Magnetic Resonance Imaging 53 (2018) 164–172 Contents lists available at ScienceDirect Magnetic Resonance Imaging journal homepage: www.elsevier.com/locate/mri Original contribution Evaluation of stacked resonators to enhance the performance of a surface receive-only array for prostate MRI at 3 Tesla T ⁎ Jorge Chacon-Calderaa, , Alexander Fischera, Matthias Malzachera, Yannik Vettera, Mathias Davidsa, Martina Flöserb, Christopher Stumpfc, Lothar R. Schada a Computer Assisted Clinical Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany c Work Microwave, Holzkirchen, Germany b A R T I C LE I N FO A B S T R A C T Keywords: RF receive coils Coil arrays Stacked resonators Prostate MRI Prostate MRI is an important tool to diagnose and characterize cancer. High local sensitivity and good parallel imaging performance are of paramount importance for diagnostic quality and eﬃciency. The purpose of this work was to evaluate stacked resonators as part of a surface receiver array for prostate MRI at 3 Tesla. A base array of 6-channels consisting of a ﬂexible anterior and a rigid posterior part were built each with three loop coils. A pair of stacked resonators was added concentrically to the center loops (anterior and posterior) of the base array. The evaluated stacked resonators were butterﬂies, composites and dipoles which yielded a total of three 8-channel arrays. The arrays were compared using noise correlations and single-channel signal-to-noise ratio maps in a phantom. Combined signal-to-noise ratio maps and parallel imaging performances were measured and compared in vivo in 6 healthy volunteers. The results were compared to the base and a commercial array. The SNR values in the prostate yielded by all the arrays were not statistically diﬀerent using fully sampled k-space. However, signiﬁcant diﬀerences were found in the parallel imaging performance of the arrays. More speciﬁcally, up to 88% geometric factor reduction was found compared to the commercial array and up to 83% reduction compared to the base array using butterﬂy coils. Thus, signal-to-noise ratio improvements were observed with stacked resonators when using parallel imaging. The use of stacked elements, in particular butterﬂy coils, can improve the performance of a base array consisting solely of single loops when using parallel imaging. We expect prostate MRI at 3 Tesla to beneﬁt from using combinations of single loops and stacked resonators. 1. Introduction The high incidence of prostate cancer requires robust and reliable methods to diagnose and characterize the disease. MRI has long been recognized as an important diagnostic tool for this purpose. Furthermore, it has recently been demonstrated that MRI can also be used to rule out cases where prostate cancer is falsely suspected, thus avoiding the need for biopsies . The main pre-requisite for prostate MRI is: high local sensitivity in the prostate with eﬃcient acquisition times to image the larger abdominal cavity. In terms of sensitivity, the location of the prostate complicates the signal detection due to the coil sensitivity drop as the distance from the coil increases. When using surface coils, this distance is determined by the anatomical location of the prostate with respect to the surface of the body. An approach to overcome this limitation is the use of tailored endorectal coils which can be placed near the prostate using the rectal cavity . However, there exists controversy in the literature regarding the advantages of using this type of coils [3–7]. Moreover, they are uncomfortable for patients. For these reasons, arrays of surface coils are often preferred despite an arguable reduction in signal-to-noise ratio (SNR) in the prostate. The limit in SNR has an impact on image quality and it could compromise the diagnostic quality of the images. Therefore, it is important to maximize sensitivity of the surface arrays. Likewise, achieving eﬃcient scanning times, is of high clinical relevance. For this Abbreviations: SNR, Signal-to-noise ratio; G-factor, Geometric factor; REF, Reference/base; COM, Commercial; B3SL, Base array with a pair of butterﬂy coils; C3SL, Base array with a pair of composite coils; D3SL, Base array with a pair of dipole antennas; AP, Anterior-Posterior; LR, Left-Right; SD, Standard deviation; SENSE, Sensitivity encoding; FOV, Field-of-view; ROI, Region-of-interest; A, Anterior; P, Posterior; L, Left; R, Right; S, Stacked; C, Center; Dir, Direction; Acc, Acceleration ⁎ Corresponding author at: Computer Assisted Clinical Medicine, Medical Faculty Mannheim, Heidelberg University, Theodor-Kutzer-Ufer 1-3, 68167 Mannheim, Germany. E-mail address: email@example.com (J. Chacon-Caldera). https://doi.org/10.1016/j.mri.2018.07.010 Received 26 March 2018; Received in revised form 8 June 2018; Accepted 21 July 2018 0730-725X/ © 2018 Elsevier Inc. All rights reserved. Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. Fig. 1. Schematic design of the tested stacked resonators and their reception chain. Their position and dimensions compared to the center loop is shown (Top). The stacked arrays are shown in a representative design and representative setups for phantom, and in vivo measurements (Bottom). . This approach has been more explored as a standalone element  and in combination with single loop coils [31, 32]. Gains have been demonstrated in arrays using both novel approaches combined with single loop coils e.g. composite coils in simulations and measurements of a cuboid phantom  and dipole antennas in body imaging at 7 T . Moreover, using simulations the dipole has shown to have a similar performance as the single loop in the target depth of the prostate at a range of magnetic ﬁeld strengths . The aim of this work is to experimentally investigate three diﬀerent stacked elements in a simple setup with low element count. This was designed to facilitate the characterization of the elements and to test their performance as part of a base array composed of single loop coils. Moreover, a comparison to a commercial solution is provided to ultimately obtain gains in SNR when using parallel imaging in prostate MRI at 3 T. purpose, parallel imaging techniques are the most common approach used in clinical. However, parallel imaging techniques yield an intrinsic SNR penalty determined by the receiver array. This penalty is characterized in terms of the geometric factor (g-factor) and must also be considered in the design and evaluation of RF coil arrays . Standard approaches to the design of RF coil arrays rely on loop coils decoupled by overlap [9–12]. The design in that case can be focused on placing the maximum number of single loop coils that ﬁt an area determined by the application [13–15]. This is feasible due to the large number of available reception channels in modern MRI systems. For body imaging, arrays of up to 128 single loop coils have been built with limited SNR increase in the center of the body but improved parallel imaging performance . However, this approach entails several technical limitations such as a reduction in individual performance of the coils due to coupling with next nearest neighboring coils, higher contribution to noise from coil resistance and reduced penetration depth. Several works have outlined that the maximum sensitivity in the center of the body (calculated as a homogeneous cylinder) can be well approximated with a limited number of single loop coils following the calculations of ultimate intrinsic SNR [17, 18]. Nevertheless, it has been proposed that additional resonators could complement the B1-ﬁeld components of the single loop coils and contribute further towards the sensitivity of RF arrays. Therefore, a combination of these resonators with single loops could yield SNR values closer to the ultimate intrinsic SNR. The impact has been investigated theoretically by several authors in simulations and experimental setups [19, 20]. Three main types of resonators have been investigated: butterﬂy and composite coils as well as dipole antennas. In this work we refer to these resonators as stacked elements to simplify the coil/antenna terminology. These stacked elements also allow geometrical decoupling when placed concentrically with loop coils. This represents an advantage over single loop coils which can only be decoupled geometrically using controlled lateral placement. Butterﬂy coils have been traditionally used in combination with single loops in MRI for quadrature detection [21–26]. Composite coils are single loop coils placed orthogonally with respect to the standard single loop coils and they have been seldom investigated [27, 28]. Another recently introduced approach is the use of dipole antennas 2. Materials and methods All methods were performed according to the relevant guidelines and regulations. The in vivo prostate scans were approved by the local Institutional Review Board and acquired with prior written informed consent. 2.1.1. Arrays design An array composed of a ﬂexible anterior and a rigid posterior part was used for the base of our analysis and also for reference (REF array). Each part of the REF array was composed of three single loops decoupled by overlap. The dimensions of the three loops were 180 × 180 mm2 for the center loop and 120 × 180 mm2 for the side loops. This resulted in a total dimension of 360 × 180 mm2 for each part of the array. The dimensions of the center loop were experimentally optimized for the depth of the prostate. The side loops were reduced in size (x-axis) compared to the center loop to minimize coupling between anterior and posterior parts of the array. The dimensions were also chosen to achieve full loading of the loops to maintain sample dominated noise. Stacked elements were added at the center of the two 165 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. height and 340 mm in maximum width (rcylinder = 110 mm, Wcuboid = 120 mm). The phantom contained homogeneous material properties with the dielectric parameters: ε′ = 50.3, ε″ = 66.6 and σ = 0.46 S/m. Additionally, the measurements included in vivo scans of healthy volunteers using a FOV of 200 × 200 mm2. 