Golden-Angle Radial Sparse Parallel (GRASP) MRI differentiates head & neck paragangliomas from schwannomas

Abstract

Purpose:

The aim of this study was to demonstrate the value of DCE MRI with high spatiotemporal resolution (GRASP) for differentiating paragangliomas and schwannomas in the head and neck.

Methods:

In a retrospective PACS search of in total 410 patients who had undergone head & neck GRASP-MRI, we identified 6 patients with biopsy proven cervical paragangliomas (n = 3) and schwannomas (n = 3). Conventional MRI features were evaluated, lesion size was determined. Postprocessing in 4D-GRASP datasets was performed (1) based on reconstructions with a temporal resolution (Tres) of 4.1 s, qualitative time-intensity curve classification and semiquantitative parameter (T_peak, PH, ER_max and Slope_max) analysis, and (2) voxel-based mapping and qualitative and semiquantitative perfusion modeling based on reconstructions with a Tres of 1.6 s. Additionally, GRASP perfusion analysis was performed in another set of 5 patients with presumed cervical paragangliomas (n = 3) and schwannomas (n = 2) based on conventional imaging criteria and was correlated with conventional imaging findings. Due to the small sample size, both groups were compared qualitatively.

Results:

In the time intensity curve classification of 4D GRASP reconstructions (Tres 4.1 s), biopsy proven paragangliomas were consistently characterized by a type-III rapid inflow wash-out pattern, compared to a type-I inflow pattern in the schwannoma group. In both temporal resolutions, semiquantitative analysis of time intensity curves demonstrated rapid wash-in, wash-out, and higher peak signal intensities in paragangliomas compared to schwannomas. In 5 presumed (non-biopsy-proven) paragangliomas and schwannomas, time intensity curves improved diagnostic certainty.

Conclusions:

Visual time intensity curve classification and semi-quantitative analysis of GRASP-MRI were, in this small retrospective series, sufficient to differentiate cervical paragangliomas from schwannomas. Utilization of this technique may further improve diagnostic confidence in lesions lacking conventional imaging features.

1. Introduction

Paragangliomas (PGs), deriving from the extra-adrenal neuroendocrine system, constitute a rare tumor entity in the head and neck (H&N) region (0.6% of all H&N neoplasms). PGs most commonly originate from the paraganglia or glomus cells of the carotid body, middle ear and/or jugular foramen (together 80%), and vagal nerve (5%) [1].

From an imaging point of view, PGs are known to enhance avidly in the arterial-/late-arterial phase in post-contrast sequences and in dynamic contrast-enhanced imaging studies, such as CTA/CTP, MRA, and DSA [2,3], including a ‘signal drop-out’ behavior on dynamic contrast-enhanced MRI [4]. These findings have been correlated pathologically to intrinsic AV-shunts with early draining veins and have also been related to characteristic T2w-hypointense flow-voids, together with punctate T1w-hyperintesities (‘salt and pepper’ pattern). In 10% of patients, PGs tend to be multifocal and, despite being WHO Grade I, lesions may present with regional or distant metastases in 6–19% of cases [5,6].

Depending on the location of the lesion, especially along the carotid sheath, the main imaging differential diagnosis constitutes benign vagal schwannoma (SC), which is also characterized by T2w-hyperintense signal and intense contrast enhancement. Further differentiating morphologic features, as e.g. the splaying of the carotid bifurcation, have been extensively described in PGs [7–9]. Both entities constitute the most common primary masses in the carotid space.

Despite characteristic conventional imaging features, differential diagnosis can be challenging. Biopsy is performed in non-evident cases, which, however, is accompanied by a significant periprocedural risk with severe bleeding in typically highly vascularized PGs and/or due to injury of adjacent structures. Surgical treatment is the therapy of choice in PGs [9]. Postsurgical outcome in PGs depends on total resection, which is usually more successful in carotid body PGs compared to tumors originating at the skull base. Surgery may be accompanied by significant morbidity due to intraoperative hemorrhage and lesions of adjacent neuronal structures, resulting in transient or persistent deficits. This underlines the necessity for high diagnostic accuracy in diagnosing PGs, especially in cases in which biopsy is unfavorable.

