measured on a CT. Mean geometric accuracy is shown to be superior relative to the 3 T diagnostic scanner and submillimeter on the 0.35 T MRL in both a central slice and a slice that is displaced by 5.4 cm from the central slice. Here, the mean was taken over each of the 93 measurements and over each of the three observers.

2.5. Sarcoma imaging

DWI was performed on five patients with high-grade soft-tissue sarcoma of the thigh on the institutional review board (IRB)-approved Habitat Escalated Adaptive Therapy (HEAT) protocol (NCT05301283), a phase 2 clinical trial utilizing functional imaging-defined habitats to identify radioresistant tumor subpopulations. Imaging was in treatment position, with one patient imaged at three timepoints: pre-treatment simulation (day 0), mid-treatment (day 21), and three weeks post-treatment (day 69). The other four patients were imaged during simulation (pre-treatment) only. For all patients/timepoints, same-day diffusion-weighted imaging was performed on a 3.0 T diagnostic scanner. MRL anatomical imaging used a balanced steady-state free precession (bSSFP) sequence [49], [50]. The gross tumor volume (GTV) was delineated by a radiation oncologist on the bSSFP images. The ADC maps (MRL and diagnostic) were rigidly registered to anatomical images (Mirada Medical, Oxford, UK). Mean ADC within the GTV was determined for each image. The mean and standard deviation of the mean ADC values across the five patients was determined.

3. Results

Phantom ADC quantification was accurate on the 0.35 T MRL for clinically relevant ADC values, with discrepancies beyond measurement uncertainty only in vials approaching free diffusion (Fig. 1B). Three measurements at different times are in gray and compared to reference ADC (cyan). Phantom ADC measurements at 0.35 T are in Table 1. Four of thirty-nine measurements (thirteen vials, three repetitions) on the 0.35 T MRL deviated significantly (asterisk in Fig. 1B) from reference ADC. These were all three measurements in vial 11 (p = 0.026, 0.020, and 0.029) and one measurement in vial 12 (p = 0.032). Both vials have 0 % PVP concentration and high ADC, which is expected to have diminished accuracy [51] due to lower SNR (increased signal loss due to greater diffusion). The other thirty-five measurements did not deviate significantly from reference ADC. At 3.0 T, ADC quantification was accurate (no significant deviations from reference) across all ADC values (Fig. 1B), likely due to increased SNR at higher field strength. ADC quantification was repeatable on the 0.35 T MRL, with small discrepancies again only seen at high ADC. Mean RC for the 0.35 T MRL versus 3.0 T diagnostic scanner was 68x10^-6 mm²/s versus 57x10^-6 mm²/s.

Table 1.

		*All ADC measurements are given in units of 10-6 mm2/s
Vial	PVP Concentration	ADC Measurement 1	ADC Measurement 2	ADC Measurement 3	Reference ADC
1	50 %	278 (±74)	294 (±65)	296 (±65)	293 (±9)
2	50 %	293 (±53)	300 (±64)	291 (±68)	293 (±9)
3	40 %	553 (±45)	571 (±42)	578 (±39)	545 (±14)
4	40 %	553 (±45)	571 (±42)	578 (±39)	545 (±14)
5	30 %	870 (±29)	884 (±45)	883 (±38)	886 (±21)
6	30 %	853 (±41)	844 (±54)	853 (±40)	886 (±21)
7	20 %	1221 (±38)	1190 (±60)	1222 (±52)	1258 (±28)
8	20 %	1165 (±48)	1189 (±59)	1215 (±33)	1258 (±28)
9	10 %	1516 (±76)	1579 (±47)	1593 (±56)	1640 (±36)
10	10 %	1483 (±70)	1630 (±32)	1649 (±40)	1640 (±36)
11	0 %	1781 (±95)	1750 (±89)	1849 (±72)	2106 (±45)
12	0 %	1890 (±93)	1840 (±83)	1897 (±93)	2106 (±45)
13	0 %	2010 (±80)	2040 (±53)	2032 (±58)	2106 (±45)

*All ADC measurements are given in units of 10^-6 mm²/s.

In the central slice, average system-dependent geometric distortions were 0.35 (±0.02) mm and 0.85 (±0.02) mm for the MRL and diagnostic system, respectively (Fig. 2G). In the slice 5.4 cm off-center, these were 0.66 (±0.04) mm and 2.14 (±0.07) mm, respectively (Fig. 2G). 0.35 T MRL ADC maps are significantly more geometrically precise than the 3.0 T ADC maps, in both the central (p < 0.0001) and off-center slice (p < 0.0001). Geometric distortion can be visualized in Fig. 2A-F.