6 healthy male volunteers aged 23 to 31 years (mean ± SD = 27.3 ± 3.1) participated in this study. The height and weight of the volunteers was 180.8 ± 4.6 cm and 75.5 ± 10.45 kg (mean ± SD), respectively. The performance evaluation was carried out in the following three stages: parts of the base array with a distance of approximately 2 mm from the center loop coils. Both, anterior and posterior parts were analyzed using the same type of stacked element. The three resultant 8-channel arrays tested were: with butterﬂies (B3SL), with composites (C3SL) and with dipoles (D3SL). In the design of the stacked elements, each butterﬂy had a distance of 3 mm between the lobes and a total dimension of 160 × 160 mm2 to achieve overlap decoupling with the side loop coils (see Fig. 1 – Butterﬂy). The composites were built with a length of 180 mm. However, the height of the two composites was diﬀerent given the limited space in the bore and the height of the housing. The anterior composite had a height of 70 mm and the posterior 40 mm (see Fig. 1 – Composite). For the dipole, a fractionated approach was taken . Namely, the dipoles were shortened using 2 inductors (air solenoids) which were inserted equidistantly in each of the two segments. Solenoids with an inductance of 237 ± 32 nF were chosen to tune the element to the resonance frequency. This resulted in four segments each with a dimension of 30 × 75 mm2 (see Fig. 1 – Dipole). 2.2. Performance in phantom Using phantom measurements, the individual-channel SNR yielded by the stacked elements and the center loops were analyzed as part of the arrays. This was done to investigate the sensitivity of the elements and their eﬀect on the performance of the center loop. For this, SNR maps were calculated and proﬁles were plotted along the y-axis in sagittal slices. Noise datasets were used to generate the correlation matrix of each array. These matrices were compared to scattering parameters (S-parameters). Flip angle maps were also plotted to assess the detuning of elements during RF transmission. These tests included measurements without receiver arrays for reference and with every tested array. For this, ﬁve scans using a gradient echo sequence with ﬂip angles of 0, 10, 30, 60 and 90° were acquired using TR = 5*T1 (2000 ms). To compute the ﬂip angle map, a sine function was ﬁtted to the signal of the images obtained with the diﬀerent ﬂip angles in each pixel. The sine of the actual ﬂip angle was then used to assess the percentage of signal expected considering the ﬂip angle errors. 2.1.2. Arrays construction The coils of the base array were built using copper tape of 3 mm in width and 0.1 mm in height. Similar to the composites, three ﬁxed capacitors were used for splitting each coil; one additional variable capacitor was used for the ﬁne tuning of the coil. Split capacitors were also used to connect and balance the coils. Fuses were added to all coils and stacked elements for the safety of our volunteers. Each coil and stacked element had its own receiver channel composed of a capacitive matching network, a cable trap, and a low input impedance and low noise preampliﬁer. An active detuning network was also incorporated in each coil and element. Given the larger size of the butterﬂy due to the extra lobe, an additional passive detuning circuit was also added for these stacked elements . In the case of the dipole, active detuning was achieved by using a diode in parallel with the solenoid which shorted the segments when forward biased. A fuse was also added in series with the inductor between the segments in each of the two conductors of the element. For all preampliﬁers, each output was individually connected to the scanner using a standard ODU cable with three cable traps. In terms of the arrays, the anterior part was made ﬂexible to adapt to the shape of the phantom and volunteers. This was achieved by employing three layers. In the ﬁrst layer, foam with 5 mm thickness was used for the comfort of the volunteers. Two layers of FR-4 material with 0.5 mm in thickness were added to support the coils. The posterior array was ﬁxed and embedded in a box made of acrylic glass tightly ﬁtting the patient bed. The top layer was 5 mm in thickness and made of polystyrene. Foam was additionally employed for comfort. Thus, the minimum distance between the nearest coil and the surface of a sample was ca. 10 mm for the anterior part and 15 mm for the posterior. A commercial solution (COM array) for prostate imaging was also used for reference in the in vivo measurements. This setup included 18 channels divided into 12 channels of a spine array and 6 channels of a ﬂexible body array (Siemens Healthineers, Erlangen, Germany). 2.3. Array performance (in vivo) The adaptive combination  of all the channels in the stacked, the REF and the COM arrays was respectively used to evaluate array sensitivity. SNR was calculated in the prostate using in vivo scans of 6 healthy volunteers with each of the stacked and reference arrays. A manual 2D selection of the prostate was determined to evaluate SNR and g-factors. Furthermore, g-factors were calculated using transversal slices with both, the anterior-posterior (AP) and left-right (LR) direction for undersampling. For this calculation, coil sensitivities were obtained using the eigenvalue approach . G-factor maps were calculated based on the SENSE formalism . Mean g-factor values were then measured in the prostate. After the quantitative evaluation of g-factors in the prostate, the g-factor maps were multiplied by a mask. The mask was obtained by thresholding magnitude images and the multiplication was used for qualitative evaluation only. Finally, the overall performance of each array at acceleration factors 2, 3 and 4 and both encoding directions was evaluated in terms of SNR. For this, mean SNR values in the prostate were calculated, following (8), as: SNRreduced = SNRfull/(g − factor∗√ (acceleration factor)). (1) Representative magnitude images for qualitative assessment were also reconstructed using SENSE with a sum-of-square combination for acceleration factors 2 and 3 in both encoding directions. ANOVA was used to test the statistical diﬀerences between the arrays in terms of SNR using fully sampled k-spaces and g-factors with acceleration factor 2. 