Can whole-body MRI replace CT in management of metastatic testicular cancer? A prospective, non-inferiority study

Concerns of imaging-related radiation exposure in young patients with high survival rates have increased the use of magnetic resonance imaging (MRI) in testicular cancer (TC) stage I. However, computed tomography (CT) is still preferred for metastatic TC. The purpose of this study was to compare whole-body MRI incl. diffusion-weighted whole-body imaging with background body signal suppression (DWIBS) with contrast-enhanced, thoracoabdominal CT in metastatic TC. A prospective, non-inferiority study of 84 consecutive patients (median age 33 years) with newly diagnosed metastatic TC (February 2018–January 2021). Patients had both MRI and CT before and after treatment. Anonymised images were reviewed by experienced radiologists. Lesion malignancy was evaluated on a Likert scale (1 benign–4 malignant). Sensitivity, specificity, positive predictive value, negative predictive value and accuracy were calculated on patient and lesion level. The primary outcome was demonstrating non-inferiority regarding sensitivity of MRI compared to CT. The non-inferiority margin was set at 5%. ROC curves and interobserver agreement were calculated. On patient level, MRI had 98% sensitivity and 75% specificity compared to CT. On lesion level within each modality, MRI had 99% sensitivity and 78% specificity, whereas CT had 98% sensitivity and 88% specificity. MRI sensitivity was non-inferior to CT (difference 0.57% (95% CI − 1.4–2.5%)). The interobserver agreement was substantial between CT and MRI. MRI with DWIBS was non-inferior to contrast-enhanced CT in detecting metastatic TC disease. www.clinicaltrials.gov NCT03436901, finished July 1st 2021.


Introduction
Testicular cancer (TC) is a common cancer in young men of western societies (Laguna et al. 2020). TC metastases are predominantly located in the retroperitoneum and lungs (Laguna et al. 2020). The treatment of metastatic TC with chemotherapy or radiotherapy is very effective resulting in very high 5-year survival rates of 90-95% (Laguna et al. 2020).
The surveillance strategy for TC includes clinical examination, serum tumour markers and, not least, regular imaging. European guidelines recommend four contrast-enhanced abdominopelvic computed tomography (CT) scans in the first 5 years after chemotherapy (Laguna et al. 2020).
As most patients with TC are young and relatively healthy and their life-expectancy is high, they are at risk of developing a secondary radiation-induced cancer. Even though the radiation doses used in imaging are diminutive compared to the doses used for radiation therapy, the cumulative imagingrelated doses in TC are a concern e.g. (Tarin et al. 2009;Silva et al. 2012;Pandharipande et al. 2013). However, 1 3 estimating long-term effects of radiation-exposure is difficult and a few studies have found no excess risk of secondary cancer in radiation-therapy-naïve TC patients (van Walraven et al. 2011;Kier et al. 2016).
The use of magnetic resonance imaging (MRI) in follow-up of TC stage I is increasing (Stephenson et al. 2019;Laguna et al. 2020;DaTeCa 2021; SWENOTECA (Swedish and Norwegian Testicular Cancer Group) 2021) and recently, a standardized MRI protocol for follow-up of TC stage I was proposed (SWENOTECA (Swedish and Norwegian Testicular Cancer Group) 2021).
All concerns regarding radiation hygiene are just as relevant for patients with metastatic TC as for TC stage I. In fact, patients with metastatic disease are more likely to be subjected to more imaging-related radiation as part of treatment evaluation and because the health care contact is longer. Several national guidelines list MRI as an option after chemotherapy (The Danish Health and Medicines Authority 2015; SWENOTECA (Swedish and Norwegian Testicular Cancer Group) 2021).
The evidence is limited regarding MRI in detection of TC metastasis. However, in a few small studies, the diagnostic accuracy of MRI was similar to results from a systematic review of CT (MRI 75-100% sensitivity, 89-100% specificity. CT 66% sensitivity, 92% specificity) (Sohaib et al. 2009;Laukka et al. 2020;Pierorazio et al. 2020;Narine et al. 2021). In addition, a prospective, multi-centre RCT on follow-up of seminoma stage I with two CT arms and two MRI arms is in progress (Sohaib et al. 2009).
The use of MRI is limited by high costs and long scan time (Su et al. 2015), but with constant technology advances, image quality is improving and scan time is reducing. For this study, a dedicated non-contrast-enhanced whole-body (WB) MRI protocol with diffusion-weighted whole-body imaging with background body signal suppression (DWIBS) was used with approx. 30 min scan time.
The aim of this prospective, non-inferiority-study, was to study the ability of WB-MRI with DWIBS to replace standard contrast-enhanced, thoracoabdominal CT in metastatic TC. This study covered the pre-treatment staging scan and the first post-treatment evaluation scan.