Available literature on the diagnostic confidence in differentiating PGs from SCs by conventional MR imaging (without dynamic contrast-enhanced imaging techniques) is controversial [3,10]. Detection and accurate lesion classification may significantly depend on reader experience. Especially in smaller carotid-space lesions, the differentiation between a small SC and a PG can be challenging. Dynamic contrast-enhanced (DCE) MRI can improve diagnostic sensitivity by revealing tissue-specific enhancement patterns. So far, however, only few studies directly compared DCE patterns in PGs and SCs, which may be explained by the rarity of PGs, sparse availability of histopathologic correlates [3], and limited availability of suited DCE MRI techniques, as most conventional DCE MRI techniques are limited with respect to the achievable spatial anatomical or temporal resolution.

Golden-Angle Radial Sparse Parallel (GRASP) MRI is a relatively new DCE-MRI technique that simultaneously provides high spatial and high temporal resolution [11]. Additionally, it offers improved robustness to motion and flow artifacts, as well as the flexibility to select the desired reconstructed temporal resolution retrospectively. GRASP has been used in the evaluation of tissue-specific enhancement dynamics, e.g. in pituitary, parotid, cardiac, liver, and breast imaging [12–15,17], and it has been evaluated in comparison to 2D FLASH, 3D FLASH, VIBE, and TWIST sequences in a phantom model, delivering better reproducible arterial input function measurements compared to less/non-accelerated sequences [18].

Based on the available phantom data and data from prior dynamic contrast-enhanced imaging in PG and SC [2–4,18], we hypothesize that GRASP can provide reliable time-intensity curves with discriminative parameters for PGs and SCs in the H&N region and, therefore, has the potential to improve the diagnostic confidence in imaging-based differential diagnosis in PGs and SCs.

The GRASP technique includes a number of improvements over existing techniques used in previous studies of DCE imaging in paragangliomas. GRASP produces high-resolution T1 weighted images with a large anatomical coverage while providing high temporal permeability images of the entire volume. GRASP uses a continuous radial stack-of-stars acquisition of k-space, and an arbitrary number of consecutive radial spokes can be combined into a single image frame to create DCE imaging of a desired temporal resolution. Additionally, by combining all spokes into a single image frame, one can create high-resolution fat-suppressed anatomic images of the entire neck with an excellent signal-to-noise ratio. Furthermore, the radial sampling pattern makes it relatively insensitive to motion, which is an advantage especially in regions where swallowing may degrade image quality. Thus, GRASP can be routinely integrated into a neck imaging protocol as a diagnostic anatomic sequence without using examination time solely to obtain permeability information.

Therefore, our study aims to 1) demonstrate the feasibility of GRASP MRI to differentiate between PGs and SCs based on GRASP MRI time-intensity curves using two postprocessing algorithms for different temporal resolutions (simple semi-quantitative (moderate temporal resolution = Tres) and semi-automated (high Tres)) in patients with histologically proven PGs or SCs (n = 6) and 2) to evaluate GRASP MRI in a cohort of presumed PGs and SCs (n = 5) or SCs (n = 6). Secondly, GRASP MRI is evaluated in a cohort of presumed PGs and SCs (n = 5).

2. Materials and methods

2.1. Patients

A retrospective review of the institutional PACS database was performed. Out of 410 patients who underwent GRASP MRI of the H&N region, 11 patients being diagnosed with a cervical PG (n = 6; n = 3 biopsy proven) and SC (n = 5; n = 3 biopsy proven) were identified. Diagnosis was independently confirmed by 2 blinded readers (2 board-certified neuroradiologists with 4 and > 10 years of experience in clinical neuroimaging at a university medical center; see paragraph below ‘qualitative image parameters’). Two patients were diagnosed with a familial PG syndrome. Our institutional review board approved this retrospective study.

