Characterization and evaluation of a flexible MRI receive coil array for radiation therapy MR treatment planning using highly decoupled RF circuits

The growth in the use of magnetic resonance imaging (MRI) data for radiation therapy (RT) treatment planning has been facilitated by scanner hardware and software advances that have enabled RT patients to be imaged in treatment position while providing morphologic and functional assessment of tumor volumes and surrounding normal tissues. Despite these advances, manufacturers have been slow to develop radiofrequency (RF) coils that closely follow the contour of a RT patient undergoing MR imaging. Instead, relatively large form surface coil arrays have been adapted from diagnostic imaging. These arrays can be challenging to place on, and in general do not conform to the patient’s body habitus, resulting in sub optimal image quality. The purpose of this study is to report on the characterization of a new flexible and highly decoupled RF coil for use in MR imaging of RT patients. Coil performance was evaluated by performing signal-to-noise ratio (SNR) and noise correlation measurements using two coil (SNR) and four coil (noise correlation) element combinations as a function of coil overlap distance and comparing these values to those obtained using conventional coil elements. In vivo testing was performed in both normal volunteers and patients using a four and 16 element RF coil. Phantom experiments demonstrate the highly decoupled nature of the new coil elements when compared to conventional RF coils, while in vivo testing demonstrate that these coils can be integrated into extremely flexible and form fitting substrates that follow the exact contour of the patient. The new coil design addresses limitations imposed by traditional surface coil arrays and have the potential to significantly impact MR imaging for both diagnostic and RT applications.

scanner (Fujita 2007). The detected MR signal is in the form of an electromotive force that is induced in the loop in accordance with Faraday's law of induction. The signal is then amplified and digitized.
Multichannel RF coils consist of multiple single coil elements that have been designed to provide uniform signal reception, with a smooth and predictable falloff as a function of distance from the coil surface across the dimensions of the composite coil. To achieve this, considerable effort is dedicated to the size and placement of individual coils as well as the amount of overlap between adjacent coils. Coil size and placement impact signal uniformity, particularly as a function of depth while coil overlap affects coupling between adjacent coil elements due to mutual inductance (Roemer et al 1990). This effect is minimized when the amount of overlap is some fraction of the coil dimension. For circular coils, the optimal theoretical overlap is equal to 0.25× the diameter of the coil (Roemer et al 1990). Due to these considerations, modern receive-only RF coils are relatively constrained both in terms of their design and rigidity, despite the development of so-called flexible surface coil arrays. The inclusion of tuning and blocking circuits close to each coil element further adds to their physical bulk, weight, and rigidity (Fujita 2007).
In 2017, a new flexible and low physical profile RF coil element design referred to as adaptive image receive (AIR) coil was described and characterized (Rossman et al 2017). The RF coil possesses unique electrical properties in which the effects of mutual inductance and capacitive coupling between coil elements are significantly reduced thereby allowing coils to be placed with higher coil density in configurations that could not be achieved using conventional RF coils. In addition, their high flexibility and low physical profile allows them to be integrated into materials such as blankets or sheets that can conform directly to individual patient contours, especially when the patient is placed into a desired radiation therapy (RT) treatment position. These new coil designs therefore hold the promise of addressing many of the limitations that conventional surface coil arrays have placed on the MR imaging of patients for RT treatment planning and simulation.
The purpose of this work was to characterize the performance of these new AIR coils by comparing them to conventional, copper based RF coil elements. Phantom studies were performed to compare the SNR of both coil types and to compare SNR maps as a function of coil overlap distance for two element coil combinations. Mutual inductance loop to loop coupling was quantified by calculating noise correlation matrices for four element coil combinations. Parallel imaging performance was assessed by calculating g-factor maps as a function of coil overlap distance. In vivo testing was performed using four and 16-element coils in both normal volunteers and a patient undergoing MR imaging for RT treatment planning of the cervical spine.

Methods
All imaging was performed on two 3 T MR scanners with bore diameters of 60 cm (Signa HDx, GE Healthcare, Waukesha WI) and 70 cm (750W, GE Healthcare, Waukesha WI). Data analysis was performed using an in-house software application (MRView, Mayo Clinic and Foundation).

