Introduction

Benign hyperplasia causes generalised thickening of the mucosa/submucosa along the nasopharyngeal walls and enlargement of the adenoid and is a common incidental finding on MRI of the adult head and neck, as well as on brain imaging [1]. Benign hyperplasia may need to be distinguished from early-stage nasopharyngeal carcinoma (NPC) confined to the nasopharynx. This is especially relevant in NPC endemic regions because the current trend of using plasma Epstein–Barr virus (EBV) DNA to screen asymptomatic subjects for NPC has led to an increase in the number of tumours detected at an early stage [2]. In most cases, these entities can be distinguished using anatomical images; lymphoid hyperplasia causes diffuse wall thickening symmetrically involving both sides of the nasopharynx, and when there is hyperplasia of the adenoid, the adenoidal stripes tend to be preserved, whereas early-stage NPC tends to cause a focal mass and loss or distortion of the adenoidal stripes [1, 3, 4]. However, in a small number of patients, lymphoid hyperplasia may cause asymmetrical thickening and the adenoidal stripes may not be identified, especially if the scan is not a dedicated scan of the nasopharynx. Furthermore, we have shown recently that 12% of stage T1 NPCs cause diffuse symmetrical thickening of the nasopharyngeal walls similar to that in benign hyperplasia [4].

The ability of MRI to accurately distinguish benign hyperplasia from NPC is also becoming more important because MRI is used more frequently to detect the 10% of NPCs that are occult on the endoscopic examination [3]. The detection of NPC by MRI is even more challenging in the screening setting, wherein plasma markers such as EBV DNA are used to identify early-stage NPCs in asymptomatic subjects [2].

Diffusion-weighted imaging (DWI) can distinguish benign from malignant head and neck tumours [5] and intravoxel incoherent motion (IVIM) DWI may be superior to conventional DWI for tissue characterisation, as the former can distinguish the process of pure diffusion from microcirculation in the capillary network. Indeed, an early study reported that IVIM DWI could distinguish NPC from benign adenoids in the nasopharynx [6]. If IVIM DWI is useful in the screening setting, it would not only improve the diagnostic performance of the current MRI protocol but potentially could replace the T1 contrast-enhanced scan which could expand the use of MRI.

Therefore, this preliminary study hypothesised that DWI has the potential to distinguish benign hyperplasia from early-stage NPC. This study was designed to examine whether IVIM-derived parameters, together with apparent diffusion coefficient (ADC) obtained from conventional DWI, are able to distinguish benign hyperplasia from early-stage NPC. The second part of the study compared diagnostic performance of these DW parameters to examine whether IVIM DWI had advantages over the conventional DWI in distinguishing between these two entities in the nasopharynx.

Methods

Patients

This prospective study was performed under approval from the local institutional review board. The patients were ethnically Chinese who were recruited between March 2015 and July 2018, and written informed consent was obtained from all patients.

Patients with early-stage NPC

Consecutive symptomatic and asymptomatic patients who met the following criteria were included: (1) newly diagnosed, biopsy-proven undifferentiated NPC; (2) primary tumour confined to the nasopharynx (stage T1) on MRI without superficial spread to the nasal cavity or oropharynx or deep spread within the nasopharynx; and (3) a primary tumour size > 5 mm.

Patients with benign hyperplasia

Consecutive asymptomatic patients who were undergoing follow-up investigations for benign hyperplasia were recruited from a previous NPC detection study with persistently raised plasma EBV DNA [2] who met the following criteria: (1) did not have NPC after the initial endoscopic examination and contrast-enhanced MRI and showed no change suggestive of NPC on the follow-up endoscopic examination and non-contrast-enhanced MRI two years later; (2) follow-up MRI examination showed benign hyperplasia > 5 mm in thickness.