2.2. MRI acquisition

GRASP MRI was performed using MAGNETOM Avanto FIT 1.5 T, MAGNETOM Skyra 3 T, and MAGNETOM Skyra FIT 3 T scanners (Siemens Healthineers, Erlangen, Germany) utilizing 20-channel head/neck coils. All patients were prepared in the scanner in a supine position. The conventional exam protocol consisted of an axial T1-weighted (T1w) TSE sequence (slice thickness ST = 4 mm), axial and coronal T2w FS/TIRM sequences (ST = 4 mm), and DWI acquisition. Sensitizing diffusion gradients were applied sequentially in the x, y, and z direction with b-values set to 0 and 1000 s/mm². Contrast-enhanced sequences included a dynamic GRASP acquisition with simultaneous contrast injection (injection speed 2.0 ml/s via 20G IV catheter, injection delay 20s, total dynamic acquisition time 4:23 min, 1.5 mm isotropic resolution), and an axial and coronal T1w FS post-contrast sequence (ST = 4 mm, Fig. 1). GRASP-specific sequence parameters are presented in Table 1.

Fig. 1.

Current in-house H&N MRI protocol in a (histologically proven) L carotid space paraganglioma including axial T2w FS (A), T1w axial (B), TOF-MRA axial (C), DWI/ADC axial (D), post-contrast T1w axial (E), and post-contrast T1w GRASP axial.

Table 1.

GRASP MRI acquisition parameters.

Scanners	MAGNETOM Avanto FIT 1.5 T, MAGNETOM Skyra 3 T, MAGNETOM Skyra FIT 3 T (Siemens Healthcare, Erlangen, Germany)
Coils	20-channel head & neck coil array
Sequence type	FS T1w stack-of-stars radial 3D GRE with golden-angle ordering (GRASP)
TR/TE/FA	4.1 ms / 1.9 ms / 12°
FOV	240 × 240 mm
Voxel size	1.5 mm isotropic
Slice thickness	1.5 mm
Total acquisition	4:23 min
Fat saturation	Q-FS
Contrast agent	Single dose of Gd-DOTA (Dotarem®) or Gadobutrol (Gadovist®) at injection rate of 2 ml/s, followed by a cc saline flush with power injector

FA = flip angle; FOV = field of view; Gd-DOTA = Gadoteric acid; TR = repetition time; TE = echo time.

2.3. Qualitative image analysis

Assessed conventional imaging characteristics included location, morphology, and T1/T2 signal characteristics, as well as T1w-post-contrast enhancement pattern and pattern on vascular displacement. In addition, the report findings were recorded based on the primary assessment of a board-certified radiologist (Table 2, column 10).

Table 2.

Patient characteristics.

Case	Sex	Age in years	Location	T1WI	T2WI	Flow Voids on T2WI	Enhancement pattern on Post-Gd T1WI	Pattern on vascular Displacement	Report	Histopathology
1	F	70	L CS	Isointense with punctate hyperintensities	Hyperintense	Y	Heterogenous with central sparing	Splaying of ICA and ECA	PG	n.a.
2	F	27	R CS	Isointense with few punctate hyperintensities	Hyperintense	Y	Heterogenous with central late-hyperenhancement	Anterior CCA. Anterolateral IJV.	PG	PG
3	F	70	L JF/CS	Heterogenous with multiple punctate hyperintensities	Hyperintense	Y	Heterogenous	IJV obliterated	PG	n.a.
4	F	60	R CS	Isointense	Hyperintense	Y	Homogenous	Splaying of ICA and ECA	PG	PG
5	F	51	L CS	Heterogenous with multiple punctate hyperintensities	Hyperintense	Y	Heterogenous	Splaying of ICA and ECA	PG (LN-Met)	PG
6	F	55	L JF/tympanon	Heterogenous with multiple punctate hyperintensities	Hyperintense	Y	Heterogenous	–	PG	n.a.
7	M	56	R CS	Isointense	Hyperintense	N	Heterogenous peripheral enhancement	Anteromedial ICA and ECA, anterior IJV	SC	SC
8	F	68	R CS	Isointense	Hyperintense	N	Heterogenous peripheral enhancement	Anteromedial CCA, anterior IJV	SC/NF (PG)	SC
9	F	67	R C1/2 Foramen	Isointense	Hyperintense	N	Homogenous	–	SC	n.a.
10	M	78	L C3/C4 Foramen	Isointense	Hyperintense	N	Homogenous	–	SC	n.a.
11	F	52	R CS	Isointense	Hyperintense	N	Heterogenous peripheral enhancement	No displacement	SC/NF	SC

