Author's personal copy Acta Neurochir (2015) 157:855–861 DOI 10.1007/s00701-015-2366-z

CLINICAL ARTICLE - PEDIATRICS

Tumor burden evaluation in NF1 patients with plexiform neurofibromas in daily clinical practice L. Pratt & D. Helfer & L. Weizman & B. Shofty & S. Constantini & L. Joskowicz & D. Ben Bashat & L. Ben-Sira

Received: 12 November 2014 / Accepted: 29 January 2015 / Published online: 14 March 2015 # Springer-Verlag Wien 2015

Abstract Background Existing volumetric measurements of plexiform neurofibromas (PNs) are time consuming and error prone, as they require delineation of PN boundaries, a procedure that is not practical in the typical clinical setting. The aim of this study is to assess the Plexiform Neurofibroma Instant Segmentation Tool (PNist), a novel semi-automated segmentation program that we developed for PN delineation in a clinical context. PNist was designed to greatly simplify volumetric assessment of PNs through use of an intuitive user interface while providing objectively consistent results with minimal interobserver and intraobserver variabilities in reasonable time. Presentation at a conference: L. Pratt, D. Helfer, L. Weizman, B. Shofty, S. Constantini, L. Joskowicz, D. Ben Bashat, L. Ben-Sira. "PNist”: a novel semi-automated volumetric method for easy segmentation of plexiform neurofibromas—a practical tool for the clinicians. Poster presentation at EANS 2013, 11–14 November 2013, Tel-Aviv, Israel L. Pratt (*) : L. Ben-Sira Imaging Division, Tel Aviv Sourasky Medical Center, 6 Weizmann Street, Tel Aviv 64239, Israel e-mail: [email protected] D. Helfer : L. Weizman : L. Joskowicz School of Engineering and Computer Science, The Hebrew University of Jerusalem, Jerusalem, Israel B. Shofty : S. Constantini Department of Pediatric Neurosurgery, and the Gilbert International Neurofibromatosis Center, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel S. Constantini : D. Ben Bashat : L. Ben-Sira Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel D. Ben Bashat Functional Brain Center, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel

Materials and methods PNs were measured in 30 magnetic resonance imaging (MRI) scans from 12 patients with neurofibromatosis 1. Volumetric measurements were performed using PNist and compared to a standard semi-automated volumetric method (Analyze 9.0). Results High correlation was detected between PNist and the semi-automated method (R2 =0.996), with a mean volume overlap error of 9.54 % and low intraobserver and interobserver variabilities. The segmentation time required for PNist was 60 % of the time required for Analyze 9.0 (360 versus 900 s, respectively). PNist was also reliable when assessing changes in tumor size over time, compared to the existing commercial method. Conclusions Our study suggests that the new PNist method is accurate, intuitive, and less time consuming for PN segmentation compared to existing commercial volumetric methods. The workflow is simple and user-friendly, making it an important clinical tool to be used by radiologists, neurologists and neurosurgeons on a daily basis, helping them deal with the complex task of evaluating PN burden and progression. Keywords Plexiform neurofibroma . Neurofibromatosis 1 . Tumor burden . Volumetry

Introduction Plexiform neurofibromas (PNs) are one of the primary features of neurofibromatosis 1 (NF1) [1, 2]. The reported incidence ranges between 30 and 40 % of NF1 patients. This is probably an underestimation, as internal PNs are often undetected without appropriate imaging [1–3]. These tumors have a significant size range, but typically are large and extensive,