Study design
A prospective, non-inferiority study of MRI with DWIBS (new test) compared to contrast-enhanced CT (reference test) was designed. A power analysis on lesion level was performed for a non-inferiority study with a paired design (patients being their own controls) with binary outcome (metastases or no metastases on a lesion basis) (Liu et al. 2002). Assuming 80% CT sensitivity and 74% MRI sensitivity, and choosing a 5% non-inferiority margin and 80% power, a sample size of 235 lesions was calculated. The 5% non-inferiority margin was justified by the reduced radiation dose and the fact that a delay in diagnosis of a small TC relapse is not likely to effect the overall survival.
The primary outcome was demonstrating non-inferiority regarding sensitivity of MRI compared to CT. The secondary outcome was calculating the sensitivity of MRI compared to CT.

Patients
Consecutive patients initiating treatment for metastatic TC were included at our institution from February 2018 to January 2021. Patients with primary metastatic TC, relapse during follow-up or pure serologic relapse were included. Patients with age < 18 years, claustrophobia or MRI contraindications were excluded.
The patients were approached with both oral and written study information at the oncological department after diagnosis of metastatic TC. Written consent was obtained after at least 24 h of consideration unless immediate treatment was imperative.
Treatment was in accordance with national guidelines and was not altered by study participation (DaTeCa 2021).

Imaging
All patients had CT and MRI of the thorax, abdomen and pelvis performed prior to treatment initiation and after treatment (three weeks after last chemotherapy or three months after completing radiotherapy in accordance with national guidelines (The Danish Health and Medicines Authority 2015; DaTeCa 2021). After study participation, patients returned to standard follow-up with CT.
Clinical CTs were used as much as possible. If a recent contrast-enhanced thoracoabdominal CT was available, no additional CT was performed. Some diagnostic contrastenhanced CTs were retrieved from pre-treatment PET-CTs.
A dedicated WB-MRI protocol was developed based on the existing clinical MRI protocol of the retroperitoneum and pelvis for follow-up of TC stage I  with extension to the upper abdomen and thorax.
A 1.5T MRI system was used (Ingenia, release 5.3 with DDAS spectrometer, Philips Medical Systems, Best, The Netherlands) with the built-in posterior coil, dS head-neck coil and flex coverage anterior coil.
CTs were evaluated in consensus by two oncological radiologists (18/30 years' experience). MRIs were evaluated independently by two oncological radiologists (14/23 years' experience) with subsequent consensus reading. Except for the diagnosis of metastatic TC, no clinical information was provided. To mimic clinical routine, the pre-treatment scan was evaluated alone first, whereafter the post-treatment scan was evaluated side-by-side with the pre-treatment scan.
The five largest lesions were recorded in eight anatomical regions: lungs, supraclavicular, mediastinum, retroperitoneum along the aorta, retroperitoneum along the iliac vessels, groin, other lymph node metastases and other metastases. For each lesion, anatomical position, pre-and post-treatment description, pre-and post-treatment diameter (shortest lymph node axis on CT/T2), and pre-and posttreatment Likert score of malignancy were recorded. Presence of other pathological findings was recorded.
All data were managed using REDCap (Vanderbilt University, Tennessee, USA).