2.1.3. MRI measurements The measurements were performed using a 3 T Siemens MAGNETOM Trio. A 2D turbo-spin-echo sequence was used for all acquisitions with the following parameters: TE/TR = 95/3500 ms, ﬂip angle = 137°, thickness = 3 mm, matrix size = 320 × 320, bandwidth = 200 Hz/px, turbo factor = 27, echo trains per slice = 12 and echo spacing = 10.6 ms. Per measurement, two scans were acquired using this protocol with full k-space sampling with the second being a noise scan (same parameters but with no excitation RF pulse). All SNR calculations were performed following the standard method presented by Kellman et al. . Stacked elements were investigated as part of the array in a ﬁrst instance using a phantom using a FOV of 300 × 300 mm2. The phantom was composed of a cuboid with two half cylinders on two sides. This structure was 500 mm in length, 220 mm in 3. Results 3.1. Performance (phantom) 3.1.1. Sensitivity In regards to the individual channels of the array, the center single loop coils showed higher SNR than any of the tested stacked elements in 166 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. Fig. 2. 2D sagittal views of single-channel SNR maps measured in the phantom (Top). The maps correspond to the center loop and the stacked element in each tested array with full channel count (8-channel B3SL, C3SL and D3SL, 6-channel REF see Subsection 2.1.1). The lines in the SNR maps indicate the position where the SNR plots were taken. Line plots illustrate the SNR variations of the tested stacked elements of each of the arrays (B3SL, C3SL and D3SL) and center loops including the reference array (REF) according to the depth (Bottom). The depth of interest was zoomed for improved visualization. the prescribed ﬂip was approximately 8° when the COM setup was in place. Of the stacked arrays, the dipole showed the maximum deviation ca. 3%. Moreover, the reduction in signal due to errors in ﬂip angle maps was < 1.1% for all tested arrays (see Fig. 4). the phantom at the depth of interest (100 mm). For instance, the minimum SNR value of a center loop was 9.1 yielded by the posterior one in the D3SL. The butterﬂies and the dipoles showed minimum SNR values of 5.7 and 7.1 respectively. The composites showed the lowest sensitivity at the depth of interest with a minimum SNR value of 4.4 and 5 (anterior and posterior, respectively). The diﬀerence in SNR between stacked elements and loop coils was more clearly observed in the anterior part of the arrays. Additionally, the SNR values yielded by the center loops were consistently higher for the B3SL compared to the rest of the arrays in both anterior and posterior arrays at the depth of interest. Compared to the loop in the REF array, the SNR values of the center loops in C3SL and D3SL were similar in the anterior array but lower in the posterior array (see Fig. 2). 3.2. Array performance (in vivo) 3.2.1. Sensitivity In our in vivo evaluation, the ANOVA test showed that the diﬀerences between all tested arrays were not signiﬁcant (P = 0.77). However, the mean SNR values of the stacked arrays were consistently higher than those of the REF array. The array that yielded the closest SNR to the REF was the C3SL which yielded a 2.7% higher mean SNR. The greatest diﬀerence was 11% yielded by the D3SL compared to the REF (mean SNRs: 9.43 and 8.53, respectively; see Fig. 5a). A greater variability in SNR values was found amongst volunteers than amongst arrays e.g. the maximum diﬀerence was 42% between volunteer 2 and 3 (mean SNRs: 10.72 and 5.51 respectively; see Fig. 5b). The highest variability in SNR values of a single array was measured in the B3SL (SD = 2.06). In contrast, the minimum variability was yielded by the COM array (SD = 1.21). Qualitative prostate images yielded by all the tested arrays can be observed in Fig. 5c–d. 3.1.2. Noise correlation Between stacked elements and center loops, the lowest correlation was found in the C3SL. The values were 5.6% for the anterior and 4.5% for the posterior part of the array. The noise correlations in the D3SL were 17.3%, and 30.4% (anterior and posterior). The highest overall correlation was found between the center loop and the butterﬂy in the posterior part of the array. This correlation value was 66.5%. However, in the anterior part of the array, the correlation between the single loop and the butterﬂy was considerably lower (28.9%). Likewise, the B3SL showed higher overall coupling between the rest of the coils compared to the D3SL and C3SL (see Fig. 3). The S-parameters showed higher coupling of the center loops with the adjacent loops (S21 ≈ −12 dB) than with any of the stacked arrays. Here, the maximum coupling was the center loop with the butterﬂy in the anterior array of the B3SL (S21 ≈ −19 dB) and the minimum was found similarly in the anterior composite and dipole with the center loop (S21 ≈ −34 dB both). All Sparameters from the tested arrays are summarized in Table 1. 