Longitudinally acquired in vivo sarcoma images for both systems provided sufficient image quality to visualize intratumoral spatially-varying ADC features. ADC maps for all three imaging timepoints (axial and coronal planes) for one patient along with MRL anatomical bSSFP images are shown with the radiation oncologist-delineated GTV (Fig. 3) The general region of enhanced ADC on 0.35 T MRL images demonstrates marked conformity with the anatomically derived GTV. Mean GTV ADC measurements of 800x10^-6 (±450x10^-6) mm²/s, 1020x10^-6 (±490x10^-6) mm²/s, and 820x10^-6 (±470x10^-6) mm²/s were obtained for the three timepoints, respectively. Moreover, the general pattern of ADC heterogeneity in the diagnostic ADC maps is qualitatively reproduced in MRL ADC maps. DWI on the MRL is robust to motion, as motion-artifacts in the post-treatment MRL anatomical images are not discernable in resulting MRL ADC maps. However, tumor ADC heterogeneity and anatomical features outside of GTV are less discernible at 0.35 T than at 3.0 T. In five sarcoma patients imaged on the 0.35 T MRL pre-treatment, mean GTV ADC was 910x10^-6 (±100x10^-6) mm²/s. MRL anatomical bSSFP images, 3.0 T ADC maps and 0.35 T ADC maps are shown for each patient (Fig. 4).

Fig. 3.

The feasibility of in vivo diffusion imaging on a 0.35 T MRL and its longitudinal potential are illustrated in a sarcoma patient who received same-day diagnostic diffusion scans at three timepoints, shown in axial and sagittal views. The region of enhanced signal in the ADC maps acquired at 0.35 T exhibits notable conformity with the tumor contours delineated on an anatomical bSSFP image (shown in green) and similarity with tumor ADC features seen in the ADC maps acquired at 3 T. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4.

Five sarcoma patients with disease of the upper thigh are imaged. The top row displays the set of anatomical bSSFP images with the tumor contours delineated in green. The middle row is a set of ADC maps acquired on a 3.0 T diagnostic scanner. The middle row is a set of ADC maps acquired on a 0.35 T MRL on the same day as their corresponding diagnostic ADC map. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Discussion

In this study, an EPI-based diffusion protocol demonstrated notable improvement in ADC quantification and repeatability. These ADC maps were geometrically precise (submillimeter precision) despite the susceptibility of EPI-based protocols to eddy current-induced geometric distortion. Also, the in vivo potential of this technique is illustrated.

To the best of the authors’ knowledge, there has been only one previous EPI-based study performing DWI on a 0.35 T MRL. Lewis et al. [40] quantified ADC accuracy as a function of gantry position and assessed geometric distortions. The current study, however, utilizes a new diffusion acquisition protocol and demonstrates a considerable improvement in ADC quantification accuracy. For example, for PVP concentrations providing ADC values in the range 200-300x10^-6 mm²/s, Lewis et al. reported minimum ADC deviations from reference of 26.6 %, while the current study observed mean ADC deviations from reference of 0.4 %. Similarly, for PVP concentrations providing ADC values in the range 800-900x10^-6 mm²/s, Lewis et al. reported minimum ADC deviations from reference of 7.2 %, while the current study observed mean ADC deviations from reference of 2.4 %. Additionally, the current study provides robust geometric distortion assessment with distances between 93 pairs of structures in two planes assessed (three pairs of structures in the central plane were analyzed in Lewis et al. [40]). Specific sequence details were not provided in Lewis et al. to speculate why the current study outperforms their results. The novelty of the current study lies not in the EPI pulse sequence applied, as this is commonly used in DWI, but in that this is the first study to demonstrate accurate/repeatable ADC mapping on a low-field MRL.

Mean ADC across five sarcoma tumors was 910x10^-6 (±100x10^-6) mm²/s, which in agreement with 1.5 T published values. Oka et al. [52] measured mean ADC of malignant soft-tissue tumors to be 920x10^-6 (±139x10^-6) mm²/s. In a later study, the same group [53] measured mean malignant soft-tissue tumor ADC of 880x10^-6 (±200x10^-6) mm²/s. Razek et al. [54] measured malignant soft tissue tumors of extremities mean ADC of 1020x10^-6 (±300x10^-6) mm²/s.. The analogous value of 910x10^-6 in this work on a 0.35 T MRL is in agreement, suggesting promising in vivo accuracy.

An alternative turbo spin echo (TSE)-based DWI approach [55] on a 0.35 T MRL [56] produced distortion-free ADC maps on the tri-cobalt system. TSE-based approaches suffer less from geometric distortion than EPI-based approaches (multiple 180° refocusing pulses allow less phase accumulation than during the long single-shot EPI echo train [57], [58]. However, this TSE-based approach on the MRL was plagued by signal-dropout artifacts [59] that may have been from eddy currents produced by linac gantry electrical components or concomitant gradients (extraneous magnetic gradients produced to satisfy the Ampere-Maxwell equation [60]), but further investigation is necessary to confirm this.

DWI has been performed using a 1.5 T MRL [51], [61], [62], [63] and may be an option at MR simulation [64]. While the 1.5 T MRL system allows ADC accuracy/repeatability quantification [61] similar to the current study, geometric accuracy results are not reported. The current study demonstrates not only ADC accuracy/repeatability but also submillimeter precision even in objects > 7 cm (Euclidean distance) from isocenter. Although DWI on a 0.35 T MRL is not yet as established as on a 1.5 T MRL, spatial precision dependence on field strength [65] alluded to in this work may indicate an advantage of incorporating DWI on the lower-field MRL system due to the importance of spatial precision in radiotherapy.