Statistical analysis
A Likert scale was constructed (1 benign, 2 probably benign, 3 probably malignant, 4 malignant) (Norman 2010). Likert 1-2 was considered benign and Likert 3-4 was considered malignant. A category representing uncertainty was avoided. The post-treatment Likert score included evaluation of treatment response in each lesion using each patient as their own control.
The reference for malignancy was defined as Likert 3-4 on post-treatment CT. The presence of malignancy was evaluated on patient and lesion level.
Sensitivity, specificity, positive predictive value, negative predictive value and accuracy were calculated with 95% confidence intervals (CI). On a lesion level, data were adjusted for clustering in patients. Lesions too small for measurement were included if measured on the other modality. ROC curves were constructed using different cut-off Likert scores for malignancy.
For non-inferiority testing, MRI and CT results could not be intermingled. Instead, the pre-treatment scan was compared to the treatment-response evaluation on the post-treatment scan within each modality. Due to the highly efficient treatment, we expected any truly malignant lesion to shrink or necrotize during treatment. The difference in sensitivity was calculated with two-sided 95% CI (Liu et al. 2002). The non-inferiority margin was 5%.
Interobserver agreement was estimated using kappa statistics (Norman 2010) in describing each lesion as benign (Likert 1-2) or malignant (Likert 3-4) after review of both scans.

Results
Ninety patients were included, but one tumour was reclassified as rhabdomyosarcoma and one suspicion of relapse was dropped. Four patients dropped out due to treatment The majority of malignant lesions were located in the retroperitoneum and lungs (Table 4).
The difference between CT and MRI sensitivities was 0.57% (95% CI − 1.4 to 2.5%) and thus MRI was non-inferior to CT in detecting TC metastasis.
In total, 238 lesions were described on both CT and MRI. The median difference in pre-treatment size on CT and MRI was 0 mm [IQR -3 to 1 mm]. Directly compared to CT on lesion level, MRI had 96% sensitivity (95% CI 94-99%) and 87% specificity (95% CI 69-92%). Four lesions were false positive on MRI: a lung lesion and three lymph nodes ≤ 10 mm. Eight lesions were false negative on MRI: a lung lesion, six lymph nodes ≤ 10 mm and a 12 mm lymph node without restrictive diffusion. Details Table 2D  In total, 259 lesions were described by one modality only. Different selection in regions with > 5 lesions accounted for 72 lesions. In general, the remaining lesions were < 10 mm and located in the retroperitoneum or lungs.
In one patient, Th12 invasion of a retroperitoneal conglomerate was only visible on MRI. In another patient, a liver lesion disappeared between the pre-and post-treatment scans. Although described as benign before treatment by both modalities, neither CT nor MRI could exclude a liver metastasis. Both findings lead to relevant clinical risk stratification and treatment (image examples Fig. 2).
Eleven patients were operated for residual tumour. In ten patients, the residual tumour was visible on both CT and MRI. In the last patient, a four mm necrotic lesion was described by CT without a corresponding lesion described by MRI (teratoma removed 5 months after post-treatment scans).  Table 3). Every second line is a zoom-in of the image above. Pre-treatment computed tomography (CT) and pretreatment magnetic resonance imaging (MRI) [T2, b800 and apparent diffusion coefficient (ADC)]. a False negative in retroperitoneum: 8 mm lymph node (Likert 4) on CT. Several small, unmeasurable lymph nodes (Likert 2) on MRI. b False negative in retroperitoneum: 22 mm lymph node (Likert 4) on CT. 12 mm lymph node without diffusion restriction (Likert 2), almost indistinguishable from the intestines on MRI. c False positive in retroperitoneum: 8 mm lymph node (Likert 3) on MRI. No lesions described by CT, but a benign lymph node is visible ◂   AUC for ROC curves was 0.97 for pre-vs. post-treatment CT, 0.96 for pre-vs. post-treatment MRI and 0.95 for posttreatment MRI vs. post-treatment CT.
The interobserver agreement was substantial between CT and MRI (kappa 0.78).