3.2.2. Parallel imaging In this evaluation, statistical signiﬁcance was found in the diﬀerent g-factors yielded by the tested arrays (e.g. P = 0.009/P = 0.0031 AP/ LR both with acceleration factor 2). Moreover, the stacked arrays yielded g-factors in the prostate which were lower than the ones from the REF array for all tested undersampling directions and acceleration factors. The B3SL performed consistently better than the rest of the tested arrays aside from one single case (acceleration factor 2 in the AP direction) where its g-factor was only 0.2% higher than the one from the C3SL. For low accelerations, the performances of all the arrays were closer to each other than for higher acceleration factors. For instance, with an acceleration factor 2, the mean g-factors of the tested stacked 3.1.3. Flip angle maps For a prescribed 90° excitation, the actual ﬂip angle without coils was 94.4° at the region-of-interest (ROI). A maximum deviation from 167 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. Fig. 3. Noise correlations found in phantom for each tested array. The coils and elements in the anterior (A) and posterior (P) arrays were: left loop (L), right loop (R), stacked element (S) and center loop (C). The white square indicates the noise correlation between the center loop and the stacked element. The commercial array (COM) was composed of 6 channels of a surface body array (BO6) and 12 channels of the spine array (SP12). Table 1 Scattering parameters of all tested arrays. The coils and elements were: left loop (L), right loop (R), center loop (C) and stacked element (S). Anterior REF B3SL C3SL D3SL L R C L R C S L R C S L R C S Posterior L R C −39 −21 −38 −35 −34 −31 −13 −12 −42 −10 −13 −38 −32 −32 −35 −14 −12 −36 −35 −28 −37 −14 −12 −41 S −13 −9 −19 −42 −15 −15 −34 −33 −12 −13 −34 −40 L R C −35 −21 −36 −30 −20 −35 −15 −15 −38 −12 −15 −45 −30 −30 −32 −15 −15 −40 −32 −28 −35 −15 −16 −47 S −11 −9 −24 −32 −15 −15 −33 −35 −12 −12 −26 −44 arrays were within 6% diﬀerence (AP/LR: B3SL = 1.063/1.034, C3SL = 1.061/1.037, D3SL = 1.071/1.039, REF = 1.116/1.055 and COM = 1.092/1.076). However, for acceleration factor 3, the g-factor of the worst performing array in AP (REF, g-factor = 2.42) was 44% higher than the best performing one (B3SL, g-factor = 1.68). This trend continued for acceleration factor 4 where the REF yielded a g-factor 61% higher than the B3SL in AP and 83% in LR. Furthermore, in LR, the COM yielded the worst performance which peaked at an acceleration factor 4 where the g-factor was 88% higher than the one of the B3SL array. Between the other two stacked arrays, the C3SL performed better than the D3SL in AP, particularly for an acceleration factor 3 where the D3SL yielded a 1.11-fold higher mean g-factor. However, in LR, the D3SL outperformed the C3SL by a maximum factor of 1.06 in acceleration factor 4. This can also be appreciated in representative g-factor maps and mean g-factor plots (see Fig. 6). Mean g-factors in the prostate and their standard deviations for all tested arrays, acceleration factors and directions are summarized in Table 2. In the calculation of SNR at diﬀerent acceleration factors, the B3SL and the D3SL showed similar performance in both AP and LR directions Fig. 4. Flip angle maps for a prescribed 90° excitation and the signal expected due to the actual ﬂip angle in percentage (100% corresponds to optimal ﬂip angle). The tested setups were A: without receiver arrays, B: COM, C: REF, D: B3SL, E: C3SL, F: D: D3SL. for an acceleration factor 2 (SNR diﬀerence < 1%). However, the gains in SNR were clearly higher for the B3SL for acceleration factors 3 (SNR gains = 28 and 16%, AP and LR) and 4 (SNR gains = 24 and 30%, AP and LR). Compared to the REF array, the maximum SNR gain (yielded by the B3SL) was calculated to be larger than a factor of 2. The COM array was also consistently outperformed by the B3SL. The range of the gain in this case was from 2% (acceleration factor 2, AP) to up to 94% (acceleration factor 4, LR; see Fig. 7). Sample reconstructions using SENSE can be observed in Fig. 8. 168 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. Fig. 5. a) SNR performance of the arrays as measured in 6 volunteers. Mean and standard deviation (SD) values yielded by the arrays are displayed on the top. The mean SNR values are also marked on the plot (+). b) Bar plot of SNR values found in the prostate of each of the scanned volunteers with the tested arrays. c) Representative transversal images of the prostate acquired with all the tested arrays showing d) a zoom of the prostate. COM Fig. 6. Representative g-factor maps yielded by all the tested arrays are displayed for an acceleration factor of 3 (Left). Mean g-factors calculated in the prostate of the 6 volunteers using anterior-posterior (AP) and left-right (LR) undersampling for diﬀerent acceleration factors are also shown (Right). Table 2 Mean g-factor values in the prostate of our 6 volunteers for all tested arrays. Dira Acc.b B3SL APc 2 3 4 2 3 4 1.063 1.683 2.908 1.034 1.329 2.833 LRd a b c d C3SL ± ± ± ± ± ± 0.03 0.22 0.41 0.02 0.07 0.24 1.061 1.958 3.383 1.037 1.536 3.966 D3SL ± ± ± ± ± ± 0.03 0.22 0.41 0.01 0.08 0.45 1.072 2.176 3.659 1.039 1.556 3.731 Undersampling direction. Acceleration factor. Anterior-posterior. Left-Right. 