Multichannel RF coils description
All multichannel AIR coils were constructed using multiple identical single loop coils. The diameter of each coil element was equal to 110 mm and consisted of a malleable conductor with an approximate diameter of 0.6 mm resulting in a flexible low profile form factor loop that can conform precisely to complex and irregular surface contours. Figure 1(a) is a schematic of the AIR coil design that includes the resonant loop consisting of four linked resonators. For a given coil, the number of resonators is determined by its circumference (as in the case of a circular loop).The loop is created from a flexible link resonator structure using a process that is proprietary to the MR scanner manufacturer (GE Healthcare, Waukesha, WI) with the length of each resonator being no greater than 1/10th of the wavelength of the resonant RF field (Veenstra and Long 2008). This ensures that the resonator does not behave as a transmission line but rather a lumped element circuit thereby ensuring a uniform charge distribution and internally confined irrotational electric fields within the resonator (Sutherland 1999). The ensemble of resonators creates a zero reactance, low loss and low interaction loop that can be modeled electrically as a lumped circuit but without any physical lumped components providing high flexibility and durability. The resonant loop is connected to a printed circuit board module (dashed box in figure 1(a)) of dimensions 26.5 mm × 26.5 mm × 10 mm. The module provides impedance matching, signal amplification, and an RF choke and replaces conventional coil circuitry including the preamplifier, decoupler, matching circuit, and baluns. The preamplifier is low noise and is tolerant of a wide range of loop loading conditions. The combination of the conductor and electronics module provides a complete imaging coil element which exhibits lowered interaction to adjacent elements, extreme flexibility, and high performance. A single triaxial cable is used to connect the coil loop to the MR scanner so as to receive the MR signal. While a fixed diameter of 110 mm was used for all coil elements in this work, any diameter loop can be fabricated and constructed.
For the purposes of comparison a second multichannel RF coil set was constructed using conventional copper based coil elements of equal diameter as shown in figure 1(b). Individual elements consisted of a single 110 mm diameter copper loop and a pair of high Q capacitors as shown. The coil element was tuned to the resonant frequency of 127.92 MHz using a network analyzer (Agilent Technologies Japan, Hyogo, Japan). Multichannel coils were created by combining multiple elements of the single loop design.  performance A series of experiments was performed to evaluate the SNR of a single and double loop AIR coil configuration and to compare these values to those obtained from equal diameter conventional RF coil(s). For the double coil, both coil types included overlaps of 0 mm, 24 mm, 30 mm and 50 mm, with the 24 mm overlap being the nominal overlap distance for 110 mm diameter coils. This value is slightly less than the value of 27.5 mm estimated by Roemer et al (1990) and was obtained by minimizing the mutual inductance experimentally using a network analyzer (Agilent Technologies Japan, Hyogo, Japan).

Coil characterization
All SNR measurements were conducted by placing the RF coils on top of a non-loading rectangular phantom (Dielectric Corporation, USA) of dimensions 380 mm × 155 mm × 155 mm filled with a dimethyl siliconegadolinium solution. The phantom and coil combinations were imaged on the aforementioned 60 cm bore diameter 3 T MR scanner using a T 1 -weighted spin echo pulse sequence whose parameters are listed in table 1 under the row denoted Experiment Number 1. A 2D imaging plane bisecting the middle of the coil element(s) and orthogonal to the plane of the coil was prescribed. For each acquisition, an SNR image was calculated according to the National Electrical Manufacturers Association (NEMA) standard MR 9-2008 (National Electrical Manufacturers Association (NEMA) 2008) by dividing the entire image by the standard deviation of a region of interest of 25 × 25 pixels in the background (i.e. air) region of the image. The SNR maps were then normalized by dividing the SNR images by the maximum SNR value from the AIR and conventional coil data.