MRI acquisition

MRI was performed using a Philips Achieva TX 3.0-T scanner (Philips Healthcare) with a body coil for radiofrequency transmission and a 16-channel Philips neurovascular phased-array coil for reception. IVIM DWI was acquired using a fat-suppressed, single-shot, spin-echo, echo-planar imaging sequence (number of slices: 9, repetition time/echo time, 2000/50 ms; field of view, 230 mm × 230 mm; section thickness, 4 mm; echo train length, 55; sensitivity encoding factor, 2; number of signals acquired, 4) and 14 b-values (0, 20, 40, 50, 60, 80, 100, 200, 300, 400, 500, 600, 800 and 1000 s/mm2). The total IVIM DWI scan time was 2 min, 30 s.

Anatomical MRI in all patients included a fat-suppressed T2-weigthed turbo-spin-echo sequence and a T1-weighted turbo-spin-echo sequence. The patients with early-stage NPC were also scanned using a T1-weighted turbo-spin-echo sequence following a bolus injection of 0.1 mmol of gadoteric acid (Dotarem; Guerbet) per kilogramme of body weight.

Imaging analysis

Olea Sphere (version 3.0; Olea Medical SA) was used for all the post-processing steps by implementing a Bayesian probability–based algorithm using all 14 b-values to fit a bi-exponential diffusion model to calculate the pure diffusion coefficient (D); the pseudo-diffusion coefficient (D*), which reflects blood microcirculation in the capillary networks; and the perfusion volume fraction (f), which correlates with capillary geometry and blood velocity [7], and using six b-values (0, 200, 400, 600, 800, 1000 s/mm2) to fit a mono-exponential diffusion model to calculate the conventional ADC (ADC0–1000). In addition, b-values of 300, 400, 600, 800 and 1000 s/mm2 were used fit a mono-exponential diffusion model for the purpose of comparing conventional ADC300–1000 with D in the IVIM model. On the ADC map, the region of interest was manually drawn around the whole primary tumour or benign hyperplasia with reference to the corresponding anatomical images by a radiologist with more than 20 years of experience with specialisation in the head and neck region and a researcher with 4 years of experience of treating NP disease. The region of interest on the ADC map was then fitted automatically to the other diffusion maps. Necrotic and cystic regions were excluded from the analysis during the delineation phase. The total volume of the abnormalities in the nasopharynx was obtained by manually outlining the cross-sectional area on each slice, summating the cross-sectional areas, and multiplying by the slice thickness of 4 mm.

Statistical analysis

Distributions of DW parameters in the early-stage NPC and benign hyperplasia groups were examined using Kolmogorov–Smirnov test. Differences in the DW parameters between the two groups were compared using Student’s t test (data normally distributed) or Mann–Whitney U test (data not normally distributed). A receiver-operating characteristic curve analysis and subsequent area under the curve (AUC) calculation were used to identify the optimal thresholds for the significant DW parameters by maximising the sensitivity plus specificity, and the diagnostic performance of these thresholds was calculated and the significance of these thresholds was re-evaluated with the chi-square test. The AUCs of significant DW parameters were compared using the Delong test. The above analyses used the average values of the parameters obtained from the 2 observers. Intra-class correlation coefficients (ICCs) with 95% confidence intervals (95% CIs) were calculated to assess the inter-observer agreement regarding the IVIM-derived parameters and ADC0–1000. ICC was also calculated to assess the agreement between D from the IVIM model and ADC300–1000.

All of the statistical tests were 2-sided, and a p value < 0.05 was considered to indicate a statistically significant difference. Analyses were performed using the statistical analysis software SPSS (version 25.0; IBM) and MedCalc Statistical Software (version 14.8.1; MedCalc Software).

Results

Patients

Thirty-nine patients with early-stage NPC confined to the nasopharynx and 32 patients with benign hyperplasia were recruited. Of these, 7 were excluded because of degraded images (NPC = 5 and benign hyperplasia = 2). The characteristics of the remaining 34 patients with NPC and 30 patients with benign hyperplasia are shown in Table 1.

