Differentiation of peripheral nerve sheath tumors in patients with neurofibromatosis type 1 using diffusion-weighted magnetic resonance imaging

Abstract

Background

We sought to determine the value of diffusion-weighted (DW) magnetic resonance imaging (MRI) for characterization of benign and malignant peripheral nerve sheath tumors (PNSTs) in patients with neurofibromatosis type 1 (NF1).

Methods

Twenty-six patients with NF1 and suspicion of malignant transformation of PNSTs were prospectively enrolled and underwent DW MRI at 3T. For a set of benign (n = 55) and malignant (n = 12) PNSTs, functional MRI parameters were derived from both biexponential intravoxel incoherent motion (diffusion coefficient D and perfusion fraction f) and monoexponential data analysis (apparent diffusion coefficients [ADCs]). A panel of morphological MRI features was evaluated using T1- and T2-weighted imaging. Mann–Whitney U-test, Fisher’s exact test, and receiver operating characteristic (ROC) analyses were applied to assess the diagnostic accuracy of quantitative and qualitative MRI. Cohen’s kappa was used to determine interrater reliability.

Results

Malignant PNSTs demonstrated significantly lower diffusivity (P < 0.0001) compared with benign PNSTs. The perfusion fraction f was significantly higher in malignant PNSTs (P < 0.001). In ROC analysis, functional MRI parameters showed high diagnostic accuracy for differentiation of PNSTs (eg, ADC_mean, 92% sensitivity with 98% specificity, AUC 0.98; D_mean, 92% sensitivity with 98% specificity, AUC 0.98). By contrast, morphological imaging features had only limited sensitivity (18–94%) and specificity (18–82%) for identification of malignancy. Interrater reliability was higher for monoexponential data analysis.

Conclusion

DW imaging shows better diagnostic performance than morphological features and allows accurate differentiation of benign and malignant peripheral nerve sheath tumors in NF1.

Key Points.

1. DWI allows sensitive and specific identification of malignancy in NF1.

2. Morphological features have limited accuracy for identifying malignant transformation.

Importance of the Study.

Malignant transformation of PNSTs is the leading cause of mortality in patients with NF1. The only curative approach is early detection followed by radical resection. Morphological features of tumors as determined by MRI have limited diagnostic accuracy. We aimed to assess the value of DW MRI for characterization of PNSTs in NF1 as malignant or benign. Our results show that DW imaging is better suited for identification of malignant transformation. Functional MRI should therefore be implemented as a complementary diagnostic tool when screening for malignant PNSTs in NF1 patients.

Neurofibromatosis type 1 (NF1) is an autosomal-dominantly inherited tumor predisposition syndrome with an incidence of 1 in 2500 newborns.¹ A key feature of NF1 is the development of peripheral nerve sheath tumors (PNSTs).² PNSTs undergo transformation into malignant peripheral nerve sheath tumors (MPNSTs) in 8–13% of NF1 patients.³ Malignant PNSTs are a leading cause of death in NF1 (up to 55%)⁴ and contribute to the increased mortality of these patients.⁵ The only curative approach to malignant PNSTs is early detection and resection.^2,6 Symptoms suggestive of malignant transformation such as tumor growth, neurological deficits, or pain are nonspecific, not observed in all cases, and often signs of advanced disease.⁷

Therefore, NF1 patients are repeatedly imaged with whole-body magnetic resonance imaging (MRI), allowing evaluation of tumor burden^8–10 and thus identification of patients at risk of developing malignant PNSTs.^11,12 Morphological features of PNSTs at MRI such as intratumoral heterogeneity, intratumoral cystic changes, and ill-defined tumor margins cannot reliably discriminate between benign and malignant PNSTs,^13,14 advocating a role for functional parameters for detection of malignancy, such as increased ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) consumption at positron emission tomography (PET).^15–17 However, PET is associated with radiation exposure, and false positive results in plexiform neurofibromas have been described in 5%.⁷