For the paired coil SNR maps, a sub region equal to a width of 385 mm and length of 80 mm with the width being parallel to the coil surface and length being perpendicular was chosen. Histograms of the normalized SNR values were calculated over the range of 0-1.0 with bin widths equal to 0.1. For each histogram the mean and standard deviation of the distribution were calculated. For SNR values within the region of interest, a one way analysis of variance and Tukey's range test were performed for each coil type and overlap distance using the Orig-inPro 9 data analysis and graphing package (OriginLab Corporation, Northhampton, MA USA).

Assessment of noise correlation
To estimate the amount of correlated noise between various coil elements, and by inference the amount of mutual inductance between them, a noise correlation matrix was calculated using the method described by Ohliger and Sodickson (2006). Noise correlation matrices were calculated for four channel multichannel coil combinations using both AIR and conventional coils. Both coil sets were assembled in a linear array and were placed on a torso shaped non electrically loading phantom (Dielectric Corporation, Milwaukee, WI) whose widest dimensions were 450 mm × 260 mm × 100 mm and filled with a dimethyl silicone-gadolinium solution as shown in figure 2. For each coil set five separate experiments at coil overlaps of 0 mm, 10 mm, 25 mm, 40 mm and 50 mm were performed. Data were obtained using the methodology described (Kellman and McVeigh 2005) in which the noise within each RF channel was sampled in the absence of RF and spatial encoding gradient fields using an ultra-fast 3D spoiled gradient echo imaging sequence whose parameters are listed in table 1 as Experiment Number 2. All imaging was performed on the previously described 60 cm bore diameter 3 T MR scanner.

Assessment of parallel imaging performance
Assessment of parallel imaging performance was determined by calculating so-called g-factor maps according to the methodology described by Pruessmann et al (1999). These maps were derived from the data acquired to estimate noise correlation matrices previously described. The g-factor maps utilize the full noise correlation matrices, with the noise sample data acquisition integrated into the scan pre-excitation.

In vivo testing 2.3.1. Evaluation of coil flexibility
To demonstrate the flexible and form fitting nature of the AIR coils, four coil elements were sewn around the sides and posterior of a Lycra ® balaclava. No elements were placed anteriorly due to the opening for the face. There was no preference given for overlaps between adjacent coil elements and the overlaps were equal to 5 mm (coils 3 & 4), 12 mm (coils 1 & 2) and 20 mm (coils 2 & 3). The balaclava design provided a flexible and exact form fitting coil design free of any rigid housings used in conventional coils. The coil combination was then placed on the head and neck of a male volunteer as shown in figure 3. Axial T 1 FLAIR and T 2 multi echo spin echo images were acquired on the aforementioned 3 T 60 cm MR scanner using the parameters listed in table 1 as described in Experiment Number 3. All imaging was performed under an Institutional Review Board approved research protocol with written informed consent being obtained prior to imaging.

RT planning coil
To assess the performance of the multichannel AIR coils for the purpose of MR imaging of patients undergoing RT treatment planning, a sixteen channel RF coil was constructed by combining individual AIR coil elements. Coils were grouped into four, four-channel paddles. For each paddle, the four coils were placed in a clover leaf configuration as shown in figure 4 with an overlap of 30 mm between adjacent coil elements and no overlap between any of the four elements at the center of the array configuration. The coils and resonant modules were sandwiched between two 3 mm thick sheets of polyester backed neoprene rubber (Foam Order Factory, San Pablo, CA) of dimensions 215 mm × 280 mm. A normal female volunteer and male patient undergoing MR imaging for RT treatment planning were imaged using the 16 channel AIR coil and a manufacturer provided combination 30 channel US Food and Drug Administration (FDA) approved RF coil set designed for RT treatment planning (RT Suite, GE Healthcare, Waukesha, WI). The FDA approved coil configuration consisted of a four coil combination that provided a total 30 individual coil elements and included a posterior array consisting of 8 coil elements encased within a rigid housing, two 3-element flex coils designed to be placed around the head and a 16 element rectangular flexible array coil. All imaging was performed under an Institutional Review Board approved research protocol with written informed consent being obtained prior to imaging. For both individuals two identical image data sets were obtained; the first with the FDA approved RT treatment planning RF coils and the second with the 16-channel AIR RF coil array. All imaging was performed on a 70 cm 3 T MR scanner previously described. Table 1 lists the scan parameters used to acquire the 3D data sets for the volunteer and treatment planning data and are listed as Experiment Number 4 and 5 respectively. For both data sets parallel imaging using the technique known as autocalibrating reconstruction for cartesian sampling (ARC) (Brau et al 2008) was used with acceleration factor of 2 in both  phase and slab encoding directions resulting in a total acceleration factor of 4. For the volunteer study a sagittal T 2 -weighted non fat saturated acquisition was acquired while a sagittal T 1 -weighted 3 point iterative Dixonbased fat saturation (Skinner andGlover 1997, Reeder et al 2004) acquisition was used for the patient study.