Table 1 Characteristics of the 64 included subjects
Full size table

DW parameters

The DW parameters of subjects with early-stage NPC and benign hyperplasia are shown in Table 2. DW parameters in each group was normally distributed (p = 0.072–0.200), and compared with the subjects with benign hyperplasia; those with early-stage NPC had significantly lower D mean, ADC0–1000 mean (Fig. 1) and ADC300–1000 mean values (Fig. 2a–c, Table 2) and a higher D* mean value (Fig. 2d, Table 2). However, no significant inter-group difference was observed in the f mean values (Fig. 2e, Table 2).

Table 2 Analysis of DW parameters calculated for study subjects
Full size table
Fig. 1
figure 1

Pure diffusion (D) and apparent diffusion coefficient0–1000 (ADC0–1000) maps and histograms of a representative subject with early-stage nasopharyngeal carcinoma (NPC) (a–d) and a representative subject with benign hyperplasia in the nasopharynx (e–h). The respective D mean and ADC0–1000 mean values were 0.59 × 10−3 mm2/s (b) and 0.82 × 10−3 mm2/s (d) for the subject with early-stage NPC and 0.76 × 10−3 mm2/s (f) and 0.93 × 10−3 mm2/s (h) for the subject with benign hyperplasia

Full size image
Fig. 2
figure 2

Side-by-side box plots of the mean true diffusion coefficient (D), apparent diffusion coefficient0–1000 (ADC0–1000), ADC300–1000, pseudo-diffusion (perfusion) coefficient (D*) and perfusion volume fraction (f) values in subjects with early-stage nasopharyngeal carcinoma (NPC) and benign hyperplasia. Patients with early-stage NPC had a significant lower D mean (a), ADC0–1000 mean (b) and ADC300–1000 (c), and a significantly higher D* mean (d), compared to patients with benign hyperplasia. The difference in the f mean (e) between the two groups was not significant. The optimal thresholds (grey dotted lines) of D mean, ADC0–1000 mean, ADC300–1000 and D* mean for distinguishing early-stage NPC from benign hyperplasia were 0.75 × 10−3 mm2/s, 0.74 × 10−3 mm2/s, 0.86 × 10−3 mm2/s and 27.4 × 10−3 mm2/s, respectively

Full size image

Table 3 and Fig. 3 present the diagnostic performance of the optimal thresholds for the above-listed significant DW parameters, as well as the AUCs of these parameters for distinguishing early-stage NPC from benign hyperplasia. Notably, the D mean was best able to distinguish these two entities in the nasopharynx, although when compared to ADC0–1000 mean and D* mean, there were no significant differences between the AUCs. Analysis of the inter-observer agreement yielded the highest ICC values for the D mean and ADC0–1000 mean (0.963 (95% CI, 0.940–0.978)), followed by the D* mean (0.953 (95% CI, 0.922–0.971)), and f mean (0.920 (95% CI, 0.868–0.951)) (all p < 0.001). The ADC300–1000 mean had a similar threshold and AUC (Table 3 and Fig. 3) (p = 0.823) to the D mean for distinguishing early-stage NPC from benign hyperplasia with a high ICC of 0.968 (95% CI, 0.948–0.981).

Table 3 Diagnostic performance of DW parameters in distinguishing early-stage NPC from benign hyperplasia
Full size table
Fig. 3
figure 3

Analysis of areas under the curves (AUCs) of the mean values of the true diffusion coefficient (D), pseudo-diffusion coefficient (D*), apparent diffusion coefficient0–1000 (ADC0–1000) and ADC300–1000 to determine the ability to distinguish early-stage nasopharyngeal carcinoma from benign hyperplasia. The D mean and ADC300–1000 had the highest AUCs of 0.985 and 0.988, respectively (blue curve and black curve, respectively), followed by the ADC0–1000 mean (0.948; green curve) and D* mean (0.926; red curve). This suggests that the D mean and ADC300–1000 were best able to distinguish early-stage nasopharyngeal carcinoma from benign hyperplasia