Author's personal copy 856

with irregular complex shapes. Their growth rate is erratic and unpredictable [4]. They can involve different parts of the body and may infiltrate, displace, or compress the surrounding structures [1, 2]. Thus, in addition to esthetic disfiguration, they may lead to substantial morbidity [3–5]. In 10 % of the cases these lesions transform into malignant peripheral nerve sheath tumors (MPNSTs) [2, 3]. This serious complication is associated with higher tumor burden and rapid tumor progression. Therefore, tumor burden assessment is essential for detecting aggressive lesions at an early stage and monitoring response to therapy [6, 7]. Magnetic resonance imaging (MRI) is the imaging modality of choice to detect and characterize PNs, which appear bright in relation to adjacent normal tissues when using fatsuppression techniques such as STIR (short-T1 inversion recovery) [8–10]. Currently, evaluation of tumor progression and treatment decisions are based on tumor volume assessment on MRI scans, through a process which is both time consuming and error prone. Delineation of tumor boundaries is often difficult due to tumor inhomogeneity, blurred tumor margins, and signal intensity overlapping with surrounding tissues [5]. In addition, there are technical difficulties related to the quality and consistency of the MRI images, such as inhomogeneity of the magnetic field. Thus, traditional unidimensional and bidimensional measurements are not suitable for efficient, reliable PN volume assessment. Several semi-automatic segmentation volumetric methods have previously been described in the literature, aiming mainly to address the difficulties in PN volume measurement [4, 5, 11]. These methods require intensive and time consuming user interaction during the entire segmentation process for multiple noncontinuous or complex lesions, and have been described mainly as research tools. In previous publications, we presented semi-automatic methods for PN segmentation [12, 13]. PNist (PN Instant Segmentation Tool) [13] is an advanced method that enables easy volumetric quantification of PNs on MRI scans with minimal user intervention. It is designed to be user-friendly and compatible with the busy daily routine of a clinician. PNist relies on simple, easy-to-learn, user interaction to allow accurate PN delineation regardless of size and complexity. This paper describes the validation study of PNist by comparing it to standard-of-care manual measurements, and its ability to evaluate tumor progression, estimating tumor burden volume changes on consecutive scans.

Materials and methods Tel-Aviv Medical Center Institutional Review Board (Helsinki Committee) approved this study. The data collection and volumetric measurements have been performed between the years 2011 to 2012.

Acta Neurochir (2015) 157:855–861

Study population MRI scans of 12 NF1 patients with PNs were retrospectively obtained. All of these patients are routinely followed by the Gilbert Israeli Neurofibromatosis Center (GINFC). Patients’ ages ranged from 4 to 24 years. A total of 30 MRI scans were included in the study. Ten patients had two or three MRI studies in different time intervals that included STIR sequences; two patients had only one MRI study each (Table 1). MRI methodology MRI scans were acquired by a 1.5-T MR System (Signa Excite HDx; GE, Milwaukee, WI, USA), at Tel Aviv Sourasky Medical Center. Images were acquired in coronal or axial planes, and included STIR sequences. The number of slices in each sequence varied between 14 and 48. Voxel sizes varied between 0.4 × 0.4 × 3.3 mm3 and 1.9 × 1.9 × 9.0 mm3. The scans showed lesions in various locations: scalp, neck, shoulder, spine, abdomen, pelvis, and calves. All scans were performed between the years 2006 to 2012. Tumor classification Tumors were classified by radiologist evaluation into three categories, based on tumor complexity and morphological patterns: 1. Simple tumor: Lesion with sharp boundaries, without extension or infiltration to adjacent tissues. 2. Intermediate tumor: Lesion with sharp boundaries, with extension or infiltration to adjacent tissues. Table 1

Study cohort: demographics and tumor location

Patient

Gender

No. of MRI studies

Tumor location

Type

1 2 3 4 5 5 5 6 7 8 9 10 11 12

M F F F F F F F M M F F F M

2 3 2 3 2 2 1 3 2 3 3 2 1 1

Abdomen Scalp Abdomen Scalp Calves Pelvis Total body Scalp Shoulder Neck Neck Neck Neck Abdomen

Intermediate Simple Intermediate Simple Simple Intermediate Intermediate Intermediate Intermediate Complex Complex Complex Complex Complex

Author's personal copy Acta Neurochir (2015) 157:855–861

857

3. Complex tumor: Lesion with blurred margins, in which it was difficult to distinguish between tumor and surrounding tissues, with extension or infiltration to adjacent tissues.