Functional MRI with diffusion-weighted imaging (DWI) has been suggested for differentiation between malignant and benign soft-tissue tumors to increase the diagnostic accuracy.¹⁸ A recent retrospective study demonstrated that DWI-derived parameters may hold promise to distinguish malignant from benign PNSTs.¹⁴ However, that study excluded many tumors in NF1 patients (eg, tumors <5 cm), leaving the applicability for early detection of malignant transformation unanswered. Furthermore, only a monoexponential apparent diffusion coefficient (ADC) model based on a limited number of b-values was applied. However, monoexponential ADC models do not consider the physiological process of microcapillary perfusion. By contrast, the biexponential intravoxel incoherent motion (IVIM) model enables further separation of tissue perfusion and diffusivity.^19,20 IVIM has shown promise for distinguishing malignant from benign pediatric abdominal tumors²¹ and in the differentiation of high- from low-grade hepatocellular carcinoma.²²

Therefore, we aimed to assess the value of DW MRI for characterization of benign and malignant PNSTs in patients with NF1, and to compare the diagnostic performance of biexponential IVIM with that of monoexponential data analysis and established morphological features of malignant transformation.

Materials and Methods

This prospective study is compliant with the Health Insurance Portability and Accountability Act and the Declaration of Helsinki and has been approved by the local institutional review board. Written informed consent was obtained from all patients.

Patient Cohort

Patients who were referred for MRI screening for malignant PNSTs between August 2014 and September 2017 were eligible for inclusion in this study. Inclusion criteria were fulfillment of the National Institutes of Health diagnostic criteria for NF1,²³ changes in texture or size of known tumors in screening whole-body MRI scans or onset of clinical symptoms, and minimum tumor size of ≥20 mm. Patients were consecutively and prospectively enrolled if none of the following exclusion criteria were present: unwillingness or disability to give informed consent.

Histopathological evaluation after tumor resection and/or radiological follow-up served as the reference standard. Resected tumors were classified according to the grading systems of the World Health Organization (WHO) and the Fédération Nationale des Centres de Lutte Contre le Cancer (FNCLCC).^24,25 Tumors were considered benign when no changes in size or appearance were present in follow-up examinations within ≥12 months.

Image Acquisition

Imaging was performed at 3T (Philips Ingenia). MRI sequences consisted of a T1-weighted gradient recalled echo localizer, a coronal T2-weighted turbo spin echo sequence, an axial T2-weighted turbo spin echo sequence with fat suppression, and a 3D T1-weighted multipoint Dixon (mDixon) turbo spin echo sequence with in- and opposed phase (Supplementary Table 1). DWI consisted of axial respiratory-triggered echo-planar sequences with 11 diffusion gradient b-values (0, 10, 20, 30, 50, 70, 100, 300, 400, 600, 800 s/mm²). Intravenous contrast material was not administered. Scan time was 35 minutes.

Image Analysis

Images were independently analyzed in a randomized blinded fashion by 2 radiologists with 7 (J.S.) and 4 years (L.W.) of experience in MR imaging. Image analyses were undertaken on a workstation with a picture archiving and communication system (Centricity Universal Viewer, v6.0, GE Healthcare).

Image Processing and Data Analysis Using DWI

Data were processed with a self-developed image-analysis framework (qMapIt)²⁶ extending the open source software ImageJ (National Institute of Mental Health). Quantitative parametric maps were calculated by nonlinear regression with pixelwise fitting of signal intensities over the spectrum of b-values to the corresponding model. A monoexponential function was applied for ADC determination. The IVIM model was fitted in 2 steps to a biexponential function. First, the tail of the curve with b-values larger than 150 s/mm² was fitted to a monoexponential function and the diffusion coefficient D was extracted. Next, D was kept constant when fitting for the determination of the perfusion fraction f and the pseudo perfusion D*. The value of D* was determined in an initial study performing the analysis over regions averaging pixel values improving the signal-to-noise ratio. D* was consequently fixed to 20000 µm²/s to further increase numerical stability.

