Whole-body magnetic resonance imaging in paediatric Hodgkin lymphoma — evaluation of quantitative magnetic resonance metrics for nodal staging

¹⁸Flourine-2-fluoro-2-deoxyglucose (FDG) positron emission tomography (PET) fused with CT scan (FDG PET/CT), performed on a hybrid system, has become the gold standard modality for evaluating lymphoma including Hodgkin lymphoma, with proven improvement at staging and follow-up monitoring compared with either CT or FDG PET alone [1, 2]. However, FDG PET/CT imparts a considerable dose of ionising radiation, which is of particular concern within the paediatric population [3]. Even with low-dose protocols, paediatric patients undergoing a single FDG PET/CT examination are exposed to ionising radiation on the order of 10–20 mSv [4], equivalent to 700–750 chest radiographs. Compared with an adult older than 30 years, exposure to ionising radiation in early childhood causes roughly a tripling in lifetime cancer risk by comparison [5]. Children are inherently more radiosensitive than adults and exposure to ionising radiation is more likely to increase the risk of secondary malignancies (radiation-induced malignancies) in children because of their smaller body size, repeated studies to assess treatment response or relapse, and longer survival following treatment [6, 7]. In paediatric and young adults with Hodgkin lymphoma, more than 90% of patients survive 10 years [8], and hence radiation-induced mortality and risk of developing secondary cancers are of more concern in children and adolescents with Hodgkin lymphoma. Whole-body MRI provides a radiation-free alternative for the staging of paediatric lymphoma [9, 10]. Yet, similar to CT, nodal classification remains dependent predominantly on size measurement, with image contrast used to help localise nodes [11]. Diffusion-weighted imaging (DWI)-derived apparent diffusion coefficient (ADC) has been reported as a potential imaging biomarker for evaluating treatment response in lymphoma [12, 13]. Several studies have suggested that ADC values for malignant lymph nodes are lower compared with benign lymph nodes [14, 15]. However, little has been reported on combining the quantitative parameters provided by sequences such as DWI with standard size measurement to determine whether this provides improved diagnostic accuracy for nodal disease classification [16]. The aim of this study was to determine the potential added diagnostic value of quantitative parameters derived from whole-body MRI for classifying lymph nodes in children and adolescents with Hodgkin lymphoma compared to the FDG PET/CT reference standard.

Our institutional review board approved the study and waived the requirement for individual patient consent for use of anonymous clinical and research patient datasets (R&D No: 12/0195 date:16/06/2012).

Thirty-seven children and young adults (male/female: 16/21; mean age 16.1 years; range 12.8–18 years) all had required baseline and follow-up whole-body MRI sequences and were therefore eligible for inclusion in the study. Staging whole-body MRI was performed within a median of 3 days (range 0–19 days) of FDG PET/CT, without any complications and before treatment in all patients. Across the cohort, 3 children had Stage I, 11 had Stage II, 2 had Stage IIE, 8 had Stage III and 13 had stage IV disease. One hundred fifty-two nodal sites were positive on reference standard FDG PET/CT. Eighteen and 23 nodes were 5- to 9-mm short-axis diameter and positive according to reference standard FDG PET/CT for Reader 1 and Reader 2, respectively.

Based on the Kolmogorov–Smirnov test, both ADC and normalised T2-signal-intensity metrics were not normally distributed. Median ADC and normalised T2 signal intensity for FDG PET/CT-positive and FDG PET/CT-negative lymph nodes are shown in Fig. 1 and Table 2 for both radiologists. For lymph nodes with 5- to 9-mm short-axis diameter, median ADC was significantly lower (P<0.0001) for FDG PET/CT-positive compared with FDG PET/CT-negative lymph nodes, whilst there was no significant difference in median normalised T2 signal intensity (Fig. 2, Table 2). Median ADC was significantly higher and median normalised T2 signal intensity significantly lower in nodes with 5- to 9-mm short-axis diameter compared with nodes with ≥10-mm short-axis-diameter (Fig. 3 and Table 2). Bland-Altman plots for interobserver and inter-sequence size measurement agreements (DWI b=500 s/mm² and STIR half-Fourier-acquisition single-shot turbo spin echo) are shown in Fig. 4. Bland-Altman plots of interobserver agreements for ADC and normalised T2 signal intensity measurements are represented in Fig. 5.