CCA = common carotid artery; CS = carotid space; ECA = external carotid artery; ICA = internal carotid artery; IJV = internal jugular vein; JF = jugular foramen; L = left; n.a. = data not available; LN-Met = Lymph Node-Metastasis; NF = Neurofibroma; PG = paraganglioma; R = right; SC = Schwannoma; T1WI = T1-weighted image; T2WI = T2-weighted image.

2.4. Postprocessing & semi-quantitative analysis

Collected GRASP raw datasets of both groups were retrospectively reconstructed with two different temporal resolutions, using custom-developed postprocessing-infrastructure [19]:

1. High temporal resolution of Tres = 1.6 s

2. Medium temporal resolution of Tres = 4.1 s

Further postprocessing of the reconstructed images was performed (1) using Syngo Tissue 4D (Siemens Healthineers, Erlangen, Germany) by calculating relative time-concentration curves and a pharmacokinetics model based on a presumed arterial input function (AIF) based on T1-maps with an assumed tissue value of 1000 ms. Three options are available for the population-averaged AIF (fast, intermediate, and slow), and the optimum AIF was chosen based on the volume of interest (VOI) curve of each patient. 2D ROIs in (1) and (2) were drawn at the same plane into the solid center (mid-part) of a lesion, demonstrating the greatest degree of enhancement in dynamic TICs, excluding vessels and cystic or necrotic parts.

Similarly to previous studies [2], in (2) signal intensity (SI_pre, SI_max) and time parameters (T_pre, T_peak) were extracted from each TIC. SI_pre was defined as the mean pre-contrast signal intensity. SI_max was the maximum signal intensity after contrast injection. T_pre and T_peak were the corresponding time points. Furthermore, PH, ER_max, and Slope_max were calculated according to recent literature [20–23]:

$$\text{PH}=\mathrm{SImax}-\text{SIpre}$$

$$\text{ERmax}=\frac{\text{SImax}-\text{SIpre}}{\text{SIpre}}$$

$$\text{SLOPEmax}=\frac{\text{SImax}-\text{SIpre}}{\text{SIpre}\text{x}(T\text{peak}-T\text{pre})}$$

PH = peak height; SImax = maximal signal intensity; SIpre = pre contrast signal intensity; ERmax = maximal Enhancement Ratio; SLOPEmax = maximal slope; Tpeak = Time to peak enhancement; Tpre = upslope Timepoint.

TIC metrics including T_peak, PH, ER_max, and Slope_max were normalized to values obtained from the ipsilateral common carotid artery.

The analysis included biopsied lesions as well as unbiopsied lesions that had been classified as probable PG or SC based on current imaging criteria (see Fig. 4 and Table 3).

Fig. 4.

Signal-ratio time intensity curves (TICs) of H&N paragangliomas (PG, red) and schwannomas (SC, blue). Dashed lines: presumed paragangliomas and schwannomas, respectively; based on current conventional MR imaging criteria. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 3.

Time intensity curve (TIC) analysis (Tres = 4.1 s) including non-biopsy proven lesions.