ImageJ was used for analysis of parameter maps (Fig. 1A). For all parameters, regions of interest (ROIs) were placed in diffusion images (b-value 50 s/mm²). ROIs were placed along tumor margins in 3 adjacent slices with largest axial tumor diameter (Supplementary Fig. 1). After manual placement of ROIs along tumor margins, the software automatically reduced the radius of ROIs by 3 mm to eliminate the risk of exceeding tumor margins. ImageJ then automatically extracted values from the previously created ADC and IVIM parameter maps. Averaged values from the 3 adjacent ROIs were calculated to determine ADC_mean, ADC_min, D_mean, and D_min. Perfusion fraction f was determined likewise, and termed f_center. Additionally, a single ROI was placed along tumor margins at lowest signal intensity, to determine ADC_dark and D_dark. The rationale for this approach was that malignant PNSTs arise within benign PNSTs and are associated with foci of high cellularity.² Therefore, minimum ADC values or areas of low signal intensity might improve diagnostic yield.^14,27 Finally, ring-like ROIs—containing the area between tumor margin ROIs and ROIs with automatically reduced radius—were used to obtain f_margin and D_margin, respectively (Supplementary Fig. 1).

Fig. 1.

(A) MR images of a 49-year-old male patient with a malignant (M)PNST, left column) and of a 39-year-old female patient with a benign (B)PNST, right column). T2-weighted images (upper row) and parametric maps (ADC, D, and f, respectively) calculated from DWI data. The thoracic MPNST is characterized by lower signal intensity in T2w imaging as well as lower values in ADC and D parameter maps, and higher perfusion fraction f compared with the BPNST. (B–E) Box-and-whisker plots of ADC (B) and IVIM-derived parameters (C–E) for characterization of MPNSTs and BPNSTs. Significant differences between MPNSTs and BPNSTs were observed for all approaches. (F) T2-weighted MR images of different patients with malignant (upper row) and benign (lower row) PNSTs demonstrating the evaluated morphological features.

Evaluation of Morphological Features

Morphological evaluation was performed using T1- and T2-weighted sequences. After measurement of tumor diameter, each PNST was evaluated for the presence or absence of 7 distinct morphological features (Fig. 1F): shape (round vs irregular), margin definition (well vs ill-defined), intratumoral lobulation, intratumoral heterogeneity, cystic changes (ie, high homogeneous T2w- and low homogeneous T1w signal), target sign (ie, hyperintense margin with hypointense center on T2-weighted images), and peritumoral edema.^13,28,29

Statistical Analyses

Continuous variables are expressed as mean ± standard deviation (SD). Categorical variables are presented with absolute and relative frequencies. For comparison of continuous data, P-values were calculated from Mann–Whitney U-tests. For categorical variables, P-values were computed from contingency tables using Fisher’s exact test. Receiver operating characteristics (ROC) analyses were performed to determine the area under the ROC curve (AUC) and to determine optimal cutoff values for differentiation of benign from malignant PNSTs. Spearman correlation analysis was used to correlate findings between IVIM-derived values D and f. Cohen’s kappa was used to evaluate interrater reliability. P-values <0.05 were considered statistically significant. Statistical analysis was performed using GraphPad Prism 5.0 for Windows.

Results

Study Population and Reference Standard

Twenty-six patients (13 men and 13 women) with a mean age of 34.2 years (17–54 y) were imaged, and a total of 67 PNSTs were analyzed.

Histopathological evaluation following resection was available in 42 of 67 PNSTs (62.7%). All (12 of 12) malignant PNSTs were verified by histopathology (6 WHO grade III/FNCLCC grade II and 6 WHO grade IV/FNCLCC grade III tumors). Thirty of 55 benign PNSTs (54.5%) were verified by histopathology (29 plexiform neurofibromas WHO grade I and one neurofibroma), whereas the remaining 25 benign PNSTs (45.5%) were confirmed by follow-up.