PNist overview We used PNist for segmentation of the tumors (for a detailed technical description of the method, see [13]). PNist relies on predefined tumor models obtained from a training phase. The input to the training phase is a set of STIR MR scans and manual delineation of PNs in those scans. Scans and delineations are used to create models of expected intensity distribution in the vicinity of tumors, which are then stored in a histogram database. Information from the database, together with a user scribble on a new scan is then fed into the interactive phase of the method to segment PNs in the new STIR MRI scans. Figure 1 illustrates the flow of the algorithm. The segmented area from one slice can be copied automatically to the adjacent slice and used as input there, effectively working in three-dimensions. The interactive phase is repeated as many times as required, with an option of manual correction, until the user is satisfied that all tumor voxels were labeled. A simple to use graphical user interface (GUI) is used to delineate the tumors (Fig. 2). In addition to basic draw and erase operations, it has a Smart Draw operation: instantly segmenting bright objects in the vicinity of the first simple mouse stroke. For illustration clip please refer to http:// youtu.be/XNcIl7_l29E.

0.1

0.1

0.1

0.05

0.05

0.05

0

0

10

20

30

40

50

0

0

10

20

30

40

50

0

0.1

0.1

0.1

0.05

0.05

0.05

0

0

10

20

30

40

50

0

0

10

20

30

40

50

0

0.1

0.1

0.1

0.05

0.05

0.05

0

0

10

20

30

40

50

0

0

10

20

30

40

50

0

0.1

0.1

0.1

0.05

0.05

0.05

0

0

10

20

30

40

50

0

0

10

20

30

40

50

0

0.1

0.1

0.1

0.05

0.05

0.05

0

0

10

20

30

40

50

0

0

10

20

30

(a)

40

50

0

0

10

20

30

40

50

0

10

20

30

40

50

0

10

20

30

40

50

0

10

20

30

40

50

0

10

20

30

40

50

(b)

Fig. 1 Stages of the segmentation algorithm. a Learning phase: Image intensity histograms of previously segmented PN tumor environments are normalized and clustered into 15 groups, creating our database of samples. This initialization process is performed only once. b User input: Given a new coronal STIR MRI of the abdomen, user scribbles over the tumor to sample the local intensity distribution. The histogram of the pixels under the thick red line is computed and matched against the samples database. c Thresholding: Using the most similar histogram in

Study methodology To evaluate the new PNist segmentation method, an expert radiologist (LP) segmented the PNs on STIR sequences in each MRI study from the data set. A total of 90 independent segmentation sessions were completed (three segmentations for each of the 30 MRI scans). All 30 scans were then segmented using a commercial general-purpose volumetric segmentation tool (Analyze 9.0). The measurements were performed on different days for each scan. The interaction time required for 21 PNist segmentation sessions was recorded and compared to the interaction time required for Analyze 9.0 segmentation. To evaluate inter-observer variability, a second expert radiologist (L.B.S.) segmented ten scans from the data set, chosen arbitrarily.

Results Correlation between PNist and analyze 9.0 The volumes measured for all PNs, using both methods, ranged from 5.1 to 611 cm3. There was high correlation (R2 =0.996) between the volumetric measurements done with PNist compared to Analyze 9.0 (Fig. 3). The correlation was lower (R2 =0.887) for smaller tumors (less than 50 cm3). Volume overlap error Overlap is a measure of similarity between segmentations, regarding their location in the space. The volume overlap difference (volume overlap error [VOE]) between the sets

(c)

(d)

the samples database, the algorithm automatically finds a threshold grey level that distinguishes the tumor from its background. The thresholding mask acknowledges the fact that tumors are not the only structures brighter than the selected threshold. d Component filtering: Bright pixels are classified as tumors only if they belong to a connected component that intersects with the original user input (in this case, the red line in b)

Author's personal copy 858

Acta Neurochir (2015) 157:855–861

Fig. 2 PNist segmentation process graphical user interface. a Scan loading: Coronal STIR MRI of the shoulder. b Manual tracing of the tumor (user input). c Segmentation created immediately (red coloring), including the hyperintense components of the tumor

measured with both algorithms was calculated using the following formula: 2⋅jA∩Pj 1−








A
þ
P
where |·| denotes set size, A is the Analyze 9.0 segmented volume, and P is the PNist segmented volume. The mean VOE was computed for the 30 scans that were segmented three times, for every pair of segmentations (three segmentations made with PNist were paired with the matching single segmentation made with Analyze 9.0, yielding three different pairs). The simple tumor group had a lower average overlap error value (6 %) than the intermediate and complex tumor groups (11.9 and 10.2 % respectively). The overall average volume overlap error was 9.54 %.