Case	Diagnosis	Histologic proof	Sizea(mm)	TIC pattern	Tpeak (s)	PH	ERmax	Slopemax
1	PG	N	59	III	0.80	0.71	3.05	0.45
2	PG	Y	36	III	0.83	0.97	1.20	0.93
3	PG	N	22	III	1.00	0.75	1.02	0.75
4	PG	Y	26	III	0.80	0.97	3.25	0.74
5	PG	Y	26	II/III	0.83	1.21	2.49	0.84
6	PG	N	18	III	1.00	0.64	0.80	0.66
7	SC	Y	44	I	9.00	0.36	0.52	0.33
8	SC	Y	24	I	8.67	0.42	0.96	0.32
9	SC	N	28	I	9.25	0.36	0.39	0.43
10	SC	N	33	I	6.00	0.96	0.87	0.86
11	SC	Y	38	I	8.33	0.29	0.74	0.23

PG = paraganglioma; SC = schwannoma; the longest diameter on the axial slices; TIC = Time intensity curve; Tpeak = Time to peak enhancement; PH = peak height; ERmax = maximal Enhancement Ratio; Slopemax = maximal slope. Metrics normalized to values obtained from ipsilateral CCA.

2.5. Qualitative time-intensity curve classification

Time Intensity Curves (TICs; x-axis: time, y-axis: signal intensity) based on 4.1 s reconstructions were categorized according to previous studies [2] by two board-certified neuroradiologists with 4 and > 10 years of clinical neuroradiology expertise into three distinct patterns: the persistent (type-I) pattern with straight or curved line and continuous enhancement over the entire dynamic study; the plateau (type-II) pattern with a relatively prominent slope and a final intensity of 90–100% of peak intensity; the (type-III) wash-out pattern with a rapid slope and a final intensity lower than 90% of peak grade.

2.6. Statistical analysis

Considering the small sample size and exploratory nature of the study, a detailed qualitative analysis was performed. Analysis and scatter plots were performed with Excel 2013 (Microsoft Inc., Redmont, WA, USA) and SPSS Statistics 22 (IBM, Armonk, NY, USA). The DCE-MRI parameters were summarized numerically as mean ± standard deviation (SD).

3. Results

3.1. Patient demographics

In our database, 11 patients (9 female, 2 male) were identified who underwent conventional MRI including GRASP MRI of the H&N that revealed either a cervical PG (n = 6) or a SC (n = 5) based on current imaging criteria (presumed PG and SC). Of those, n = 3 PG and n = 3 SC lesions were biopsy proven. Patient demographics are summarized in Table 2.

3.2. Anatomical imaging

Overall mean lesion size was 32 mm. The lesion size of biopsy proven lesions was (mean ± SD) 29 ± 6 mm in PG and 35 ± 10 mm in SC. Lesion size including non-biopsied lesions was 32 ± 15 mm in PG, and 33 ± 8 mm in SC. Individual measurements are presented in Table 3.

4/6 PGs were located in the carotid space (3 biopsy proven), 1/6 at the jugular foramen (non-biopsy proven), and 1/6 in a jugulotympanic location (non-biopsy proven). 3/5 SC lesions were located in the carotid space (all histologically proven); 2/5 lesions were located in a neuroforamen (C1/2, C2/3) and biopsy was not performed due to typical signal characteristics including ‘dumbbell-shape’, lack of a ‘dural tail’, partial cystic degeneration, diffuse/rim enhancement in solid parts, and bone remodeling, highly suggestive for SCs according to current literature [24,25]. T2-hypointense flow-voids were diagnostic in 6/6 PGs and not present in any SC patient (‘Pepper Pattern’). Focal T1w-hyperintensities (‘Salt Pattern’) were present in 5/6 PGs and 0/5 SCs. On conventional postcontrast imaging, enhancement patterns in PGs varied from homogenous to heterogeneous with central hypo-/hyperenhancement, in SCs from homogenous to heterogeneous and peripheral enhancement patterns (Table 2). Splaying of ICA/ECA was present in 3/6 PGs. In 1/6 PGs, the internal jugular vein was obliterated. Anterior displacement of the IJV was present in 2/6 PGs and 1/5 SCs.