PNSTs were more frequently observed in female patients (40 tumors; 60%) than in male patients (27 tumors; 40%). Among these, malignant PNSTs were also more frequently found in female patients (8 tumors; 67%) than in male patients (4 tumors; 33%). Considering the topographic distribution of tumors, we observed 9 cervical tumors (13%; 1 MPNST/8 benign PNSTs), 23 thoracic tumors (34%; 8 MPNSTs/15 benign PNSTs), 31 abdominal tumors (46%; 2 MPNSTs/29 benign PNSTs), and 4 tumors located within the lower extremities (6%; 1 MPNST/3 benign PNSTs).

Diffusion-Weighted Imaging of Peripheral Nerve Sheath Tumors

ADC and D values were significantly lower in malignant PNSTs (ADC_mean, 1.23 ± 0.25 × 10⁻⁶ mm²/s; D_mean, 1.03 ± 0.28 × 10⁻⁶ mm²/s) (all P < 0.0001) compared with benign PNSTs (ADC_mean, 2.09 ± 0.30 × 10⁻⁶ mm²/s; D_mean, 1.88 ± 0.27 × 10⁻⁶ mm²/s) (Fig. 1, Table 1). The average ADC, D, and f values of malignant and benign PNSTs are shown in Table 1.

Table 1.

ADC and IVIM-derived parameters of malignant (M) and benign (B)PNSTs

Parameter	MPNSTs	BPNSTs	P
ADC mean	1.23 ± 0.25	2.09 ± 0.30	<0.0001
ADC dark	1.21 ± 0.26	2.03 ± 0.30	<0.0001
ADC min	0.60 ± 0.33	1.60 ± 0.38	<0.0001
D mean	1.03 ± 0.28	1.88 ± 0.27	<0.0001
D dark	0.99 ± 0.30	1.82 ± 0.30	<0.0001
D min	0.32 ± 0.33	1.40 ± 0.31	<0.0001
f center	21 ± 11	11 ± 5	0.0003
f margin	24 ± 9	14 ± 6	<0.0001
D margin	0.92 ± 0.23	1.77 ± 0.25	<0.0001

ADC and D values are shown as (value) × 10^–6 mm²/s ± standard deviation; f values are shown as %.

Both malignant and benign PNSTs demonstrated a significantly higher perfusion fraction in the tumor periphery than in the tumor center (MPNSTs, f_center 21 ± 11% vs f_margin 24 ± 9%; P = 0.0038; benign PNSTs, f_center 11 ± 5% vs f_margin 14 ± 6%; P < 0.0001). Perfusion fraction f was significantly higher in malignant PNSTs, which was observed both in the center (21 ± 11% vs 11 ± 5%; P = 0.0003) and in the periphery of tumors (24 ± 9% vs 14 ± 6%; P < 0.0001) (Table 1). There were no significant ADC, D, or f differences between WHO grade IV and grade III malignant PNSTs (Supplementary Table 2).

There was an inverse correlation between diffusion coefficient D and perfusion fraction f (D_min vs f_center; r = −0.488, P < 0.0001).

Interrater reliability

Interrater reliability was “very good” for all ADC-derived parameters (kappa ADC_mean, 0.898 − ADC_dark, 0.906), whereas interrater reliability was “good to very good” for IVIM-derived diffusion coefficients (kappa D_mean, 0.646 − D_dark, 0.831), and “moderate” for perfusion fraction (kappa f_center, 0.467 − f_margin, 0.468) (Supplementary Table 3).

Morphological Features of Peripheral Nerve Sheath Tumors

Malignant PNSTs were significantly associated with irregular shape (P = 0.0133), ill-defined margins (P < 0.0001), intratumoral lobulation (P = 0.0293), intratumoral heterogeneity (P < 0.0001), intratumoral cystic changes (P = 0.0006), and peritumoral edema (P < 0.0001) (Fig. 1 and Table 2). The target sign was absent in all malignant PNSTs, but also in a substantial number of benign PNSTs (P = 0.055). Morphological features were not significantly different between WHO grade IV and grade III malignant PNSTs (P > 0.05 in all cases) (Supplementary Table 4). The most common morphological features in both WHO grade IV and grade III tumors were peritumoral edema and intratumoral heterogeneity (grade IV: 6/6 [100%]; grade III: 5/6 [83%]).