PNist was shorter when compared to Analyze 9.0. The average user interaction time for the 21 measurements made with PNist was 360 s, compared to 900 s with Analyze 9.0. This constitutes an overall reduction of 60 % with PNist compared to Analyze 9.0. Table 2 shows that the user interaction time was shorter for all subgroups (simple, intermediate, and complex). Intraobserver variability The intraobserver variability of the volume being measured (coefficient of variation, CV=standard deviation/mean), calculated for the 90 segmentations generated from the 30 MRI scans, ranged from 0.17 to 6.28 % (mean 3 %, STD 1.8 %). Fig. 3 illustrates the proximity of the maximal, minimal, and average volumetric values measured with PNist for each MRI study. Interobserver variability

Segmentation time Twenty-one segmentations were performed while documenting the work-flow, including measuring the duration of the segmentation process. Overall, the segmentation time measured with Fig. 3 Correlation between PNist and Analyze 9.0. We analyzed the correlation between volumetric values measured with Analyze 9.0 and the maximal, minimal, and average volumetric values measured with PNist (PNmax, PNmin, and PN Average, respectively) for each study

There was high correlation between measurements of the volume made by the two radiologists (R2 =0.997). The interobserver variability ranged from 0.8 to 11.4 % (mean 4.25 %, STD 6.4 %).

Author's personal copy Acta Neurochir (2015) 157:855–861 Table 2 type

859

Average segmentation time (in seconds) according to lesion

Tumor type

PNist time (seconds±SD)

Analyze time (seconds±SD)

Simple Intermediate Complex

125 (±53) 273 (±179) 840 (±364)

240 (±66) 394 (±162) 1,350 (±642)

Comparison of tumor volume over follow-up time To evaluate the accuracy of PNist when assessing tumor volume changes on follow-up studies (the “Delta” between sequential measurements), we defined a Delta Comparison (DC) quantifier: . P P0 1 PNist△ Delta Comparsion ¼ ¼ . Analyze△ A A 1 0 Where P0 and A0 denote a previous volume measurement with PNist and Analyze 9.0 respectively, and P1 and A1 denote the consecutive volume measurement with PNist and Analyze 9.0 respectively. Ideally, DC should be as close to 1 as possible, indicating that both tools measure volume changes identically. Ten study patients had two or three MRI scans of the same PN. We computed the DC for every pair of consecutive exams, a total of 15 pairs. Figure 4 illustrates the high accuracy of the segmentation tools, with a mean DC=0.983 and a standard deviation of 8.7 %. Systematic and/or random errors that might exist in the absolute measurements might decrease or increase the accuracy of the measured tumor volume change. To evaluate this scenario, we computed the following absolute and relative values:


 

P1

P0



Abs Error ¼ max
−1
;
−1

 100% A1 A0 and 2
.
3
P P

1 0
PNist△ −1  100% ¼ 4

.

−15  100% Relative Error ¼ Analyze△
A A


Abs Error represents the maximal mismatch between the segmentation methods in the absolute measurement values of the same tumor on consecutive studies. Relative Error represents the delta mismatch between the tools. Figure 5 illustrates the relationship between absolute and relative errors. The points below the diagonal slope represent measurements in which the delta mismatch is smaller compared to the individual absolute volume values obtained with the two algorithms. Note the two measurements in the bottom right corner of the graph: both have an absolute volume measurement mismatch between the tools that exceeded 25 %. However, the delta mismatch for these measurements was smaller than 3 %. This indicates a systematic mismatch between the tools, which canceled out in the relative delta measurements. For some measurements the volume change mismatch is larger than the individual values mismatch (points above the diagonal line), however, they are still much lower than the worst case line, which denotes the worst case summation of the errors.