3.3. Medium resolution & semi-quantitative analysis

Visual and semi-quantitative analysis of the TICs from reconstruction with medium temporal resolution (Tres = 4.1 s) revealed two different enhancement patterns in the groups of biopsy-proven PG and SC lesions: A wash-out (type-III) pattern was found in 2/3 PGs, represented by a steep initial signal rise (T_peak 0.82 ± 0.02), high peak signal intensity (1.05 ± 0.14), high maximal Enhancement Ratio (2.31 ± 1.04) and maximal slope (0.84 ± 0.10). Of these three patients, one (case 5) demonstrated a type II/III enhancement curve (Fig. 3) with similar individual values to other biopsy proven PGs (T_peak 0.83; PH 1.21; ER_max 2.49; Slope_max 0.84). In contrast, a persistent (type-I) pattern was found in 3/3 SC, with no steep initial signal rise (T_peak 8.67 ± 0.33), lower peak signal intensity (0.36 ± 0.06), lower maximal Enhancement Ratio (0.74 ± 0.22), and maximal slope (0.29 ± 0.05). TIC and semi-quantitative parameters are presented in Fig. 3 and Table 5.

Fig. 3.

Signal-ratio time intensity curves (TICs) of 6 cases of biopsy proven H&N paragangliomas (PG, red) and schwannomas (SC, blue). Mean time-to-peak (TTP) enhancement biopsy proven [PG/SC] was at 18.8 s (SD 2.4) vs. 213.5 s (SD 11.5). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 5.

Time intensity curve (TIC) analysis in biopsy proven lesions by using reconstructions with Tres = 4.1 s.

	Paraganglioma (n = 3) [Mean ± SD]	Schwannoma (n = 3) [Mean ± SD]
TIC pattern	III/III/(II/III)	I/I/I
Tpeak (s)	0.82 ± 0.02	8.67 ± 0.33
PH	1.05 ± 0.14	0.36 ± 0.06
ERmax	2.31 ± 1.04	0.74 ± 0.22
Slopemax	0.84 ± 0.10	0.29 ± 0.29

TIC = Time intensity curve; Tpeak = Time to peak enhancement; PH = peak height; ERmax = maximal Enhancement Ratio; SLOPEmax = maximal slope. Metrics normalized to values obtained from ipsilateral CCA.

Visual analysis and semi-quantitative parameters of the TICs obtained with high Tres = 1.6 s and TICs obtained with a medium Tres = 4.1 s were comparable regarding the identification of the characteristic TIC patterns for PG and SC lesions (exemplary Fig. 2 and Fig. 3).

Fig. 2.

Perfusion maps (wash-in) and time concentration curves/automated qualitative curve-model fittings in a cervical paraganglioma (A) and schwannoma (B).

Exemplary dynamic contrast-enhanced series (Tres = 4.1 s) are shown for two right cervical biopsy proven lesions in Movies 1 & 2.

3.4. High resolution & postprocessing

In the semi-quantitative analysis of GRASP-MRI TICs with high time resolution (Tres = 1.6 s), PG lesions showed a high positive wash-in (23.04 ± 39.25) and a high peak enhancement intensity (7.08 ± 9.64) (type-III pattern), while SC lesions were characterized by low positive wash-in (0.09 ± 0.02) and a mean positive wash-out (0.08 ± 0.03) and low peak enhancement intensity (0.51 ± 0.14) (type-I pattern). Exemplary TICs for PGs and SCs are presented in Fig. 2. Semi-quantitative parameters (for Tres = 1.6 s) are presented in Table 4.

Table 4.

Syngo Tissue 4D qualitative perfusion analysis in biopsy proven lesions by using reconstructions with Tres = 1.6 s.

	Paraganglioma (n = 3) [Mean ± SD]	Schwannoma (n = 3) [Mean ± SD]
TIC pattern	III/III/(II/III)	I/I/I
Wash-In	23.04 ± 39.25	0.09 ± 0.02
Wash-Out	−0.35 ± 0.42	0.08 ± 0.03
TTP	0.60 ± 0.28	0.87 ± 0.03
PEI	7.08 ± 9.64	0.51 ± 0.14
iAUC	2.04 ± 2.92	0.04 ± 0.01

TIC = Time intensity curve. TTP = time-to-peak-enhancement; PEI = peak enhancement intensity; iAUC = area under the curve; SD = standard deviation.