Discussion
In this prospective, non-inferiority study, the lesion level sensitivity of MRI was non-inferior to CT. The number of lesions calculated in the power analysis was reached.
On a patient level, MRI had 98% sensitivity and 75% specificity compared to CT. Disagreement only existed in three patients and concerned lymph nodes of which only one was > 10 mm. On subsequent post-chemotherapy CT, all these lymph nodes diminished in size indicating malignancy (Table 3).
On lesion level using the post-treatment scan as reference within each modality, comparable sensitivities (MRI 99%, CT 98%) and slightly differing specificities (MRI 78%, CT 88%) were found. Disagreement only existed for 12 lesions in 11 patients (Table 3). MRI and CT performed equally well in detecting residual tumours.
Sensitivity is preferred over specificity from a clinical perspective. Final diagnosis is rarely based on imaging alone but needs confirmation by biopsy or serum tumour markers (DaTeCa 2021). As early diagnosis is vital to prognosis, missing a lesion is much more problematic than raising unwarranted suspicion.
Only few studies have investigated MRI in metastatic TC. Laukka et al. (2020) compared MRI with DWI to contrastenhanced CT in 50 TC patients (47 with retroperitoneal TC). Using clinical data as reference, the sensitivity was comparable (MRI 98%, CT 96%), but the specificity differed (MRI 100%, CT 75%). Sohaib et al. (2009) studied 52 patients (30 with retroperitoneal TC), and found MRI to have 78-96% sensitivity and 91-100% specificity compared to CT. Two other MRI studies included a few patients with metastatic TC but did not calculate diagnostic accuracy (Pfannenberg et al. 2004;Mosavi et al. 2015).
The high MRI sensitivity was in line with our results, however, our MRI specificity was slightly lower.
Due to motion, MRI is usually not well suited for lungs (Khalil et al. 2016). To improve detection of lung lesions, we applied a fast, respiratory-triggered DWIBS b200 sequence. The b value was high enough to suppress flow from pulmonary veins while obtaining a high SNR. A similar approach has previously been useful in detecting lung metastases in colon cancer (Sivesgaard et al. 2020).
Although only few lung lesions were present, we found MRI to have 96% sensitivity in detection of lung lesions. This is in line with two meta-analyses in which MRI with DWI had 80-83% sensitivity and 80-93% specificity in detecting pulmonary nodules (Chen et al. 2013;Li et al. 2014). However, further studies of pulmonary MRI in TC are necessary.
The lack of gold standard such as histopathology for each lesion was the main limitation. Instead, either CT was used as reference or the treatment-response evaluation on post-treatment scans within the same modality was used as reference. The assumption was that malignant lesions would shrink or necrotise during the highly effective treatment, whereas benign lesions would appear unchanged. Known limitations would be malignant lesions not responding to treatment (very rare) or reactive lymph nodes shrinking without being malignant (rarely described as malignant).
A major limitation was the lesions described by either, but not both modalities. As an example, most benign lesions (except for benign lymph nodes) do not have increased signal on DWIBS and are not as visible on MRI due to the lower resolution, whereas structures with restrictive diffusion on MRI are not prominent on CT. The different number of benign lesions described created a bias in specificity.
The five largest lesions in each region were recorded, but the readers often disagreed on which were largest. As a result, several of these lymph nodes were only described by one modality.
Knowledge of the diagnosis created a bias towards judging lesions as malignant and thus increasing sensitivity and decreasing specificity.

Conclusions
In this prospective, non-inferiority study, we found MRI with DWIBS to be non-inferior to contrast-enhanced CT in detecting metastatic TC disease.
It seems realistic that patients with metastatic TC may benefit from non-ionizing MRI as follow-up imaging in the future.

Conflicts of interest
The authors have no relevant financial or nonfinancial interests to disclose.
Ethics approval This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Central Denmark Region Ethics Committee (1-10-72-179-17) and the Danish Data Protection Agency (1-16-02-720-17).
Informed consent Informed consent was obtained from all individual participants included in the study. The authors affirm that patients provided informed consent for publication of the images in Figs. 1 and 2.