Whole-body MRI is being increasingly advocated as an alternative or adjunct to FDG PET/CT for staging and response monitoring in paediatric lymphoma. Previously published works have shown that morphological MRI and DWI can be implemented in whole-body MRI protocols for assessing the extent of disease [9, 13]. MRI signal itself can be made sensitive to the presence of pathology, and as such, MRI offers the opportunity to use signal characteristics as well as morphological features to classify disease.

Size measurement is conventionally used to determine nodal positivity. We investigated whether the performance of whole-body MRI for nodal classification, based on size measurement, was different when measurements were made on whole-body DWI rather than anatomical whole-body STIR half-Fourier-acquisition single-shot turbo-spin-echo imaging. We found that measurements made on either sequence yielded a comparable diagnostic accuracy for detecting involved lymph nodes and that size measurement provides the best classifier for nodal disease status.

In our study, diseased lymph nodes had significantly lower ADC values than benign lymph nodes (as highlighted by the FDG PET/CT reference standard); this finding supported the results of prior investigators [21, 22]. For instance, Perrone et al. [23] demonstrated that the mean ADC value for malignant cervical lymph nodes was significantly lower than that of benign lymphadenopathy (0.85×10⁻³ mm² s⁻¹ and 1.44×10⁻³ mm² s⁻¹, respectively, P<0.01). In another study, the authors showed that compared to benign lymphadenopathy and metastatic lymphadenopathy from head and neck cancers, lymphomatous lymph nodes had a significantly lower mean ADC value [21]. Whilst our results demonstrate good classification by ADC value (ROC/AUC of 0.67–0.74), we note that the performance of the ADC was not greater than simply measuring nodal size (ROC/AUC 0.80–0.81).

We also noticed that for nodes deemed negative for disease based on size criteria (measuring 5–9 mm) there was significant difference in ADC between disease-positive and disease-negative nodes (P<0.0001). It is not surprising, therefore, that after applying ADC cut-off values [20] to 5- to 9-mm lymph nodes, there was an overall increase in whole-body DWI’s sensitivity for both readers. However, the increase in whole-body DWI’s sensitivity came at the expense of noticeable decrease in specificity. We conclude that early nodal involvement (where size remains <1 cm) cannot be reliably classified using DWI and hypothesise that this likely reflects insufficient increase in cellularity to significantly alter water diffusion [24].

Our result for interobserver reproducibility of ADC measurements for lymph nodes in children and adolescents with Hodgkin lymphoma emulates previous findings for nodal ADC reproducibility in healthy volunteers [25, 26]. Moreau et al. [26] found that the inter-reader reproducibility for their ADC measurements showed an absolute bias of 0.045×10⁻³ mm² s⁻¹ (level of agreement −0.146; 0.056). Although both studies used a different protocol compared to our study, they highlighted that ADC measurement in healthy volunteers might not always be adequately reproducible and that a reliable use of ADC values requires further technical advances and systematic quality control [25, 26].

Short tau inversion recovery half-Fourier-acquisition single-shot turbo-spin-echo images generate contrast to highlight water whilst nulling signal from fat [27]. In theory, replacement of the normal fatty hilum by cellular infiltrate as occurs in pathology should increase the STIR half-Fourier-acquisition single-shot turbo-spin-echo signal from diseased lymph nodes [28]. This was reflected in our results, with FDG PET/CT-positive lymph nodes demonstrating approximately 20% increased signal on STIR half-Fourier-acquisition single-shot turbo-spin-echo images compared to FDG PET/CT-negative nodes. As a univariate classification parameter, normalised T2 signal-intensity classification performance (ROC/AUC 0.66–0.72) was not as high as ADC or nodal size measurement parameters.