3.5. Comparison between biopsy-proven and presumed PG & SC

Semi-quantitative parameters and visual analysis of GRASP TICs of presumed PG and SC lesions revealed the same characteristic TIC pattern as the biopsy-proven lesions (Fig. 4 and Table 3).

4. Discussion

In this study, we evaluated the potential role of GRASP MRI to differentiate between Head & Neck PGs and SCs. We demonstrated that GRASP MRI is feasible in the H&N region and provides reliable TICs that show specific enhancement patterns for PGs and SCs. To the best of our knowledge, this is the first study using GRASP MRI to differentiate between these tumor entities.

Visual and semi-quantitative analysis of GRASP-MRI TICs was, in this small retrospective study, sufficient to differentiate both tumor entities. PGs and SCs demonstrated characteristic TIC patterns, which is in accordance to previous studies including PGs [2–4,26] and SCs [21]. Semi-quantitative parameters based on TIC, especially T_peak and to a lesser degree PH, ER_max, and Slope_max, were characteristic for either PG or SC lesions and represent microstructural features as highly arterialized vascularization and intrinsic AV-shunting in PGs. Therefore, dynamic contrast-enhancement patterns are additional imaging features to improve non-invasive diagnostic specificity, especially in cases where conventional imaging might be ambiguous or where biopsy is not favorable. This point can be illustrated with case 4 (histologically proven PG), in which punctate hyperintensities on T1w images were not present (Table 2). Furthermore, our series contains two cases, in which biopsy was not performed due to anatomic location at the jugular foramen (cases 3 and 6; presumed PGs; Table 2).

Theoretically, the higher temporal resolution postprocessing (1.6 s) should be superior to the lower (4.1 s), as this acts as a low-pass filter potentially missing subtle oscillations below the minimal frequency capturable with a 4.1 s temporal resolution. In our study however, voxel-based perfusion mapping and qualitative modeling of reconstructions with high temporal resolution was not superior in delineating typical PG perfusion characteristics; even the reconstructions with lowest temporal resolution of 4.1 s resulted in distinct TICs, with all but one PG presenting with characteristic Type III-washout curves. Only one biopsy-proven PG case (No. 5) presented with a Type II/III curve. However, the enhancement curve even in this case showed a clear peak with a short but definite decrease in intensity that none of the SCs showed. The fact that this lesion presented with a Type II/III curve remains debatable. Possible causes are: the possible structural heterogeneity of PG expressed by a different extent of existing intrinsic AV shunts, patient-specific factors such as an extended cycle time, a bolus-related cause in case of too low contrast agent flow as well as a too low temporal resolution of the reconstruction. Regarding the latter point it can be said that even in the increased time resolution (1.6 s) this lesion showed a type II/III curve, which makes this argument seem unlikely. The fact that this lesion shows semi-quantitative parameters that fit well with other biopsied and suspected PGs suggests that in addition to the visual qualitative classification of TICs, the calculation of normalized semi-quantitative parameters is useful and necessary.

Limitations of our study include potential of selection bias due to retrospective data analysis and because histologic correlation was available in only 6/11 cases. However, these numbers reflect clinical reality, as biopsy is waived under certain circumstances. Reasons are multifold. Biopsy may be refused due to classic conventional imaging features or typical lesion location. In our cohort, biopsy was not performed in 2/5 SC patients due to diagnostic shape and location (dumbbell-shaped lesion within a cervical neuroforamen). Biopsy was waived in 3/6 PG lesions, which showed characteristic T2-hypointense flow voids and consistent presence of ‘Pepper (and Salt)’ pattern.