The cervical spines of both the volunteer and RT patient were imaged using the two coil configurations shown in figures 5(a) and (b). Both subjects were first positioned on top of a MR compatible indexed flat table top insert (CIVCO Radiotherapy, Orange City, IA, USA). For the AIR coil combination (figure 5(a)) a single four element paddle was placed below the cervical spine. Two four element paddles where placed against the left and right side of the subject's head while the final four element paddle was placed on the anterior chest at the level of the clavicles. Individual paddles were held in place using thin Velcro straps. Figure 5(b) shows the configuration used when imaging with the RT Suite and involved placing the two 3 element coils directly on top of the face of the subject, the 16 element flexible coil around the sides of the patient's head with the 8 element posterior array inserted into the table of the MR scanner below the flat table insert. Figure 6 shows SNR plots as a function of perpendicular distance from the coil surface for the single coil (figure 6(a)) and 24 mm overlap double coil (figure 6(b)) configurations for both coil types respectively. For the single coil the profile was taken at the center of the coil while for the double coil comparison profiles were taken at the middle of the overlap between the two coils. In the single coil comparison (figure 6(a)) The AIR coils provided improved SNR compared to the conventional coil up to a depth of approximately 25 mm. For the two coil comparison (figure 6(b)), the conventional coil SNR was 12% higher at the surface of the coil. However, at distances beyond one millimeter from the surface of the coil both coils exhibited similar SNR values as a function of depth. Figure 7 shows the SNR maps for the two element coil configuration as a function of overlap distance for both the conventional and AIR coils. The gray scale image at the top of the figure shows a representative magnitude image from which the SNR maps were derived. The yellow dashed line shows the region from which the eight SNR maps were obtained. All maps have been normalized to the maximum SNR value obtained from all data which was equal to the AIR coil SNR map at an overlap of zero millimeters. Coil overlap is identified by the red 'H' symbol above each map. Inductive coupling between adjacent coil elements due non-ideal coil overlap is seen in the non-uniformity of the coil SNR maps, particularly close to the surface of the coil. The conventional coil maps exhibit greater non-uniformity compared to the AIR coils. At the optimal overlap distance of 24 mm, both coil types exhibit similar SNR maps. However, at distances of 0 mm and 30 mm the AIR coil SNR maps are visually closer to the 24 mm overlap map compared to the conventional coils. At a maximum overlap of 50 mm, the SNR maps of the two coil types are approximately equivalent, demonstrating that at some distance, both coils will suffer from degraded SNR due to mutual inductance between adjacent coil elements. Of note, the SNR plots shown in the figure have been quantized into 7 individual bins with equal bin sizes of 0.1 from 0 to 0.6. The final bin size is equal to 0.4 and includes the SNR range from 0.6 to 1.0. Table 2 lists the mean and standard deviation of the histogram of SNR values when binned in 0.1 increments starting at a value of 0.0 while table 3 lists the results of performing Tukey's range test for two coil combinations of conventional and AIR coils versus coil overlap distance. For each comparison an alpha value (significance level) of 0.05 was used. In table 3 the first column lists the coil comparison being tested where C and A denote the  conventional and AIR coils respectively. The numerical value denotes the overlap distance. For example, A_24: C_30 tests the SNR values of the AIR coil with an overlap of 24 mm compared to the conventional coil combination at an overlap of 30 mm. The second column describes the difference in means of the two SNR distributions. A positive difference denotes that the first coil combination's mean was greater than the second coil combination. A negative value denotes the opposite. The results of this statistical test