Diagnostic value of retrospective PET-MRI fusion in head-and-neck cancer

Clinical examination and imaging is routinely performed for the staging of head-and-neck cancer (HNC) in order to establish the tumour extent and size, to assess nodal involvement, to evaluate detailed tumour spread, e.g. bone infiltration and perineural infiltration, and to identify distant metastasis [1]. Accurate imaging is essential for staging, treatment planning and follow-up, since surgical intervention and neo- or adjuvant therapy modalities depend on the outcome of the diagnostic results [2]. For head-and-neck tumours, magnetic resonance imaging (MRI) and computed tomography (CT) provide accurate anatomical information with good resolution but have well-known limitations, especially in the staging of the nodal involvement of the neck [3, 4] and recurrent disease [5, 6]. Furthermore, the differentiation between malignant bone infiltration and infectious bone reaction is also of poor accuracy [7]. An advantage of MRI is that fewer artefacts occur from (dental) metallic implants, which often interfere with CT interpretation [8].

Nowadays multimodal imaging by using positron emission tomography/computed tomography (PET/CT) has gained a wide acceptance as a powerful imaging tool, especially in recurrent tumour disease. The combination of morphological and functional imaging (multimodal imaging) has been shown to reduce false positive or false negative results [9, 10]. Driven by the success of PET/CT in HNC treatment [11, 12], hybrid PET/MRI scanners are now available to combine the high soft-tissue contrast of MRI with the molecular and/or metabolic information of PET. Previous reports have described the development of reliable PET/MR imaging protocols [10, 13‐15] for HNC staging. However, to date, only a few sequential or fully integrated PET/MRI scanners are in clinical use. Most departments can still only offer the use of the two modalities (PET/CT and MRI) each as a single device. The separately gained images can be additionally analysed in two different ways: 1) side-by-side analysis and 2) retrospective software-based image fusion. The latter can be performed manually by shifting and rotating the images by using landmarks as reference [16, 17] or fully automatically [18]. Retrospective image fusion is, because of its complexity, time-consuming, hardware- and software-dependent and thus problematic in clinical routine. Furthermore, differences in the patient’s position during each imaging process limit the accuracy of the retrospective fusion. On the other hand, this method has been shown to lead to a more precise evaluation and treatment of HNC in complex cases [14, 19].

The purpose of this retrospective study was to assess the clinical value of retrospective PET-MRI fusion and to compare the diagnostic accuracy with side-by-side analysis MRI and ¹⁸F-FDG PET alone.

In order to determine the clinical benefit of PET-MRI fusion in HNC, we have analysed the value of retrospective PET-MRI fusion and compared the results with side-by-side analysis and PET or MRI alone. Although in our small patient population no statistically significant differences in the performance of each technique could be determined for local tumour assessment or lymph node staging, there was a trend towards a better performance of multimodal imaging (side by side analysis or retrospective image fusion) especially in the diagnosis of recurrent HNC and inconclusive findings in the single modality. Based on these data, this finding now has to be confirmed in larger prospective studies.

To our knowledge, no previous study has compared the value between retrospective image fusion and side-by-side analysis up to now. A diagnostic benefit has to be proven for multimodal imaging, since morphological imaging by using MRI or CT alone has nowadays reached a high accuracy in patient follow-up and even allows the detailed analysis of structures such as mucosal invasion [3, 23]. Accurate assessment of the local tumour extent is important for the surgeon, as the imaging results will determine the extent of tumour resection, the type of reconstruction and the adjuvant therapy.

In several studies, the beneficial use of PET data in primary T-staging has been reported [24, 25] and has led to the assumption that PET and MRI analysis further complement each other [6, 15]. In our study, one false negative and one false positive rating in MRI could be changed to the correct diagnosis through side-by-side analysis and elevated the sensitivity/specificity to 85%/100%. Retrospective image fusion had the highest sensitivity/specificity rates in T-staging of 89%/100%. It was shown to be of evident value in a suspected recurrent tumour (Figure 3). As is well-known, morphological imaging alone (MRI or CT) in previously operated regions is challenging due to sometimes difficult differentiation between non-neoplastic and neoplastic change. The reasons are scar tissue, loss of symmetry, side shift and unspecific contrast enhancement [5]. PET has gained wide acceptance in staging of suspected recurrent tumour disease, because of its high negative predictive values of up to 95%. On the other hand, it is also associated with false positive findings shortly after operation and should therefore be combined with CT or MRI and should be performed at a time interval of 10–12 weeks after surgery [26‐28].