Discussion In this study we propose a novel tool that allows for easy, fast, tumor burden assessment of PNs. Using this tool, the treating physician and radiologist are able to rapidly assess tumor burden and progression in their office setting, with minimal interruption to their daily workflow. The simplicity, ease of use and accuracy of PNist makes it unique among other volumetric methods in this field. Previously described tools usually rely on delineation of a region of interest, which requires more time and a steady hand, or involves seed initiation, which may be tiresome when many small non-continuous masses are involved. The input to PNist is a free-hand brush over the object or objects of interest, performed in less than a second and requires less mental exertion and accuracy than other types of input. Few articles have been published regarding volumetric methods designed specifically for accurate PN measurements.



1

0

Fig. 4 Delta comparison between PNist and Analyze 9.0

Fig. 5 Relative versus absolute error measurements

Author's personal copy 860

Solomon et al. [4] described a segmentation algorithm based on histogram analysis, in which the analyst finds a threshold value that distinguishes between intensity levels of normal tissue and lesions. Solomon’s two-dimensional volumetric program requires the user to define a region of interest in each MRI slice containing the tumor. The program then automatically segments all tumors within the region of interest. This method was proven reproducible with coefficient of variation between 0.6 and 5.6 %, and correlated with manual tumor tracing (R=0.999). However, the algorithm eliminates connected components smaller than a predetermined size, resulting in an inability to detect small tumor lesions. This may result in overlooking active disease foci or missing significant data when assessing treatment response. Cai et al. [11] measured tumor burden in patients with neurofibromatosis 1 and 2 and schwanomatosis through whole-body MRI, using coronal STIR scans. They estimated tumor burden with a three-dimensional segmentation algorithm called the “dynamic–threshold level set” in which the user has to identify the center of each PN. Their method then expands automatically from the user’s predefined central point to segment the entire lesion in three dimensions, according to statistical measures. Cai et al.’s method was reliable when compared to manual segmentation (ricc =0.99), less labor intensive, and more repeatable. Nevertheless, these methods are based on prolonged user-interaction and do not fit well into most radiologists’ daily clinical work schedule. Our results show that the PNist algorithm reliably defines and measures PN lesions when compared to an existing commercial method (Analyze 9.0), with high repeatability and reproducibility. Comparable results for interobserver variability [4, 5] and intraobserver variability [4] have been described in previous publications. PNist was found to be reliable for absolute volume values, as well as for three-dimensional shape of created segmentations as measured by the VOE. The mean VOE between PNist and the ground truth was 9.54 % in the present study. In our earlier work [12], the mean VOE between a semi-automated PN segmentation algorithm and the ground truth was 27 %. No other published articles regarding proposed segmentation algorithms provided information regarding overlap quality in PN segmentation. Another advantage of PNist is the short computation time—a clear improvement in the required segmentation time compared to Analyze 9.0. There are no previous data reported regarding the required segmentation time with the aforementioned alternative volumetric methods, but we estimate that our method is less time consuming, especially for measuring complex tumor subgroups or lesions that contain many small unconnected masses. There are a few limitations to PNist. Although we found a high correlation between measurements obtained with PNist in comparison to those obtained with Analyze, the correlation was slightly lower when comparing only small volume tumors