are also shown in figure 8 for the various coil combinations as a function of overlap distance. Statistically significant differences in means between various coil combinations are shown in red while non-significant differences in means are shown in black. A total of six combinations had mean differences greater than 0.02. The difference in mean SNR values for the coil at an overlap of 50 mm and conventional coil at overlaps of 0 mm, 24 mm and 30 mm were negative and greater than 0.02 indicating that the 50 mm overlap distance provided the lowest SNR compared to the others and that these differences were statistically significant. For positive differences greater than 0.02 the AIR coil mean at overlap distances of 0 mm, 24 mm and 30 mm was statistically significantly greater than the conventional coil combination at an overlap of 50 mm. Figure 9 shows the noise correlation matrices for the four coil linear combination of the conventional (top row) and AIR coil (bottom row) RF coils. The diagonal elements of the matrix quantify the total noise power of a given element while the off-diagonal elements describe noise correlations between multiple elements (Ohliger and Sodickson 2006). For each matrix, the normalized correlation values are shown. These values were normalized to a maximum value of 1.0 by taking the absolute value, and then dividing individual correlation values by the largest value within the matrix. Given these characteristics, it is observed that the amount of correlated (off-diagonal) noise is less in the AIR coil elements compared to the conventional coil elements. The off-diagonal correlation values of the AIR coil matrix are consistently near zero for all overlaps indicating that the full SNR  benefit of all coil elements is realized independent of overlap distance. At an overlap of 25 mm the conventional coil combination showed almost zero non diagonal elements indicating, again at the optimal overlap distance the SNR of the coil combination is maximized. However, as overlap distance deviates from the 25 mm value, nondiagonal values increase denoting coupling between coil elements. Overall, the AIR coil combination showed near-zero off-diagonal values for all overlaps, demonstrating an SNR advantage compared to conventional coils independent of overlap distance. Table 4 lists both the average and standard deviation of the g-factor values acquired as a function of overlap distance for the AIR and conventional four coil combinations while figure 10 is a plot of the average values alone versus overlap distance for both coil types. As expected, at the optimal overlap distance of 25 mm, the coils exhibit similar performance characteristics with average g-factor values of 1.084 and 1.090 for the conventional and AIR coils respectively. At overlap distances less or greater than the optimal value for this coil diameter the parallel imaging performance of the conventional coil is degraded as seen by the increasing average g-factor value which degrades as overlap distance increases or decreases from the optimal. In contrast, the AIR coil exhibits almost no degradation in parallel imaging performance as a function of overlap distance with average g-factor values not exceeding a maximum of 1.1. Figure 11 shows axial images superior to the orbits at the level of the operculum obtained from the normal male volunteer scanned with the balaclava coil shown in figure 3. Despite the fact that the AIR coil is relatively loosely affixed to the balaclava-no more than four stiches per coil element and clear tape were used-the coils conform closely to the subject's anatomy and provide high quality MR images for both the T 1 FLAIR ( figure 11(a)) and T 2 spin echo ( figure 11(b)) acquisitions. Note that, as expected, there is signal drop off due to the absence of any coil elements covering the anterior anatomy. However, despite the lack of coil coverage, anterior anatomical structures including the right frontal gyrus (arrow) can be easily seen. Anatomical structures posterior to frontal gyrus are clearly visualized with the four coil combination and despite the variable overlaps between adjacent coil elements, the images display a high degree of image uniformity. It