169 REF ± ± ± ± ± ± 0.01 0.27 0.65 0.01 0.12 0.91 1.116 2.422 4.690 1.055 1.591 5.187 COM ± ± ± ± ± ± 0.03 0.34 0.72 0.02 0.14 1.00 1.092 1.986 4.001 1.076 1.815 5.321 ± ± ± ± ± ± 0.03 0.08 0.76 0.02 0.09 0.40 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. part of the array where it lays on the volunteer. Air solenoids were useful to shorten the dipole's length. The noise added by these inductors is given by the eﬀective series resistance  which is determined by the copper of the inductor and was negligible in our case (< 0.01 Ω). Our ﬁrst evaluation of sensitivity was carried in phantom experiments using individual channels. Here, all the stacked elements showed sensitivity at the depth of interest. This was observed in the SNR proﬁle plots which were clearly above the noise ﬂoor. This was expected from previous works [28, 30]. However, their sensitivities were found to be consistently lower than the one from the single loop coils. One diﬀerent dominant reason for this can be provided for each element. In the case of the butterﬂy, the penetration depth is reduced as the number of lobes increases . The composite is a single loop coil which is well-known to have its main B1-ﬁeld component normal to its plane . Lastly, the dipole is mainly a radiative element (curl-free) but at the tested frequency, the signal detection is based on near-ﬁeld induction (divergence-free currents dominate the contributions to the SNR ). In this work, only the center loop coil was combined with a stacked element. Stacked elements can be added to the rest of the single loops at the cost of increased construction complexity and coupling. Furthermore, considering the longer distance of the side loop coils to the prostate and the observed limited penetration depth of the stacked elements these additional stacked elements are not expected to have considerable sensitivity. The center loops showed variability with each of the stacked elements. One potential reason for this is coupling e.g. the highest coupling between stacked elements and center loops was found to be between the butterﬂy and the single loop in the anterior part of the B3SL using S-parameters. This could increase the mutual inductance and aﬀect the individual performance of the coils by adding extra loading. Other possible sources of variations are slight diﬀerences in tuning and matching, array positioning and variations in the position of the selected line plot. In terms of noise correlation, high values were found between the stacked elements and the single loops despite lower coil coupling than between the adjacent loops as observed in the S-parameters. This is because noise correlation is not only dependent on coil inductance (cross-talk between coils) but also on coil overlap as demonstrated by Roemer et al. . Considering the increased overlap, the noise correlation of the butterﬂy and the single loop coils in the B3SL was higher than between the other stacked elements and their corresponding center loop coils. On the other hand, the composite showed the lowest overall noise correlations and mutual inductances with the center loops. The dipoles Fig. 7. Mean SNR values in the prostate of the 6 volunteers calculated for diﬀerent acceleration factors are plotted for both undersampling directions anterior-posterior (AP) and left-right (LR). 4. Discussion In this work, we evaluated three stacked elements added to a base array of single loop coils. This evaluation was carried out in order to test, in the ﬁrst instance, whether SNR gains in the prostate could be obtained from a) detecting B1-ﬁeld components orthogonal to single loop coils, b) increasing coil density without decreasing individual coil size and c) from placing additional elements nearest to the prostate as possible. Their parallel imaging performance and thus, their SNR with parallel imaging was tested in the second instance. The stacked elements included in the base array were butterﬂy and composite coils as well as dipole antennas. In the design process of the arrays, diﬀerent constraints regarding the dimensions of the elements had to be taken into account. For the butterﬂy, the size of the lobes was determined by the coils adjacent to the center loop which had a ﬁxed position to achieve overlap decoupling with the center loop. Here, the x-axis dimensions of the butterﬂies were chosen to achieve overlap decoupling with the side coils. The height was the geometrical limitation imposed by the composite coils. This was due to the limited space inside of the bore and the housing of the posterior part of the array. The design of the dipole was not as restrictive but its length could not be increased without adding complexity to its positioning, especially in the anterior Fig. 8. Representative parallel imaging reconstructions of all arrays are shown. Images were reconstructed using SENSE with acceleration factors of 2 and 3 using both undersampling directions: anterior-posterior (AP) and left-right (LR). 