Introduction

Hypoxia is associated with tumour progression and treatment resistance. Oxygen-enhanced (OE)-MRI is a non-invasive method for mapping and quantifying the extent of regional hypoxia¹. Current analysis methods either measure oxygen-induced change in longitudinal relaxation rate (ΔR₁) or identify tumour regions that are perfused but insensitive to oxygen change (oxygen refractory (pOxy-R)), by combining OE-MRI with a perfusion measurement such as dynamic contrast enhanced (DCE)-MRI². Both analyses have previously detected therapy induced changes in tumour hypoxia in patients with non-small cell lung cancer³. The advent of MR-guided radiotherapy equipment has raised the potential of applying the technique on an MR Linac (MRL) system. We sought to develop OE-MRI in head and neck (H&N) squamous cell carcinoma and then perform a first-in-human translation of OE-MRI onto the MRL, with a longer term aim of using the technique for biologically adaptive radiotherapy⁴.

Methods

Participants were recruited after ethics approval and written informed consent. Imaging was performed on a 1.5 T Philips Ingenia MR-RT (MRSim) and a 1.5 T Elekta-Philips Marlin MRLinac (MRL) system. Analysis was performed in MATLAB (Mathworks).

MRSim: Optimised sequence parameters for T₁ mapping and OE-MRI are provided in table 1. OE-MRI gas delivery was administered during dynamic series: scans 1-25 (medical air; 21% O₂), 26-70 (100% O₂) and 71-91 (medical air; 21% O₂). A low-flow air-oxygen gas blender (Inspiration Healthcare) ensured 15 l/min flow via a high concentration mask (EcoliteTM, Intersurgical Ltd). The protocol was run in 6 healthy volunteers, each imaged twice (8 ± 3 days apart), and in 4 patients with H&N carcinoma, imaged 1-2 times pre-treatment and then once during chemoradiotherapy (CTRT) at weeks 2-4. In patients, DCE-MRI was acquired after OE-MRI, with sequence parameters provided in table 1 during injection of gadoterate meglumine (Dotarem, Guerbet) at the 8th image, 0.2 ml/kg at 3 ml/s.

T₁ maps enabled conversion of OE signal change to ΔR₁ (= R_1,O2 – R_1,air). In volunteers, regions of interest (ROIs) were positioned in the nasal concha (NC), tongue and brain. Mean ΔR₁ values and within-subject coefficient of variation (wCV) were obtained for each region. Patient tumour volumes were delineated on T₁ weighted post gadolinium (Gd) contrast images. ΔR₁ was measured for each tumour and NC (for quality assurance; QA). Perfused tumour voxels (DCE signal increase p < 0.05) were classified as oxygen enhancing (Oxy-E) (suggesting normoxia) or oxygen refractory (Oxy-R) (suggesting hypoxia), based on t-test analysis of air and 100% O₂ phases.

MRLinac: Hardware fitting steps included retrofitting of gas delivery in the MRL bunker; installation of gas ports, use of an MR conditional blender (Inspiration Healthcare); and installation of a contrast agent power injector (Bayer-Medrad). MRL sequences were modified slightly due to hardware differences with the intention of matching sequence parameters where possible as shown in table 1. Identical protocols for gas and contrast agent delivery were used on both systems. 4 healthy volunteers and one patient with H&N carcinoma were scanned.

NC ΔR₁ was measured in 4 volunteers and compared with those acquired in the averaged MRSim volunteer data (=12 datasets, i.e. 6 volunteers x 2 visits). The patient tumour volume was delineated on a T₁ post contrast image and ΔR₁ and pOxy-R measurements obtained. Tumour ΔR₁ curves were compared between the MRSim pre-treatment patient scan data (i.e. the average of 5 pre-treatment datasets from patients 2, 3 and 4) and the MRL patient.

Results

MRSim: Volunteer ΔR₁ curves for the two visits for the NC, tongue and brain regions are shown in Figure 1. Mean ΔR_1,NC = 0.059 ± 0.027 s^-1 (p < 0.001); ΔR_1,Tongue = 0.001 ± 0.012 s^-1 (p = 0.86); ΔR_1,Brain = 0.005 ± 0.005 s^-1 (p = 0.04) (Figure 1A-C). wCV for NC was 21.0% (Figure 1D). Patient ΔR_1,NC = 0.056 ± 0.026 s^-1 (p < 0.001). ΔR₁ increased in the primary tumour in all four patients from pre-treatment values to those measured after commencing CTRT (p = 0.04; Figure 2 A). pOxy-R was found to decrease in 3 of the 4 patients (Figure 2 B).

MRLinac: Volunteer ΔR_1,NC = 0.065 ± 0.030 (p < 0.001) had a similar trend to that measured on the MRSim (Figure 3A-B), no significant difference in ΔR_1,NC was observed (p = 0.6, unpaired t-test). Figure 3C-D shows tumour ΔR₁ curves from both systems. MRSim tumour mean ΔR₁ = 0.031 ± 0.035 (p < 0.001) (=5 visits from 3 patients). MRL tumour ΔR₁ = 0.035 ± 0.011 (p < 0.001) (n=1 patient). First-in-human MRL hypoxia maps are presented in Figure 4 and show distinct ‘hypoxic’ and ‘normoxic’ tumour sub-regions.

Discussion and Conclusion

We investigated OE-MRI for mapping of hypoxia in the context of head and neck carcinoma patients treated with chemoradiotherapy. Healthy volunteer data identified the NC as a useful QA reference to confirm successful gas delivery. We then implemented OE-MRI in 4 patients with H&N cancer all of whom had an increase in ΔR₁ in the primary tumour during CTRT, consistent with hypoxia-reduction. Finally, we translated the technique onto the MRL, a system designed to allow anatomical assessment of tumours prior to daily delivery of RT. Collectively, our work is the first to demonstrate the potential feasibility of using OE-MRI for adaptive radiotherapy.

Acknowledgements

This work was supported by funding from Cancer Research UK and the Medical Research Council.