is important to note that no surface coil Table 4. Average and standard deviation g-factor values as a function of coil overlap for both four coil AIR and conventional RF coil element combinations as a function of coil overlap distance. For both coil types a total of four coils were used.  10. Average parallel imaging g-factors as a function of coil overlap distance for the four coil AIR and conventional coil combinations. AIR coils show relatively no change as a function of overlap distance compared to the conventional coil combinations which exhibit degradation in coil performance (increasing g-factor) away from the optimal overlap distance of 25 mm. Parallel imaging performance decreases with increasing g-value with a value of 1.0 representing optimal performance. intensity correction algorithms were applied to the images, and they represent the native images acquired from the 3.0 T MR scanner. Figure 12 shows sagittal images of the cervical spine obtained from a normal female volunteer using both the 16 channel AIR coils ( figure 12(a)) and 30 channel RT Suite ( figure 12(b)). Both images are a single sagittal slice obtained from the 3D T 2 -weighted multi echo spin echo sagittal acquisition described in table 1 (Experiment Number 4). Both images have equal window and level values and demonstrate that, despite having approximately half of the number of RF coil elements, the AIR coil design provided qualitatively similar image quality when compared to the 30 channel FDA approved coil. Signal intensity variations across the field of view of both data sets are the result of placement of individual coils within each coil set. Subtle differences in slice location between figures 12(a) and (b) are due to the physical repositioning of the subject that occurred when exchanging coil sets. However, as described in table 1, identical imaging parameters were used to acquire the data represented in these figures. Figure 13 shows a single sagittal slice obtained from the 3D fat saturated T 1 -weighted acquisition described in table 1 (Experiment Number 5). Cervical spine imaging was performed for the purpose of external beam irradiation treatment planning of metastatic lesions in the cervical spine arising from adenocarcinoma of the prostate in a 78 year old male. Figure 13(a) shows the result of using the 16 channel AIR coil while 13(b) shows that from the 30 channel FDA approved coil. The arrow identifies susceptibility induced artifacts arising from the placement of spine fixation hardware. Relative differences in signal intensity between the two image data sets across the imaging field of view are due to the placement of the individual RF coils.

Discussion
Historically, the design and construction of multi-channel RF receive-only coils has been constrained by the need for a fixed physical geometry of individual coil elements, most notably the overlap amongst nearest neighbors (Roemer et al 1990). In addition, coil manufacturers have constructed either rigid or semi rigid housings to prevent damage and electrical breakdown of individual coil elements due to physical stress and distortion when placed around a patient. As a result, these coils have been ill-suited for RT treatment planning in which patients are imaged in non-neutral treatment positions which cannot be altered. The use of immobilization devices for RT treatments further complicates coil placement and can decrease image quality by requiring that coils be placed outside the device, further from the patient. The use of the linked resonator and interfacing electronics module combination (i.e. AIR coil) has the potential to address these limitations by relaxing the requirement on coil overlap distance, creating opportunities to increase or decrease coil density and allowing for overlap between adjacent coil arrays. The thin conductor construction of each coil element makes them flexible and durable while the miniaturization of the coil resonant circuitry decreases their overall form factor thereby facilitating fabrication of coils that can closely conform to the complex patient surfaces. Phantom and in vivo testing presented in this work provides quantitative evidence to support these claims.