170 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al. References demonstrated a similar behavior to that of the composites in terms of noise correlations with the center loops. However, higher noise correlations were observed between the dipoles and the adjacent loops. The impact of the higher noise correlations on SNR was minimized by using a matched ﬁlter coil combination  in which the noise correlation matrix is used to coherently weigh the coil sensitivities to maximize SNR. In terms of parallel imaging, it has been shown in previous works that the butterﬂy coils yields distinct phase information  which seems to outweigh the higher noise correlations. Moreover, it has been shown that noise correlations have limited impact on the parallel imaging reconstruction . In the ﬂip angle maps, some coupling in the periphery was found between the arrays and the transmission coil leading to ﬂip angle deviations. However, at the region-of-interest, the signal decrease due to ﬂip angle errors was negligible in all tested arrays. In the in vivo measurements, despite the lack of statistical diﬀerences, the SNR of the arrays that included stacked elements were consistently higher than the base array. In comparison to the COM array, the D3SL and B3SL showed similar sensitivity despite the higher channel count of the COM array. Unlike sensitivity for fully sampled k-spaces, the parallel imaging performance was signiﬁcantly diﬀerent for each array especially as the acceleration factor increased. In this evaluation, the B3SL showed the best performance. The improved performance using butterﬂy coils compared to a reference array of single loops is consistent with previous studies . Furthermore, in g-factors, the advantage of using stacked elements was found to be even higher than the choice between the types of element. For example, the g-factor diﬀerence between the B3SL with acceleration factor 3 and REF with acceleration factor 2 was less than the diﬀerence between B3SL and D3SL both with acceleration factor 2 (20% and 29%, respectively). The calculation of mean g-factors in the prostate allowed the evaluation of image quality degradation due to the tested arrays when using parallel imaging. Here, it was found that the SNR of the B3SL was between 12 and 15% higher than the one from the REF array for an acceleration factor 2. In comparison to the COM array, both the B3SL and the D3SL showed improved performances but the B3SL showed higher gains. The parallel imaging reconstructions were performed using standard methods; more advanced methods could improve the image quality further with the beneﬁt of the reduced g-factor found by adding the stacked elements.  Ahmed HU, El-Shater Bosaily A, Brown LC, Gabe R, Kaplan R, Parmar MK, et al. Diagnostic accuracy of multi-parametric MRI and TRUS biopsy in prostate cancer (PROMIS): a paired validating conﬁrmatory study. Lancet 2017;389(10071):815–22.  Erturk MA, Tian J, Van de Moortele PF, Adriany G, Metzger GJ. Development and evaluation of a multichannel endorectal RF coil for prostate MRI at 7T in combination with an external surface array. J Magn Reson Imaging 2016;43(6):1279–87.  Costa DN, Yuan Q, Xi Y, Rofsky NM, Lenkinski RE, Lotan Y, et al. Comparison of prostate cancer detection at 3-T MRI with and without an endorectal coil: a prospective, paired-patient study. Urol Oncol 2016;34(6):255.e7–255.e13.  Heijmink SWTPJ, Fütterer JJ, Hambrock T, Takahashi S, Scheenen TWJ, Huisman HJ, et al. Prostate cancer: body-array versus endorectal coil MR imaging at 3 T—comparison of image quality, localization, and staging performance. Radiology 2007;244(1):184–95.  Kim BS, Kim T-H, Kwon TG, Yoo ES. Comparison of pelvic phased-array versus endorectal coil magnetic resonance imaging at 3 Tesla for local staging of prostate cancer. Yonsei Med J 2012;53(3):550–6.  Barth BK, Cornelius A, Nanz D, Eberli D, Donati OF. Comparison of image quality and patient discomfort in prostate MRI: pelvic phased array coil vs. endorectal coil. Abdom Radiol (NY) 2016;41(11):2218–26.  Gawlitza J, Reiss-Zimmermann M, Thörmer G, Schaudinn A, Linder N, Garnov N, et al. Impact of the use of an endorectal coil for 3 T prostate MRI on image quality and cancer detection rate. Sci Rep 2017;7:40640.  Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42(5):952–62.  Wright SM, Wald LL. Theory and application of array coils in MR spectroscopy. NMR Biomed 1997;10(8):394–410.  Malzacher M, Kalayciyan R, Konstandin S, Haneder S, Schad LR. Sodium-23 MRI of whole spine at 3 Tesla using a 5-channel receive-only phased-array and a wholebody transmit resonator. Z Med Phys 2016;26(1):95–100.  Wiggins GC, Triantafyllou C, Potthast A, Reykowski A, Nittka M, Wald LL. 32channel 3 Tesla receive-only phased-array head coil with soccer-ball element geometry. Magn Reson Med 2006;56(1):216–23.  Edelstein WA, Glover GH, Hardy CJ, Redington RW. The intrinsic signal-to-noise ratio in NMR imaging. Magn Reson Med 1986;3(4):604–18.  Schmitt M, Potthast A, Sosnovik DE, Polimeni JR, Wiggins GC, Triantafyllou C, et al. A 128-channel receive-only cardiac coil for highly accelerated cardiac MRI at 3 Tesla. Magn Reson Med 2008;59(6):1431–9.  Wiggins GC, Polimeni JR, Potthast A, Schmitt M, Alagappan V, Wald LL. 96-channel receive-only head coil for 3 Tesla: design optimization and evaluation. Magn Reson Med 2009;62(3):754–62.  Keil B, Wald LL. Massively parallel MRI detector arrays. Massively parallel MRI detector arrays. J Magn Reson 2013;229:75–89.  Hardy CJ, Giaquinto RO, Piel JE, Rohling KW, Marinelli L, Blezek DJ, et al. 128channel body MRI with a ﬂexible high-density receiver-coil array. J Magn Reson Imaging 2008;28(5):1219–25.  Lattanzi R, Sodickson DK. Ideal current patterns yielding optimal signal-to-noise ratio and speciﬁc absorption rate in magnetic resonance imaging: computational methods and physical insights. Magn Reson Med 2012;68(1):286–304.  Schnell W, Renz W, Vester M, Ermert H. Ultimate signal-to-noise-ratio of surface and body antennas for magnetic resonance imaging. IEEE Trans Antennas Propag 2000;48(3):418–28.  