Acta Neurochir (2015) 157:855–861

of less than 50 cm3. A similar trend was reported by Solomon et al. [4]. This finding is probably related to the fact that small volume tumors are prone to large relative volume differences in consecutive segmentations. For example, an absolute volume difference of 1 cm3 between two measurements, which is hardly detectable by the user, yields a relative volume error of 25 % when the tumor’s overall volume is 4 cm3. Nevertheless, previous works reported that PN size and location are the most important factors in predicting clinical outcome and chances for malignant transformation. Thus, it is most important to monitor large-sized tumors and internal PNs that are difficult to assess with physical examination. Another limitation common to all volumetric techniques relying on diffusion of the created segmentation, has previously described by Cai et al. [11], and concerns diffusion of the segmentation to high-signal-intensity structures in the vicinity of the segmented lesion. This happens when nontumoral structures (such as cerebrospinal fluid in the spinal canal or urinary bladder), are brighter than the algorithm threshold and the structure’s connected component intersects with the algorithm input. All of these bright structures will then be mistaken for tumor. However, this can be very easily and efficiently corrected in PNist by minimal user intervention. We also have developed a predelineation option for high signal structures, thereby excluding them from segmentation even before starting the measurement process. One of the more attractive PNist capabilities is its ability to follow volumetric PN changes on consecutive MRI scans. Radiologists must detect tumor volume changes to predict clinical complications and evaluate the possibility of malignant transformation. An efficient tool is required to determine if a lesion has been stable or is changing in size over time. Therefore, in addition to evaluating PNist accuracy in measuring absolute volumes for PNs, we also assessed the method’s ability to track relative volume changes over time. We demonstrated that PNist reliably assesses relative volume changes compared to Analyze 9.0. Furthermore, even when there was a large mismatch between the absolute volumetric values measured with both tools for the same lesion at a specific point of time, the relative size change of the lesion on consecutive scans was similar (with differences of less than 13 %) between the two tools. There are not enough data in the literature regarding the natural history of PNs, probably because of the complexity of these tumors, their frequent internal location, and the cumbersome volumetric methods currently in use, which are not suitable for daily routine assessment. Most previous studies assessed growth characteristics of PNs on regional MRI scans of known or suspected lesions [3, 14]. Whole-body tumor burden and asymptomatic tumors cannot be evaluated adequately with regional techniques, since PNs can appear in different locations in the body. In addition, it has been found that different PNs have different growth rates in the same patient. Thus, a single lesion cannot represent the growth rate

Author's personal copy Acta Neurochir (2015) 157:855–861

of the other lesions, and separate follow-up is needed for each lesion [3]. Several studies have shown that whole-body MRI (WBMRI) is an efficient tool for evaluating PN tumor burden in NF1 patients [11, 15, 16]. This technique enables detection and characterization of PN size and morphology throughout the body, and can be used for both baseline assessment and future follow-up. Plotkin et al. [17] showed that WBMRI detects internal tumors in a far higher percentage of patients compared with regional imaging methods, highlighting the importance of using this technique for NF1 patients. They found that 60 % of NF1 patients had internal nerve sheath tumors identified by WBMRI, compared with 16–39 % found using regional imaging techniques. Furthermore, they characterized the distribution of lesions to distinct anatomic subgroups. To date, only one published work addresses natural PN growth dynamic using volumetric estimation of tumor burden on WBMRI scans of NF1 patients. In this study, NF1 patients with serial WBMRI scans were monitored over 1.1–4.9 years. The results showed that PN growth rate is correlated with tumor burden and inversely correlated with age, suggesting that younger patients, and patients with high tumor burden, warrant close clinical and radiological follow-up. In addition, patients with no internal PNs on the first MRI scan are unlikely to develop lesions later on. They also found that some lesions decreased in size during follow-up, but it seemed likely that most of these observations represented measurement error. They concluded that long-term natural history studies with WBMRI monitoring and clinical assessment are needed. Preliminary results show that PNist can reliably measure the tumor burden in whole-body coronal STIR MRI scans (WBMRI) in a relative short computation time, with minimal staff intervention required.

Conclusion We present the first clinical evaluation study of PNist, a novel semi-automated segmentation tool for volumetric measurements of PNs. PNist is a clinical office-based tool designed to be used by radiologists, neurologists and neurosurgeons on a daily basis, allowing them to deal with the complex task of evaluating PN burden and progression. PNist was found to be accurate for absolute volumetric measurement and for defining the shape of the created segmentation. It has been proven reliable in assessing PN size changes over-time. The workflow is simple and user-friendly, with a short, interactive computation time that can easily be incorporated into a busy work schedule. Conflicts of interest None of the authors has any conflict of interest and/or commercial stake in the evaluated software.