Table 2.

Morphological features of MPNSTs (n = 12) versus BPNSTs (n = 55) on MRI

Parameter	MPNSTs (%)	BPNSTs (%)	P
n	12 (18)	55 (82)	‒
Largest diameter, mm			0.1407
Mean ± SD	72 ± 75	35 ± 15
Range	20–290	20–103
Shape			0.0133
Round shape	6 (50)	47 (85)
Irregular shape	6 (50)	8 (15)
Margin definition			<0.0001
Well-defined	6 (50)	55 (100)
Ill-defined	6 (50)	0 (0)
Intratumoral lobulation			0.0293
Present	4 (33)	4 (7)
Absent	8 (67)	51 (93)
Intratumoral heterogeneity			<0.0001
Heterogeneous	11 (92)	5 (9)
Homogeneous	1 (8)	50 (91)
Intratumoral cystic changes			0.0006
Present	4 (33)	1 (2)
Absent	8 (67)	54 (98)
Target sign			0.055
Present	0 (0)	23 (42)
Absent	12 (100)	32 (58)
Peritumoral edema			<0.0001
Present	11 (92)	1 (2)
Absent	1 (8)	54 (98)

Diagnostic Performance of Different Imaging Approaches for Differentiation of Malignant from Benign PNSTs

Morphological features demonstrated a sensitivity of 18‒94%, and a specificity of 18‒82% to differentiate between malignant and benign PNSTs (Table 3). The highest sensitivity (94%) was achieved using intratumoral cystic changes as a criterion, albeit associated with low specificity (18%). By contrast, ADC and IVIM-derived parameters demonstrated both high sensitivity and specificity for identification of malignant PNSTs (Fig. 2, Table 4). At ROC analysis, the largest AUC was obtained for D_min (sensitivity 100% and specificity 96% at a cutoff value of 0.96 × 10⁻⁶ mm²/s, AUC, 0.99; P < 0.0001), followed by ADC_mean and ADC_dark (both AUC, 0.98, P < 0.0001). In general, large AUCs were obtained for both, all ADC and all D values (Fig. 2, Table 4).

Table 3.

Diagnostic performance of morphological features discriminating malignant from benign PNSTs

Morphological Feature	AUC	95% CI	Sensitivity	Specificity
Shape	0.52	0.42–0.61	21% (12–33)	82% (71–90)
Margin definition	0.55	0.45–0.64	91% (82–97)	18% (10–29)
Intratumoral lobulation	0.53	0.43–0.63	88% (78–95)	18% (10–29)
Intratumoral heterogeneity	0.53	0.43–0.63	24% (14–36)	71% (71–90)
Intratumoral cystic changes	0.56	0.46–0.66	94% (85–98)	18% (10–29)
Target sign	0.58	0.49–0.68	34% (23–47)	82% (71–90)
Peritumoral edema	0.50	0.40–0.60	18% (10–29)	82% (71–90)

Values in parentheses are 95% CIs.

Fig. 2.

ROC curves of ADC and IVIM-derived parameters for differentiation of benign from malignant PNSTs.

Table 4.

Diagnostic performance of ADC and IVIM-derived parameters for discriminating malignant from benign PNSTs

Parameters	AUC	95% CI	Optimal Cutoff Value	Sensitivity	Specificity
ADCmean	0.98	0.95–1.00	1.60	92%	98%
ADCdark	0.98	0.93–1.00	1.54	92%	98%
ADCmin	0.97	0.93–1.00	0.95	92%	95%
Dmean	0.98	0.95–1.00	1.41	92%	98%
Ddark	0.97	0.92–1.00	1.30	83%	100%
Dmin	0.99	0.99–1.00	0.96	100%	96%
fcenter	0.84	0.73–0.95	10	100%	55%
fmargin	0.81	0.67–0.95	16	83%	75%
Dmargin	0.99	0.97–1.00	1.15	92%	100%

AUC with 95% CI and sensitivity/specificity for optimum cutoff values. Cutoff values are shown as (value × 10^–6 mm²/s) for ADC and D, or % for f, respectively.

