Changes in multimodality functional imaging parameters early during chemoradiation predict treatment response in patients with locally advanced head and neck cancer

Radical chemoradiation (CRT) is widely accepted as the standard of care for organ-sparing treatment of locally advanced head and neck squamous cell carcinoma (HNSCC). However, it is now evident that locally advanced HNSCC represents a disease spectrum rather than a single entity, with variable response to standard CRT. Clinical variables such as tumor, node and metastases (TNM) staging and smoking history are prognostically robust but predictively deficient. Reliable prediction of outcome early during CRT is, therefore, highly desirable to avoid continuation of ineffective treatment and guide adaptation of therapy based on response.

Functional and molecular imaging (FMI) can characterize tumor phenotypes by providing quantitative parameters with radiobiological relevance. ¹⁸F–FDG-PET/CT and MRI both have the benefit of being non-invasive, allowing serial measurements during radiotherapy. Several correlation studies in HNSCC have demonstrated varied, but complementary, biological information from different FMI parameters [1, 2]. To date, most FMI biomarker studies in HNSCC have concentrated only on pre-treatment time points. Indeed, there remains a paucity of prospective data to define the optimal time point for early intra-treatment assessment in patients receiving CRT.

Here, we report the results of serial ¹⁸F–FDG-PET/CT, and diffusion-weighted (DW), dynamic contrast-enhanced (DCE) and susceptibility-weighted (SW) MRI following the first and second week of CRT in patients with HNSCC. The primary objective of this analysis is to identify the optimal timing and predictive value of early intra-treatment changes in FMI parameters for ultimate response to CRT.

We evaluated early intra-treatment assessment using multimodality FMI parameters to predict response to CRT in patients with locally advanced HNSCC. The reason for investigating the chosen study time points and not later, i.e. >2 weeks, is to allow early identification of patients who are likely or unlikely to respond so that the window of opportunity to affect therapeutic outcome is not missed. Responders could be considered for treatment de-escalation, e.g. radiotherapy dose reduction or target volume adaptation, to reduce treatment-related morbidity [8, 9]. In contrast, non-responders should be considered for treatment intensification, e.g. radiotherapy dose escalation [10], hypoxia modification [11], novel radiosensitisers [12, 13] and/or ‘bail out’ surgery.

As shown in previous studies [14, 15], treatment-induced changes in FMI parameters precede anatomical changes, allowing earlier risk stratification of patients. Our data demonstrated differing optimal times for early response assessment during CRT when FDG-PET and MRI parameters were used. Changes in tumor TLG_40% and SUV_max at week 1 were predictive of treatment outcome, but these signals, in fact, disappeared later by week 2. This was explained by the low tumor FDG uptake and lesser difference between the two groups by week 2. Another possible confounding factor is the influence of radiotherapy-induced peritumoral inflammation on FDG uptake with cumulative fractions, which may affect ROI segmentation (Fig. 3), but this phenomenon is typically observed during the latter part of radiotherapy [2, 16]. This raises uncertainties about the reliability of FDG-PET parameters in reflecting tumor response beyond the first week of CRT. It is also evident that combined, rather than isolated, analysis of PT and LN FDG-PET parameters provides a better overall representation of the tumor response. To the best of our knowledge, this is the first published data investigating the predictive value of early intra-treatment changes in FDG-PET parameters in such a setting.

In contrast to FDG-PET parameters, it was not until week 2 that treatment-induced changes in MRI parameters successfully discriminated responders from non-responders. Our data showed that a larger fractional increase in PT DW-derived ADC at week 2 (∆ > 17%) was highly predictive of favorable response to CRT. Overall, our results are consistent with previous DW MRI studies which cumulatively reported that an increase of tumor ADC (>14–24%) between week 1 and 4 of radiotherapy could predict treatment outcome [17‐20]. Kim et al. reported intra-treatment assessment at week 1 to be predictive of response, but their scans were performed on average 12 days after the start of CRT [19], which would have been defined as week 2 in our study. Thus, it is reasonable to deduce that whilst highly desirable, earlier assessment with DW MRI (e.g. <7 days from start of CRT) is premature and of limited utility in HNSCC. Moreover, our study improves on previous studies due to its homogeneity: previous studies used MRI of different strengths (1.5 and 3 T) within the same study, had less standardized scanning time points (standard deviation of >3 days) or/and included patients with early disease (stage I-II) undergoing radiotherapy only.

