Introduction

Dynamic Contrast Enhanced (DCE) Magnetic Resonance Imaging (MRI) has seen potential for predicting head and neck (H&N) tumour response to a treatment using pharmacokinetic (PK) modelling parameters [1]. PK-perfusion based parameters include the transfer constant from plasma (K^trans) and tissue volume fraction (V_e). These can be impacted when the tissue and vessels supplying it are affected by disease. The process of perfusion is dependent on the blood Haematocrit (Hct) value and Arterial Input Function (AIF), which describes the rate of contrast agent uptake into a region of interest (ROI) [2]. There exists debate in the literature to whether individualised or population-based Hct or AIF’s should be used for the generation of PK parameters for response monitoring, including for head and neck tumours [1-3]. This study aims to quantify the variations in PK-parameter values generated by implementing these different inputs to the model.

Methods

Imaging: 5 patients with diagnosed H&N cancer were imaged using a 3T MRI scanner at various timepoints along treatment: baseline, weeks (W) 2, 5, and 20 (post-3 months); using a 3T Siemens Magnetron Skyra scanner (Siemens Healthineers, Erlangen, Germany). T₁-spoiled 3D-GRE Volumetric interpolated breath-hold examination (VIBE) sequences were used to generate T₁-Maps in post-processing using the Variable Flip Angle (VFA) method [4] (FA: 2° and 15°) on 3D-Slicer. Common acquisition parameters for the VIBE and dynamic acquisition (TWIST) included: BW=440Hz/pixel, TR/TE=4.09/1.35 ms, FOV=30 cm², slices = 26 and thickness = 3 mm. The latter scan utilized a Gadovist based contrast agent (0.1ml/kg, injected at a rate of 4 ml/s after the 3rd measurement).

Analysis: Primary tumour contours were generated by a clinician on the TWIST images. These were used as the ROI from which the mean (95% confidence interval) for each PK-parameter map were extracted from. Such parameter maps were generated using the Tofts-Kermode Model [5] in 3D-Slicer, and included maps of: K^trans and V_e. Population AIF’s (PA) were provided by 3D-Slicer while individualized AIF’s (IA) were generated by using the uptake under a 61 mm³ label on the carotid artery. A population Hct (PH) of 0.42 for the patient cohort converged with the literature [3], whilst individualized Hct (IH) values were manually extracted from the patient's blood results (acquired within 3 days of scanning). The PK-parameters were generated using four input parameter combinations: PA-PH, PA-IH, IA-IH, IA-PA. Percentage differences between the baseline and subsequent weeks were normalized to the baseline value to observe the differences in each parameter over the course of the treatment, as would be required for treatment response applications.

Results

Similar trends for each combination were observed for most patients (Figure 1). The differences become apparent when taking the ratio of the average % difference increase in values (Figure 2) generated for baseline and W₂₀ for [V_e, K^trans] respectively: IAPH/PAPH= [39%, 500%] and IAIH/PAIH= [32%, 480%]. This suggests IA’s positively scale the parameters, with K^trans to a greater extent. IH variations increased for later treatment scan outputs also.

Discussion

The findings in this study suggest that the input parameters (population versus individualized) of AIF and Hct to a PK-model are significant when quantifying the generated output PK-parameter maps. Although the IA dependency was determined to be a scaled increase of the specific parameters, the PA used for the comparison could alter this finding [3]: Utilizing a PA specific to the patient cohort under investigation should be considered. Further, due to the H&N anatomy, the task of identifying of the main vessel supplying the primary tumour was simplified with the carotid artery displaying a clear uptake in most dynamic acquisitions. This could have contributed to the common trendlines observed and may not hold true for tumours located in other body regions.

An observed increase in IH variations as treatment progressed in part was caused by the reduction in IH for patients (baseline - W₂₀: ∆IH_avg= - 0.04). Thus, larger differences existed between IH and PH values. The literature in general only utilizes a patient cohort-PH generated from the baseline scan [2, 3]. A further study would involve generating a PH for each week from the patient cohort and comparing to weekly IH measurements for potential increased accuracy.

The 5 patients selected were nominated for this preliminary study since they all had successful scans acquired at the same 4 timepoints over the course of treatment. Rejections from the study included inadequate FA-scans or IH values not being obtained. Other inaccuracies potentially affecting interscan variations were not considered in this study but should be further investigated: including those generated by image distortions (B₀-inhomogeneities, and susceptibility). In particular, a future study investigating the B₁-inhomogeneity effects on the VFA acquisitions is to be completed to estimate any potential impacts to the T₁-Map and hence resultant PK-parameter values generated.

Conclusion

This study highlights the large variations and difficulties which can occur during DCE-MR image post-processing. If consistency in a methodology is followed, relative changes between weeks to predict response is possible. The small patient cohort size however in this study leaves the question for the most suited method of PK-parameter map generation (individualized or population-based inputs) unclear. Further studies with increased time-points and patient numbers are required for validation.

Acknowledgements

This work was made possible with access to the 3T MRI scanner at Liverpool Hospital provided by staff in the Cancer Therapy Centre working in collaboration with Ingham Institute for Applied Medical Research (Physics). Further, this research was in part funded by the South Western Sydney Local Health District (SWSLHD) Top-Up Scholarship (2019- Madeline Carr) and Cancer Institute NSW Early Career Fellowship (2019- Michael Jameson).