Discussion

In this study, we demonstrated that utilization of DWI-derived parameters for differentiation of malignant from benign PNSTs resulted in higher levels of sensitivity and specificity compared with morphological features.

Both the ADC and the IVIM model improved diagnostic accuracy compared with morphological MR features, and both models demonstrated that malignant PNSTs display significantly higher diffusion restriction than benign PNSTs. This is in line with the results of histopathologic evaluation of PNSTs showing a higher cellularity in cases of malignancy.³⁰ In our study, DWI-derived values of malignant PNSTs (ADC_mean, 1.23 ± 0.25 × 10⁻⁶ mm²/s) matched previously published data on malignant soft tissue tumors.³⁰ The highest levels of sensitivity and specificity were achieved by determining minimal values of the diffusion coefficient D in tumors using the IVIM model, resulting in a sensitivity of 100% and a specificity of 96%. This finding is in line with results of Demehri et al, who suggested that the minimum ADC value is a better predictor of malignancy than the mean ADC value.¹⁴

However, we observed excellent AUCs for all ADC measures, with only minor differences (eg, between ADC_mean and ADC_min). Diagnostic accuracy using the simplified ADC_dark and D_dark approach was similar to the more complex method using 3 ROIs only in case of ADC_dark, whereas D_dark showed lower sensitivity at optimal cutoff values. For ADC_dark, this simplified approach of determining DWI-derived values in areas of highest cellularity by placement of a single ROI might be better suited for use in the clinical setting. Simple visual inspection of DWI sequences did not allow for reliable discrimination between malignant and benign PNSTs in all cases (Supplementary Fig. 2), advocating quantification.

Regarding the IVIM model, best results were obtained for D_min to differentiate between malignant and benign PNSTs. D_mean and D_dark measurements of diffusion coefficient D also resulted in high levels of diagnostic accuracy, confirming previous results for the IVIM model in other entities.²¹ The diagnostic performance of ADC-derived values was similar to that of IVIM-derived values with high levels of sensitivity (ADC_mean 92%; D_mean 92%) and specificity (ADC_mean 98%; D_mean 98%).

However, interrater reliability analysis demonstrated that the monoexponential data analysis was more robust. The highest agreement was observed for ADC-derived parameters (kappa, 0.898–0.906), whereas Cohen’s kappa was substantially lower for IVIM-derived parameters (kappa D, 0.646–0.831; and f, 0.467–0.468). The lower interrater reliability of IVIM-derived parameters might be explained by the more complex calculation and fitting procedure in which minor differences in ROI placement between readers may sum to greater variability in final values, indicating that monoexponential data analysis is a more robust method and should be preferentially applied.

Perfusion fraction f was significantly increased in malignant PNSTs and inversely correlated with diffusion coefficient D. The inverse correlation between D and f can be explained by the fact that malignant transformation may cause an increase in both vascularity and cellularity, thus increasing perfusion fraction f and decreasing diffusion coefficient D within the IVIM model. Perfusion fraction f was of limited usefulness regarding tumor discrimination due to moderate levels of sensitivity (83–100%) and specificity (55–75%). Nevertheless, perfusion fraction might add relevant information regarding tumor aggressiveness or treatment response.^31,32