We also found responders to have a larger fractional increase in PT DCE-derived K^trans and V_e. Similarly, these were only evident by week 2. These observations are likely to reflect early cell degradation in responding tumors, resulting in expansion of interstitial space and increased vascular permeability. Unlike DW MRI, there is limited data on the role of intra-treatment DCE MRI to assess and predict response to CRT. This may be related to the technical difficulties, e.g. tumor motion due to swallowing and workload required to process DCE data. We are aware of only two pilot studies which assessed changes in DCE parameters during radical radiotherapy in patients with HNSCC. Cao et al. reported an increase in PT blood volume (BV) 2 weeks into CRT to be associated with local control [21]. Baer et al. subsequently investigated a novel method of using parametric response maps of DCE MRI to predict survival following CRT in 10 patients: they found patients with a large percentage of PT gross volume that decreased in K^trans after 2 weeks were more likely to have significantly reduced survival [22]. Our larger study supported their findings that intra-treatment changes in K^trans is a potential biomarker in predicting treatment response.

In this study, we also investigated the role of SW-derived R₂* as a predictive biomarker in HNSCC. SW MRI is an alternative, hypoxia-dependent, non-invasive imaging technique that exploits the paramagnetic properties of deoxyhaemoglobin in erythrocytes to create contrast. Our interest in SW MRI stems from pre-clinical data [23] and a previous clinical study in cervical cancer demonstrating the ability of baseline tumor R₂* to predict response to CRT: responders had a lower average baseline R₂* than non-responders [24]. This result was not replicated in our study and we did not find any apparent trends for alterations in R₂* during the first 2 weeks of CRT. The only other R₂* study in HNSCC was recently published by Min et al. and they also failed to demonstrate any clear pattern in its weekly changes throughout radiotherapy [25]. They did not correlate R₂* with treatment outcome, but as with our observation, it was evident that R₂* does not appear to have a predictive role in HNSCC as a standalone parameter. A possible explanation is that R₂* values are strongly dependent on tumor BV [26, 27], which is highly heterogeneous in HNSCC. An accurate and robust measurement of tumor BV is challenging [28, 29]. Therefore, additional work is required to ascertain how best to interpret R₂* measurements with BV before it can be utilized as a hypoxia imaging biomarker in HNSCC.

Attempts to combine multiple identified FMI parameters failed to yield superior predictive power over a single parameter (∆ADC at week 2) in this cohort. This may partly be due to the relatively small number of non-responders in our study (8/35, 23%). The risk of treatment failure is not truly binarised by a single parameter threshold, and in ‘real-life’ clinical practice, intra-treatment changes in other predictive parameters (TLG_40%, K^trans and V_e) may prove useful in further determining the risk in equivocal cases. Our study has provided the basic framework for early intra-treatment assessment with FMI in locally advanced HNSCC, but requires further refinement and validation with more patients. This work continues and we are expanding our PET and functional MRI database beyond the current study cohort.

There are limitations of this study. Eight patients with T1–2 tonsillar cancer did not have measurable PT following diagnostic tonsillectomy and it is unclear whether this would have an impact on the result. In addition, the cranio-caudal coverage of our MRI protocol meant that in five patients, involved LNs outside the FOV were excluded. However, the largest LN for each patient was included, which is likely to have been representative of the dominant tumor biology. Another difficulty was the requirement to exclude obviously necrotic or cystic regions of the tumor, which was performed manually.

Our study highlighted the importance of intra-treatment scanning time points when integrated into clinical practice due to its impact on prediction outcome. This study provides the framework of utilizing multimodality FMI early during CRT and could be used to inform the design of future risk-stratified adaptive interventional studies in HNSCC.