861

References 1. Mautner VF, Hartmann M, Kluwe L, Friedrich RE, Funsterer C (2006) MRI growth patterns of plexiform neurofibromas in patient with neurofibromatosis type 1. Neuroradiology 48:160–165 2. Williams VC, Lucas J, Babcock MA, Gutmann DH, Korf B, Maria BL (2009) Neurofibromatosis type 1 revisited. Pediatrics 123:124– 133 3. Tucker T, Friedman JM, Friedrich RW, Funsterer C, Mautner VF (2009) Longitudinal study of neurofibromatosis 1 associated plexiform neurofibromas. J Med Genet 46(2):81–85 4. Solomon J, Warren K, Dombi E, Patronas N, Widemann B (2004) Automated detection and volume measurement of plexiform neurofibromas in neurofibromatosis 1 using magnetic resonance imaging. Comput Med Imaging Graph 28(5):257–265 5. Poussaint TY, Jaramillo D, Chang Y, Korf B (2003) Interobserver reproducibility of volumetric MR imaging measurements of plexiform neurofibromas. AJR Am J Roentgenol 180(2):419–423 6. Nguyen R, Kluwe L, Fuensterer C, Kentsch M, Friedrich RE, Mautner VF (2011) Plexiform neurofibromas in children with neurofibromatosis type 1: frequency and associated clinical deficits. J Pediatr 159(4):652–655 7. Nguyen R, Dombi E, Widemann BC, Solomon J, Fuensterer C, Kluwe L, Friedman JM, Mautner VF (2012) Growth dynamics of plexiform neurofibromas: a retrospective cohort study of 201 patients with neurofibromatosis 1. Orphanet J Rare Dis 7:75 8. Ros PR, Eshaghi N (1991) Plexiform neurofibroma of the pelvis: CT and MRI findings. Magn Reson Imaging 9:463–465 9. Stull MA, Moser RP, Kransdorf MJ, Bogumill GP, Nelson MC (1991) Magnetic resonance appearance of peripheral nerve sheath tumors. Skelet Radiol 20(1):9–14 10. Tien RD, Hesselink JR, Chu PK, Szumowski J (1991) Improved detection and delineation of head and neck lesions with fat suppression spin-echo MR imaging. AJNR Am J Neuroradiol 12:19–24 11. Cai W, Kassarjian A, Bredella MA, Harris GJ, Yoshida H, Mautner VF, Wenzel R, Plotkin SR (2009) Tumor burden in patients with neurofibromatosis types 1 and 2 and schwannomatosis: determination on whole-body MR images. Radiology 250:665–673 12. Weizman L, Hoch L, Ben-Bashat D, Joskowicz L, Pratt LT, Constantini S, Ben Sira L (2012) Interactive segmentation of plexiform neurofibroma tissue: method and preliminary performance evaluation. Med Biol Eng Comput 50(8):877–884 13. Weizman L, Helfer D, Ben Bashat D, Pratt LT, Joskowicz L, Constantini S, Shofty B, Ben Sira L (2014) PNist: interactive volumetric measurements of plexiform neurofibromas in MRI scans. Int J Comput Assist Radiol Surg 9(4):683–693 14. Dombi E, Solomon J, Gillespie AJ, Fox E, Balis FM, Patronas ER, Korf BR, Babovic-Vuksanovic D, Packer RG, Belasco J, Goldman S, Jakacki R, Kieran M, Steinberg M, Weidemann BC (2007) NF1 plexiform neurofibroma growth rate by volumetric MRI: relationship to age and body weight. Neurology 68:643–647 15. Mautner VF, Asuagbor FA, Dombi E, Funsterer C, Kluwe L, Wenzel R, Widemann BC, Friedman JM (2008) Assessment of benign tumor burden by whole-body MRI in patients with neurofibromatosis 1. Neuro Oncol 10:593–598 16. Van Meerbeeck SFL, Verstraete KL, Janssens S, Mortier G (2009) Whole body MR imaging in neurofibromatosis type 1. Eur J Radiol 69:236–242 17. Plotkin SR, Bredella MA, Cai W, Kassarjian A, Harris GJ, Esparza S, Merker VL, Munn LL, Muzikansky A, Askenazi M, Nguyen R, Wenzel R, Mautner VF (2012) Quantitative assessment of wholebody tumor burden in adult patients with neurofibromatosis. PLoS ONE 7(4):e35711