In the literature, DWI accuracy at high b-values using a single-shot EPI approach has been controversial due to lower sensitivity at 0.35 T relative to clinical field strengths [55]. Measuring ADC map SNR is nontrivial, since it is derived from exponential fitting. This study addressed this by measuring repeatability [66] of 0.35 T MRL ADC quantification. The introduction of noise diminishes the exponential fit accuracy/stability and, subsequently, results in larger ADC variance over repeated measurements. However, despite having lower sensitivity than the 3.0 T diagnostic system, the 0.35 T MRL repeatability coefficient was only slightly inferior. Thus, sensitivity limitations inherent in operating at 0.35 T do not preclude repeatable ADC measurements.

The current study only assesses system-dependent geometric distortion. Additional distortions are created when introducing a patient into the magnetic field, due to magnetic susceptibility effects [67] and chemical shift [68]. This represents a potential limitation of this study since patient-dependent geometric distortions were not assessed. Both effects scale with magnetic field strength [69], [70] and would, thus, be smaller on a lower-field MRL. However, since EPI-based approaches are particularly prone to susceptibility effects [71], patient-dependent geometric distortion in this context should be investigated in future studies. Another limitation of the distortion analysis in this study is that it only measured the centroid-to-centroid distance between structures, which is not sensitive to distortions in object shape.

Another limitation of this study is that b-values are not optimized. Using five b-values along each direction may not increase ADC quantification accuracy, and acquisition duration could be reduced (or SNR increased with more averages) by reducing these measurements. Moreover, lower/upper b-value limits have not been investigated. At low b-values, the phenomena of diffusion and perfusion become coupled [72], and signal decrease departs from mono-exponential behavior [73]. This is typically observed at b-values less than 100–150 s/mm² [74]. This would not manifest in phantom measurements (perfusion not present), but its effect on in vivo ADC quantification needs investigation. Optimal maximum b-value should be determined. Consensus recommendations for DWI on the 1.5 T MRL [75] advise against b-values > 500 s/mm² due to gradient limitations and sensitivity concerns. While images acquired in the current study using a b-value of 800 s/mm² qualitatively looked sufficiently sensitive, a deeper investigation into how lower sensitivity at higher b-values affects ADC quantification is warranted. Other imaging parameters such as echo time, repetition time, voxel size, and bandwidth all still need to be more rigorously optimized for application-specific image quality.

Another limitation of this study is the fact that the imaging parameters for DWI on the 0.35 T MRL and the 3.0 T diagnostic scanner differ. For the latter, routine protocol parameters were used. For the MRL, these parameters were modified due to lower overall sensitivity of the low-field system. For example, a larger slice thickness and lower maximum b-value were used on the MRL because the parameters in the diagnostic scan would result in unacceptable sensitivity reduction. It is not the intent of this manuscript to directly compare the lower-field MRL and higher-field diagnostic system. Instead, it illustrates the feasibility of accurate/repeatable DWI on a 0.35 T MRL while highlighting its advantages (geometric precision, logistics) and disadvantages (sensitivity) relative to the 3.0 T diagnostic system.

It should also be noted that ADC heterogeneity is only discernible within the GTV on the ADC maps acquired at 0.35 T. While this is not entirely understood, it could be due to the fact that the SNR is highest in the tumor because the coils were placed directly adjacent to the tumor for each patient. Further investigation is warranted.

Note, the NIST-traceable ADC values used as reference were specified at a field strength of 3.0 T [76]. However, ADC quantification can change slightly with field strength [77], although other studies are less conclusive [78]. Since 0.35 T is an uncommon field strength, no quantification standards exist for ADC at 0.35 T, making the NIST-traceable values at 3 T a best approximation.

In conclusion, the acquisition of accurate, repeatable, and geometrically precise (sub-millimeter distortion; greater than two-fold improvement over 3.0 T) ADC maps is possible at 0.35 T with an EPI approach. This enables tracking of longitudinal changes in tumor cellular density over the course of treatment on a low-field MRL. This may help facilitate biologically-guided online plan adaptation based upon a tumor’s dynamic physiologic changes over the course of treatment.

Author contribution

All authors contributed to the writing and preparation of the manuscript. JW, JA, IMO, SN, KL, and GR participated in the data collection. JW, JMB, and GR analyzed the geometric distortion data. CLL helped with the statistical analyses. TA designed the pulse sequence. JRC and AON designed the diagnostic imaging protocol. JMB, JMF, KY, AON, and SAR provided the clinical perspective for the manuscript. KL, EGM, and IMEN provided the technical perspective for the manuscript. SAR and GR provided the vision for the manuscript from the clinical and technical sides, respectively. JW led the project.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: JW, the first and corresponding author, has nothing to declare. Amongst the coauthors, TA was an employee of ViewRay, Inc at the time this work was performed and owned ViewRay, Inc stocks at that time. EGM, JMF, and SAR have been supported by a grant/contract from ViewRay, Inc. KL and SAR have consulted for ViewRay, Inc.