In our case series, biopsy was performed in 3/4 carotid space lesions but none of the lesions being located at the jugular foramen and the jugulotympanic location, which may also be explained by the fact that jugular/jugulotympanic lesions are anatomically/surgically less accessible and may be accompanied by higher periprocedural complication rates, including damage to adjacent cranial nerves and bleeding complications involving closely adjacent arteriovenous structures, which has been demonstrated in a large series of PG tumor resections [27]. A biopsy to improve diagnostic confidence in neuroforaminal lesions is often not performed as it can potentially be accompanied by neuronal damage or perineural/intraspinal hematoma. These factors underline the need for highly accurate diagnostic imaging surrogates, especially in anatomically less accessible “difficult to biopsy” lesions.

The fact that biopsies were not performed in all of the cases presented is a major limitation of our study, as we also included 5 cases of presumed PGs and SCs. However, the results were clearly presented with regard to existing and non-existing biopsies. Nevertheless, the question arises as to why patients without an existing biopsy were included. The presence of characteristic features in native sequences is sufficiently described in the literature [28,29]. Nevertheless, these studies are mainly about case series in which the interrater agreements regarding the existing feature are not consequently recorded. It is therefore still under discussion whether a biopsy must be performed in every case. Since the diagnosis based on conventional imaging criteria depends strongly on the experience of the respective reader, diagnosis based on these criteria is possible in the majority of cases. Therefore, a biopsy may pose an unnecessary risk for the patient, which holds true for a number of patients in our series. Nevertheless, especially a study on GRASP in parotid lesions has shown that this technique generates objectifiable contrast-enhancement curves that greatly increase diagnostic confidence in unknown lesions, independent of the presence of conventional features which also depend on lesion size, image quality and slice thickness. It is not supported by data so far that DCE MRI is necessary in every suspected case of PG or SC, but we assume that there is a high potential to increase the diagnostic confidence, especially for less experienced readers, using GRASP MRI. Unclear cases exist and due to the fact that contrast media is given for all head and neck tumor MRI protocols, a DCE perfusion sequence during contrast injection does not expose the patient to any additional MRI contrast media and most likely increase the diagnostic accuracy.

Furthermore, our study is limited by the fact that imaging studies were performed using three different scanner models with two field strengths (same vendor), as inter-scanner variability of DCE-MRI studies is a well-known phenomenon [30,31]. However, the small sample size did not allow drawing conclusions on the inter-scanner variability of the semi-quantitative parameters in our cohort.

We have used IV catheters with a relatively small inner diameter (20G). This limits the injection rate in our patient cohort to 2 ml/s, which theoretically has a negative effect on the bolus shape. Despite this fact, the derived TICs appear to be relatively reliable, however, a higher injection rate may be advantageous.

Current data on DCE MRI in PGs and SCs are characterized by predominant usage of 2D techniques, small sample sizes, and retrospective nature. GRASP MRI-based tissue perfusion analysis provides a dynamic 3D technique with high spatiotemporal resolution and the ability to retrospectively reconstruct datasets with varying temporal resolutions. This feature results from the radial k-space sampling pattern used in GRASP, which at the same is highly insensitive to patient motion and blood flow in vessels. High spatial anatomical resolution and motion robustness represent significant advances compared to established DCE-MRI techniques. High spatial resolution by generating reconstructions of 1.5 mm isotropic with high temporal resolution may be superior to previously published 2D DCE-MRI techniques concerning lesion detection, characterization, and follow-up especially in smaller, multifocal, or metastatic lesions, which need to be investigated in larger cohort studies. These advances may also facilitate improved characterization of postoperative and post-embolization lesions by identifying necrotic parts and/or adjacent scar tissue, which may be difficult to differentiate from residual tumor on post-contrast series. Beyond that, high spatial resolution, non-dynamic diagnostic postcontrast series can be provided from the same sequence, which may also shorten diagnostic exam protocols in a ‘one-stop shop’ manner.

5. Conclusion

DCE MRI and, especially, GRASP MRI can be used to improve accuracy and sensitivity in diagnosis of cervical PGs and SCs. Simple time-intensity curve analysis with a temporal resolution of 4 s, in cases of doubt under additional consideration of normalized semi-quantitative parameters, seems sufficient to differentiate between both entities.