Previously, Corea et al (2016Corea et al ( , 2017 have described the use of screen printing techniques involving the deposition of conductive silver ink onto the opposite surface of a plastic substrate to create a flexible and thin RF coil elements. The sandwiching of the plastic substrate between the two conducting layers provides the capacitance necessary for tuning and matching circuitry while the thin nature of both the plastic and metal deposition create a low form factor and flexible RF coil. The AIR coil design reported in this work is fundamentally different from this approach for several reasons. First, the Corea design is electrically equivalent to the lumped circuit design used in existing RF coils as shown in figure 1(b). As such, the same limitations in terms of mutual inductance between coil elements exists thereby imposing the restriction on a predetermined coil overlap distance for multi element RF coils. Second, while screen printing of the circuit onto a thin plastic film provides a flexible substrate upon which to create and RF coil, the geometry of both individual coil loops and the overall coil remain unchanged. Unlike the balaclava coil shown in figure 3 in which the shape of individual coil elements can change based on the fitting of the fabric around an individual subject's head, the plastic substrate imposed a fixed geometry on the RF coil configuration. Finally, the increased decoupling between coil elements using the AIR coil technology creates opportunities to create extremely high coil density arrays that cannot be achieved with conventional coil designs, despite how thin or flexible these designs may be. Vasanawala et al (2017) have reported on the development of a 16 element 'blanket' AIR coil for pediatric imaging applications. The 16 element blanket can be used on its own or in combination with posterior coil elements from FDA approved coils from the MR scanner manufacturer (GE Healthcare, Waukesha, WI) including the pediatric posterior array, posterior cardiac array and the posterior array from a standard adult torso coil. As in this work, the authors reported similar performance characteristics of the AIR coil design compared to conventional RF coils. The authors also noted that the decoupling between coil elements as seen in the noise correlation matrix for the 16 element coil remained effectively unchanged when flexed in two dimensions. For conventional coils in which a change in shape will result in a shift in resonant frequency of the coil (Mehmann et al 2017) and result in changes in coil coupling, no change in the noise correlation matrix indicates the robustness of these elements to shape distortion. Taken together, the work described in this report and that by Vasanawala et al (2017) demonstrate the utility of this new technology and the potential application beyond the sub specialty imaging application of MR for RT treatment planning and pediatric imaging.
This work provides both quantitative and qualitative data illustrating the superior performance characteristics of AIR coils when compared to equivalent conventional copper based designs. Normalized SNR maps for two coil combinations (figure 7) illustrate the relative insensitivity of coil overlap distance for AIR coils compared to conventional ones. This is particularly evident in the region of the overlap between coils in which conventional coil SNR maps show regions of signal loss (0 mm overlap) and significant heterogeneity (overlap distances of 30 mm and greater). In contrast, the overall SNR profile for AIR coils remains relatively constant as a function of coil overlap. Table 3 provides quantitative assessment of SNR data for the two coil combination and identifies that, based on SNR values alone, for overlap distances of 0 mm and 50 mm the AIR coil mean was statistically significantly greater that the conventional coil mean. However, at overlap distances close to the optimal overlap distance for the 110 mm coil diameter, the differences in SNR means were not statistically significant (24 mm and 30 mm overlap). This is somewhat expected as the effects of mutual inductance will be minimal at these overlap distances and therefore the SNR profiles should be similar as seen in figure 6(b). The relative insensitivity to coil overlap distance is graphically illustrated in the noise correlation matrices shown in figure 9 and further supports the fact that at the optimal overlap distance, the matrices for the four coil combination for both coil types are similar. However, at distances other than the optimal overlap distance, coil-to-coil interactions become evident for the conventional coil design. The overall superiority of the new coil design is also demonstrated in the g-factor data derived from the four coil phantom experiments as seen in figure 10 and table 4. While parallel imaging performance is dependent upon overlap distance for the conventional coil design, the AIR coil design demonstrates almost no dependence on overlap distance. Finally, in vivo testing in both normal volunteers and patient data demonstrate that these coils can conform closely to an individual's anatomy and provide images comparable to commercially available RF coils with larger numbers of coil elements.
There are several limitations related to this work. First, the electrical properties of both coil types were not fully characterized in this report. Parameters including overall coil quality (Q) factor, self-inductance (L) and resistance (R) are not reported. However, these values for a two element RF coil were previously reported (Vasanawala et al 2017). As such, the authors felt that it would be redundant to reproduce the same data that had been previously reported. Secondly, the in vivo comparison of the 16 element AIR coil and the FDA approved RT coil suite was not optimal due to the disparity in the number of coil elements between the two coil types. As noted previously, the commercial coil configuration included almost twice as many RF coil elements compared to the AIR coil with the expected decrease in overall image uniformity as seen in figures 12 and 13. However, despite the decreased element count, the AIR coil images produced images with image quality subjectively comparable to those produced with the commercial coil combination. To address this discrepancy, future work will involve the development of a 32 element AIR coil design using thinner and more flexible materials similar to those used in the balaclava coil.

Conclusion
In conclusion, this work has demonstrated that new, flexible AIR coils can be used for MR imaging of RT patients and that they provide advantages not seen with conventional RF coils.