In our analysis, retrospective PET-MRI fusion in initial tumour staging of primary lesions was of minor advantage compared with a single modality or to the side-by-side analysis with regard to T-staging. No significant benefit of retrospective image fusion was seen compared with side-by-side analysis. However, multimodal imaging (fusion and side-by-side analysis) had a tendency for higher sensitivity compared to MRI or PET alone, although the difference was not significant.

Another important aspect of HNC staging is the recognition of tumour invasion of neighbouring structures, such as maxilla or mandible. As described earlier, additional PET data might be beneficial in the detection of bone infiltration compared with MR images alone [29]. In one case of our present study, retrospective PET-MRI fusion could also reduce false positive diagnoses for bone infiltration and correspondingly could reduce surgical over-treatment.

Staging for metastases to cervical lymph nodes (N-staging) plays an important role and correlates directly with the survival of the patient [30, 31]. Correct diagnoses are still associated with difficulties, especially when only morphological imaging technique is performed [32]. The reported sensitivity/specificity ranges between 78-88%/75-86% for MRI alone [4, 33]. The differentiation between benign and malignant lymph node enlargement in morphological imaging alone (CT or MRI) mainly relies on size-based evaluation. However, the sizes of lymph nodes, whether benign or malignant, vary and previously studies have shown that up to 21% of nodes smaller than 10 mm can be malignant and up to 40% of nodes larger than 10 mm can be benign [32, 34]. Our high sensitivity for MRI (94%) can be explained by only differentiating between N₀ or N₊ staging and no node-by-node comparison. PET/CT is able to raise the specificity up to 97% but is still accompanied with a low sensitivity of 50-84% [33, 34]. In our results, PET analysis alone achieved the highest sensitivity and specificity rates (94% and 91%) and was higher than that reported in the literature. Our small patient population might well be the reason for these findings.

On a "N-staging per patient basis", MRI only achieved correct staging in 70.4%, whereas PET alone, side-by-side analysis and PET-MRI fusion achieved correct staging in 81.5% (Table 4). Furthermore, side-by-side analysis and retrospective image fusion achieved the lowest under-staging rates compared with PET and MRI alone (7.4% vs. 11.1%). On the other hand, MRI had the highest rates of over-staging (18.5%) compared with multimodal imaging (11.1%). Both retrospective image fusion and side-by-side analysis corrected two initial false positive ratings to N₀ staging. PET analysis alone was associated with the lowest rate of over-staging (7.4%). As shown in our results, even with retrospective PET-MRI fusion or side-by-side analysis, malignant neck lymph nodes smaller than 10 mm remained undetected and were associated with a lower specificity than by PET analysis alone (82% vs. 91%), as caused by the false positive influence of the MRI. Ng et al. also reported a superiority of PET analysis for nodal staging compared with MRI alone. By analysing the nodal status in a side-by-side fashion, they showed a trend of increased diagnostic accuracy over the single modalities [35].

Our study has some limitations. The main limitations are its retrospective design, the inhomogeneous patient population including patients with primary and recurrent disease and the relatively small number of patients included. However, these limitations do not decisively influence the results of our study. As the observers were not aware of the histopathology report, the retrospective nature of the study should not have influenced the diagnostic performance of image fusion. Furthermore, the aim of our study was to evaluate performance of image fusion of PET and MRI in the head and neck region. We therefore chose to include patients with diseases in this region, independently of the primary or the recurrent disease. Despite the relatively small number of patients included in the study, our results confirmed the known superiority of combining morphological and functional imaging in diagnostic accuracy. However there were no beneficial gains for the image fusion in this aspect compared to side-by-side analysis.

In summary, we found positive trends of multimodal imaging in T- and N-staging of HNC cancer. Nevertheless, the proof of cost-effectiveness of an initial multimodal imaging (PET/CT or PET/MRI) for primary staging of HNC and also any evident and significant advantages in diagnoses of bone infiltration are still pending and must be analysed systematically. Another question to be answered is, whether it is contemporary to perform time consuming and in some cases difficult retrospective image fusion or side-by-side analysis when it is possible to use fully integrated PET/MRI scanners in primary staging of HNC.

Our study has shown the beneficial use of multimodal imaging by using retrospective PET-MRI fusion in selected HNC cases only. Compared with morphological MRI alone, we have seen advantages in single cases of recurrent diseases and in the ambiguous diagnosis of suspected lymphatic spread to the neck. However, the complex and time-consuming nature of retrospective image fusion hardly justifies the routine use in light of the only slight advantages compared to side-by-side analysis. However this technical limitation can be overcome by fully integrated PET/MRI scanners, which have recently become available. Their clinical value for HNC has not as yet been fully defined [14, 36, 37] and must be analysed systematically in future studies.