Full size image

Discussion

MRI has been established as the imaging modality of choice for the staging of biopsy-proven NPC [8, 9]. However, the role of this imaging modality is expanding to include the detection of endoscopically occult tumours [3, 10]. As a result of the more widespread use of plasma markers to screen for early-stage NPC, there is increasing pressure on MRI to distinguish benign hyperplasia from early-stage cancers which do not have other features such as deep spread or nodal metastases to suggest malignancy on MRI. Therefore, an MRI sequence that could improve the diagnostic performance of anatomical imaging and even replace contrast-enhanced images would be advantageous in the screening setting.

In this study, the mean values of the IVIM-derived parameters D and D*, but not f, could successfully distinguish early-stage NPC from benign hyperplasia. D mean was significantly lower and D* significantly higher in NPC than in benign hyperplasia suggesting NPC has hyper-cellularity, smaller extra-cellular space and more vascularity than benign hyperplasia. Of these two parameters, the D mean yielded the best diagnostic performance and reproducibility, but ADC0–1000 from a conventional DWI was also useful. Similar to D, subjects with early-stage NPC had a significantly lower ADC0–1000 than those with benign hyperplasia, suggesting that pure diffusion, compared with microcirculation, more strongly influences the ADC0–1000 value. Although the accuracy of ADC0–1000 mean was slightly lower than that of D, conventional DWI is still an acceptable alternative to IVIM DWI for patients, especially for those patients unable to tolerate the longer time scan. Scan time could be further reduced by limiting the conventional DWI to only those b-values between 300 and 1000 s/mm2 to acquire the ADC300–1000 which we found provided similar results to the D mean from the IVIM model. Our DWI results suggest that biopsies targeted to sites with a low D mean or low ADC300–1000 mean should have a high sensitivity for NPC, although a small number will be false positive because of an overlap in the mean values with benign hyperplasia. In addition, a turbo-spin-echo sequence, rather than an echo-planar sequence, can be used during conventional DW image acquisition. This is advantageous because the turbo-spin-echo sequence reduces susceptibility artefacts in the nasopharynx [11, 12].

DWI aids for the MRI-based characterisation of tissues in the head and neck. Most studies have reported lower ADC values in malignant tumours relative to benign processes, which reflects the more restricted diffusion in the former lesions [5]. More recently, IVIM studies in the head and neck have reported lower D values in malignant tumours relative to benign pathologies [6, 13,14,15,16,17,18,19], including comparisons of NPC with benign hyperplasia [6], which are consistent with our findings. The previously reported values for D in NPC [6, 18,19,20,21,22,23,24] and benign hyperplasia [6, 25] were also similar to those in our study. This similarity is encouraging given that our prospective study of consecutive patients in a screening setting included those with small volumes of NPC and benign hyperplasia. The other two IVIM parameters, D* and f, are likely less robust, which is supported by the greater range of values reported in studies of the head and neck [26], including NPC [6, 18,19,20,21,22,23,24].

This study had several limitations. Firstly, we were unable to assess whether there are differences in IVIM parameters between the two patterns of contrast enhancement in benign hyperplasia, because at the time of this IVIM study we only obtained plain scans in our patients who were undergoing follow-up MRI. Secondly, although the D mean yielded a superior diagnostic performance relative to the ADC0–1000 mean and D* mean, these differences did not reach statistical significance, likely due to the small sample size. Thirdly, the optimal thresholds obtained in this preliminary study will require validation in an independent data set before implementation in clinical practice.

Conclusion

In conclusion, the mean IVIM values of D and D* but not f, together with the mean ADC0–1000 and ADC300–1000 values obtained from conventional DWI, were identified as potential DW parameters that could distinguish early-stage NPC from benign hyperplasia in the nasopharynx. NPC was associated with a lower D, ADC0–1000, ADC300–1000 and a higher D* compared with benign hyperplasia, with the D and ADC300–1000 being the most promising parameters for use in the screening setting.