Morphological imaging features demonstrated limited diagnostic accuracy for differentiation of malignant and benign PNSTs with sensitivities of 18–94% and specificities of 18–82%, similar to previous observations.¹³ Malignant PNSTs were significantly associated with distinct morphological features, such as irregular shape, ill-defined margins, intratumoral heterogeneity, and peritumoral edema. Peritumoral edema was the most common feature of malignant PNSTs (92% of cases), which is in line with previous studies.^14,28 Intriguingly, the maximum tumor diameter was not a useful indicator of malignancy in this study. By contrast, other studies have suggested tumor size as suggestive of malignancy.^15,28 This discrepancy may be explained by exclusion of tumors <2 cm in this study, since other morphological features cannot be reliably evaluated in these. Regarding other established imaging techniques for screening of NF1 patients such as PET/CT, functional DWI-derived parameters may improve the diagnostic accuracy to a considerable extent. DWI may be particularly useful as a complementary method in cases of borderline glucose metabolism of tumors in PET imaging. Additionally, DWI can support planning of image-guided biopsy, since targeting areas of minimal ADC/D values can improve diagnostic yield.²⁷ Finally, functional DWI might be applied to monitor therapeutic effects of mitogen-activated protein kinase inhibitors (MEK inhibitors).^32,33 Previous studies demonstrated that DWI, and the IVIM model, can contribute to monitor therapeutic effects in pharmacological therapy.^34,35

Some limitations of this study should be acknowledged. The number of malignant tumors is relatively small, but syndromal malignant PNSTs in the context of NF1 are rare tumors, and the number of malignant PNSTs analyzed in this study is comparable to other studies.^17,36 The higher prevalence of benign PNSTs compared with MPNSTs in our study might have affected the results of the statistical analysis. However, robust and significant differences were observed between both groups, and the higher prevalence of benign tumors in NF1 patients is related to the nature of the disease. Due to the non-invasive nature of our study, we did not histopathologically confirm the diagnosis of benign PNSTs in all cases for both practical and ethical reasons. However, all tumors that were categorized as benign showed stable imaging results over at least 12 months, virtually excluding malignant transformation. We did not administer intravenous contrast material in this study. Contrast enhancement has previously been suggested to represent an indicator of malignancy.^13,15,28 However, contrast-enhanced MRI examinations have not been recommended in the International Consensus Statement on Malignant Peripheral Nerve Sheath Tumors,⁶ and a recent work demonstrated that contrast enhancement is not an independent feature of malignancy.¹³ Furthermore, considering the ongoing discussion regarding gadolinium deposits in subcortical structures, an intravenous administration of gadolinium-based contrast agents was not permitted by the institutional review board in this patient cohort, which will typically undergo multiple MR studies over time. Moreover, despite good to very good interrater reliability, the precise cutoffs obtained in our study should not be used at other institutions due to intersystem and intervendor variability of ADC measurements, which is usually minor and lies within the range of test-retest variability.^37,38 Site-specific cutoffs should be established taking into account differences between the equipment and imaging protocols. This is particularly important for biexponential data analysis, which was less robust in this study. Finally, a selection bias due to clinical evaluation of patients cannot be excluded, since our center represents a national referral center for NF1 patients, potentially contributing to the excellent diagnostic performance of functional MRI.

In conclusion, DW MRI with intravoxel incoherent motion or monoexponential data analysis improves accuracy compared with morphological MRI in the differentiation of benign and malignant PNSTs in NF1. Diffusion-weighted MRI may aid the diagnosis of malignant transformation in PNSTs. This study provides a rationale for further prospective studies elucidating the comparative diagnostic performance of DWI techniques and other imaging modalities such as ¹⁸F-FDG PET, or integration of both techniques to improve the diagnosis of malignant transformation in NF1.

Funding

This study was supported by a grant from the “Bundesverband Neurofibromatose e.V.” to J.S.

Conflict of interest statement.

Nothing to declare.

Authorship statement.

All authors were involved in the writing of the manuscript at draft and revision, and have read and approved the final version. In addition, LW and JS performed imaging studies, data analysis, and statistical analysis. MGK developed the software used to perform analysis of DWI data, and supported the statistical analysis. SF has contributed in patient acquisition and clinical evaluation of examined patients. JH contributed by planning of imaging sequences and data analysis due to his experience with previously performed DWI/IVIM projects. KIG contributed in patient imaging and preparation of data for evaluation. CH performed histopathological evaluation of resected tumors. MB was the surgeon in charge for tumor resection. PB and GA contributed in writing and revising the manuscript. VFM and TD conceptualized this study, and have revised the manuscript for important intellectual content.