Ohno et al. [29] evaluated 135 metastatic and 135 non-metastatic mediastinal and hilar lymph nodes in their cohort of 93 patients with N1-N3 non-small-cell lung cancer using 1.5-T STIR turbo-spin-echo MRI, diffusion-weighted imaging and FDG PET/CT, with the reference standard being histopathological samples. In line with our results, they showed that lymph node signal intensity as normalised to muscle signal intensity (lymph-node-to-muscle ratio) on STIR turbo-spin-echo sequence was significantly higher in metastatic lymph nodes compared to non-metastatic lymph nodes (mean lymph-node-to-muscle ratio 1.5 ± 0.3 [range 0.6–2.2] and 1.0 ± 0.3 [range 0.4–1.8] for metastatic and non-metastatic lymph nodes, respectively, P<0.0001). However contrary to our results, the performance of lymph-node-to-muscle ratio for determining metastatic lymph nodes was higher than that of ADC values in their cohort. This might be caused by two major differences between the two studies: first, the underlying malignancies were different between the two cohorts; second, Ohno et al. [29] only investigated mediastinal and hilar lymph nodes whilst we included all the measurable nodal stations above and below the diaphragm.

In our study, we included the nodal stations comprising only measurable nodes with short-axis diameter ≥5 mm, for two reasons. First, quantitative analysis of small nodes is more likely to be affected by partial volume errors compared with larger nodes. Second, and more important, almost all <5-mm lymph nodes in paediatric patients with Hodgkin lymphoma are negative on FDG PET/CT (indeed, we did not find a single FDG PET/CT-positive node in our cohort) and do not present a diagnostic challenge. Application of quantitative metrics should be for nodes in which there might be a diagnostic dilemma. Studies that include large numbers of nodes <5 mm within their population are likely to demonstrate artificially high levels of specificity for the technique being assessed.

Our results have direct clinical implications. They suggest that ADC values should not be used as a discriminator of nodal disease status in children with lymphoma and confirm the dominance of nodal size as the best classifier of disease status. Furthermore, if diffusion weighted imaging is used for staging paediatric lymphoma, it is sufficient to make measurements of nodal size directly from diffusion-weighted images (b=500 s/mm²) rather than from anatomical imaging such as STIR half-Fourier-acquisition single-shot turbo-spin-echo images. This might help to reduce scan times by providing an opportunity to replace STIR half-Fourier-acquisition single-shot turbo spin echo with a faster anatomical acquisition such as modified DIXON imaging [30, 31] or obviating the requirement for DWI based on no additional classification value. Limiting overall scan time is particularly important for paediatric patients and has direct heath care cost implications.

There were limitations to our study. First, we did not have histopathological confirmation for positivity/negativity for nodal involvement by lymphoma. We used FDG PET/CT as the reference standard because this is the current gold standard modality of imaging with excellent performance [32]. We have assumed that discordances between whole-body MRI and the FDG PET/CT reference are related to limitations in sensitivity/specificity of the whole-body MRI technique, but we acknowledge that a small number of lymph nodes might also be misclassified by FDG PET/CT. Although ethically and technically challenging, a prospective study with histological sampling to resolve discrepancies would further aid comparison between techniques. A second limitation of the study is the generalisability of whole-body MRI quantitative features across institutions. For example, absolute ADC values are known to be dependent on MRI protocol parameters and scanner platform [33, 34]. However our best-performing classifier remained nodal size, and classification was not significantly improved by ADC values. Third, we used the highest b value of 500 s/mm² for DWI disease assessment. We acknowledge that a higher b value of 800–1,000 s/mm² would have been in line with current recommendations [35]; however the ADC cut-off values were derived from previous work using a similar DWI protocol to the current study [19]. Finally, whilst we excluded nodes with short-axis diameter <5 mm from quantitative analysis to minimise the partial voluming effect, partial voluming effect on our quantitative measurements cannot be entirely excluded.

We have demonstrated that both whole-body DWI and whole-body STIR half-Fourier-acquisition single-shot turbo-spin-echo MRI size measurements had similarly very good performance for classifying nodal disease involvement in children with known lymphoma. Combined short-axis diameter and ADC thresholds marginally improve sensitivity and drop specificity compared with size classification alone. We continue to advocate the use of anatomical criteria for nodal disease classification and raise caution over the interpretation of ADC value as a classifier of disease, particularly in normal-size (5- to 9-mm short-axis diameter) lymph nodes.