Wang ZJ. Towards a complete coil array. Magn Reson Imaging 2008;26(9):1310–5.  Guerin B, Villena JF, Polimeridis AG, Adalsteinsson E, Daniel L, White JK, et al. The ultimate signal-to-noise ratio in realistic body models. Magn Reson Med 2016. https://doi.org/10.1002/mrm.26564.  Sapkota N, Thapa B, Lee Y, Kim T, Bisson EF, Shah LM, et al. Eight-channel decoupled array for cervical spinal cord imaging at 3 T: six-channel posterior and twochannel anterior array coil. Concepts Magn Reson Part B 2016;46B(2):90–9.  Ohliger MA, Greenman RL, Giaquinto R, McKenzie CA, Wiggins G, Sodickson DK. Concentric coil arrays for parallel MRI. Magn Reson Med 2005;54(5):1248–60.  Versluis MJ, Tsekos N, Smith NB, Webb AG. Simple RF design for human functional and morphological cardiac imaging at 7 tesla. J Magn Reson 2009;200(1):161–6.  Ouhlous M, Moelker A, Flick HJ, Wielopolski PA, de Weert TT, Pattynama PM, et al. Quadrature coil design for high-resolution carotid artery imaging scores better than a dual phased-array coil design with the same volume coverage. J Magn Reson Imaging 2007;25(5):1079–84.  Alfonsetti M, Clementi V, Iotti S, Placidi G, Lodi R, Barbiroli B, et al. Versatile coil design and positioning of transverse-ﬁeld RF surface coils for clinical 1.5-T MRI applications. Magn Reson Mater Phys 2005;18(2):69–75.  Kumar A, Bottomley PA. Optimized quadrature surface coil designs. Magn Reson Mater Phys 2008;21(1–2):41–52.  Wang ZJ. Improving SNR of RF coils using composite coil elements. NMR Biomed 2009;22(9):952–9.  Maunder A, Fallone BG, Daneshmand M, De Zanche N. Experimental veriﬁcation of SNR and parallel imaging improvements using composite arrays. NMR Biomed 2015;28(2):141–53.  Raaijmakers AJ, Ipek O, Klomp DW, Possanzini C, Harvey PR, Lagendijk JJ, et al. Design of a radiative surface coil array element at 7 T: the single-side adapted dipole antenna. Magn Reson Med 2011;66(5):1488–97.  Raaijmakers AJ, Luijten PR, van den Berg CA. Dipole antennas for ultrahigh-ﬁeld body imaging: a comparison with loop coils. NMR Biomed 2016;29:1122–30.  Shajan G, Mirkes C, Buckenmaier K, Hoﬀmann J, Pohmann R, Scheﬄer K. Three- 5. Conclusions In conclusion, our results demonstrated that the use of stacked elements can improve the performance of a base array consisting solely of single loops. Moreover, it was shown that the choice of stacked elements between dipoles and butterﬂies yields small diﬀerences in SNR but could have considerable impact on parallel imaging performance. The complexity of the design was negligible and the advantages were clear when using parallel imaging methods. Considering SNR with parallel imaging reconstructions, the B3SL showed the best performance overall. However, the other elements showed advantages in design and decoupling that could also be exploited. Moreover, novel methods to combine elements while minimizing the distance to the ROI could provide further advantages . These approaches must also be investigated and put into perspective compared to traditional designs. In any case, stacked elements could become a consideration in the design of coil arrays for prostate imaging at 3 T. Acknowledgments The authors wish to thank Andreas Bitz and Nicolas Behl for their assistance. This research did not receive any speciﬁc grant from funding agencies in the public, commercial, or not-for-proﬁt sectors. 171 Magnetic Resonance Imaging 53 (2018) 164–172 J. Chacon-Caldera et al.       layered radio frequency coil arrangement for sodium MRI of the human brain at 9.4 Tesla. Magn Reson Med 2016;75(2):906–16. Erturk MA, Raaijmakers AJ, Adriany G, Ugurbil K, Metzger GJ. A 16-channel combined loop-dipole transceiver array for 7 Tesla body MRI. Magn Reson Med 2017;77(2):884–94. Raaijmakers AJ, Italiaander M, Voogt IJ, Luijten PR, Hoogduin JM, Klomp DW, et al. The fractionated dipole antenna: a new antenna for body imaging at 7 Tesla. Magn Reson Med 2016;75(3):1366–74. Edelstein WA, Hardy CJ, Mueller OM. Electronic decoupling of surface-coil receivers for NMR imaging and spectroscopy. J Magn Reson 1986;67(1):156–61. Kellman P, McVeigh ER. Image reconstruction in SNR units: a general method for SNR measurement. Magn Reson Med 2005;54(6):1439–47. Walsh DO, Gmitro AF, Marcellin MW. Adaptive reconstruction of phased array MR imagery. Magn Reson Med 2000;43(5):682–90. Uecker M, Lai P, Murphy MJ, Virtue P, Elad M, Pauly JM, et al. ESPIRiT–an       172 eigenvalue approach to autocalibrating parallel MRI: where SENSE meets GRAPPA. Magn Reson Med 2014;71(3):990–1001. Kumar A, Edelstein WA, Bottomley PA. Noise ﬁgure limits for circular loop MR coils. Magn Reson Med 2009;61(5):1201–9. Lattanzi R, Wiggins GC, Zhang B, Duan Q, Brown R, Sodickson DK. Approaching ultimate intrinsic signal-to-noise ratio with loop and dipole antennas. Magn Reson Med 2018;79(3):1789–803. Roemer PB, Edelstein WA, Hayes CE, Souza SP, Mueller OM. The NMR phased array. Magn Reson Med 1990;16(2):192–225. Hajnal J, Larkman D, Herlihy D. An array that exploits phase for SENSE imaging. Proc Intl Sot Mag Reson Med 2000;8:1719. Ohliger MA, Ledden P, McKenzie CA, Sodickson DK. Eﬀects of inductive coupling on parallel MR image reconstructions. Magn Reson Med 2004;52(3):628–39. Chacon-Caldera J, Malzacher M, Schad LR. Partially orthogonal resonators for magnetic resonance imaging. Sci Rep 2017;7:42347.