Feasibility of 31P spectroscopic imaging at 7 T in lung carcinoma patients

Currently, it is difficult to predict effective therapy response to molecular therapies for the treatment of lung cancer based solely on anatomical images. 31P MR spectroscopic imaging could provide as a non‐invasive method to monitor potential biomarkers for early therapy evaluation, a necessity to improve personalized care and reduce cost. However, surface coils limit the imaging volume in conventional 31P MRSI. High‐energetic adiabatic RF pulses are required to achieve flip angle homogeneity but lead to high SAR. Birdcage coils permit use of conventional amplitude modulated pulses, even over large FOV, potentially decreasing overall SAR massively. Here, we investigate the feasibility of 3D 31P MRSI at 7 T in lung carcinoma patients using an integrated 31P birdcage body coil in combination with either a dual‐coil or a 16‐channel receiver. Simulations showed a maximum decrease in SNR per unit of time of 8% for flip angle deviations in short TR low flip‐angle excitation 3D CSI. The minimal SNR loss allowed for fast 3D CSI without time‐consuming calibration steps (>10:00 min.). 31P spectra from four lung carcinoma patients were acquired within 29:00 minutes and with high SNR using both receivers. The latter allowed discrimination of individual phosphodiesters, inorganic phosphate, phosphocreatine and ATP. The receiver array allowed for an increased FOV compared to the dual‐coil receiver. 3D 31P‐CSI were acquired successfully in four lung carcinoma patients using the integrated 31P body coil at ultra‐high field. The increased spectral resolution at 7 T allowed differentiation of multiple 31P metabolites related to phospholipid and energy metabolism. Simulations provide motivation to exclude 31P B1 calibrations, potentially decreasing total scan duration. Employing large receiver arrays improves the field of view allowing for full organ coverage. 31P MRSI is feasible in lung carcinoma patients and has potential as a non‐invasive method for monitoring personalized therapy response in lung tumors.

and has potential as a non-invasive method for monitoring personalized therapy response in lung tumors. KEYWORDS 31 P MR spectroscopic imaging, In vivo application, lung carcinoma, response monitoring, X-nuclei MRS

| INTRODUCTION
In recent years, many new molecular therapies, such as immunotherapy, have been introduced for the treatment of lung cancer 1 . Tumor cells generally use antigens to mask themselves from the immune system and immunotherapy exploits this mechanism by administering antibodies which specifically target tumor antigens. This labels the cell which allows it to be recognized by the own defense mechanisms of the body. The immune system responds by inhibiting or attacking the tumor cells, resulting in stalled tumor growth not necessarily accompanied by a decrease of tumor volume on imaging modalities 2 .
Currently, it is difficult to predict which patients show an effective response to immunotherapy based on anatomical images like computed tomography only. Although a promising new treatment strategy for non-small cell lung carcinoma, immunotherapy is expensive and severe drug side effects are observed accompanied by an apparent decrease in quality of life. Therefore, there is an unmet need for a non-invasive method that can be used to predict tumor metabolic response which is crucial for early therapy effect evaluation. By adjusting the therapy strategy accordingly, such a tool would allow for more personalized curative care with less side effects, and reduced costs.
A recent study in breast cancer showed that changes in the phospholipid metabolism in responsive tumors can be detected after a single chemotherapy session using 31-phosphorous ( 31 P) magnetic resonance spectroscopic imaging (MRSI) 3,4 . 31 P MRSI can detect the phospholipid and energy metabolites, which provides possibilities to monitor tissue metabolism non-invasively during treatment. Inorganic phosphate (Pi), phosphocreatine (PCr) and ATP (with the α-, βand γresonances) allow assessment of the energy metabolism and the phosphomonoesters (PME) and phosphodiesters (PDE) provide insight into the phospholipid metabolism 4-7 . Enhanced ratios of phosphocholine (PC) to glycerophosphocholine (GPC) and phosphoethanolamine (PE) to glycerophosphoethanolamine (GPE), are frequently observed in tumor tissue and correlated with proliferation 7-12 . Another study in breast cancer demonstrated the feasibility of the phospholipid metabolism as biomarker for therapy follow-up and additionally reported shortening of the transverse relaxation time of Pi as a biomarker 4,8,13 . As the physiological changes are present before any morphological changes have occurred, these metabolites, their ratios and individual MR properties are potential (bio-)markers for therapy response monitoring [14][15][16] .
However, the individual detection of 31P metabolites is hampered at lower magnetic field strengths (3 T and below) due to the restricted spectral bandwidth and the low detection sensitivity. By going to higher magnetic field strengths (e.g. 7 T and higher), the SNR and spectral resolution are intrinsically enhanced 17,18 . These properties have a tremendous advantage for the low abundant 31 P metabolites and even allow detection of the individual phosphomonoesters, (i.e. PE, PC) and diesters (i.e. GPE, GPC) 19 .
Unfortunately, the imaging volume in conventional 31 P MRSI is limited as small birdcage or surface coils are used 5,20 . Surface coils generally require the use of high-energetic adiabatic RF pulses to achieve flip angle homogeneity as inhomogeneous excitations lead to signal variation in the acquired spectra over the large field of view. Adiabatic RF pulses usually result in high specific absorption rates (SAR), leading to longer repetition times (TR), clinically impractical scan times for a single protocol and a limiting number of consecutive scans. Full spectroscopic coverage of large organs such as the lungs is therefore challenging due to inhomogeneous B 1 +/− fields and inhomogeneous excitation which increase with magnetic field strength.
In addition, MR imaging and spectroscopy are challenging near the lungs due to the presence of air, the relatively small amount of tissue and This allows the use of rectangular pulses, which decreases global and local SAR, creating opportunities for fast spectroscopic imaging methods. In addition, they demonstrated that this 31 P-body coil even allowed quantification of transverse relaxation times and the feasibility of obtaining high flip angle chemical shift imaging (CSI), over a large field of view. However, the use of this coil was revealed with a 30% inter-subject variation of the flip angle using a single power setting for multiple volunteers. This raises questions for the need for individual 31 P calibration, especially at low flip angles, as only the effective flip angle and not B 1 + -field homogeneity is affected. Low flip angle excitations accompanied with short repetition times (TR) can be used for fast 3D CSI. The optimal SNR per unit of time at lower flip angles is acquired when the Ernst angle (α E ) is used and any deviation from this flip angle result in additional T 1 weighting and a lower SNR per unit of time 24 . The effects of a 30% flip angle deviation to the SNR per unit of time and consequently the acquired spectra can be evaluated by simulations. Excluding B 1 calibrations can decrease the total scan duration by 10 minutes or more, subsequently increasing patient comfort or allowing for additional scans or additional sampled averages to improve SNR. 25 The primary aim of this study was to investigate the feasibility of 3D 31 P MR spectroscopic imaging at ultra-high field in combination with a 31 P whole-body birdcage coil in four lung carcinomas.
The effect of an uncalibrated excitation, that leads to a deviation from the Ernst angle (α E ) was assessed by simulating the SNR per unit of time for the α E and for α Ε with a ± 30% and ± 50% deviation over a TR/T 1 -ratios range of 10 −6 to 0.3. The latter is chosen with respect to a short TR of 60 ms and the longitudinal relaxation times (T 1 ) for 31 P metabolites of interest possibly ranging from 450 ms (α-ATP) to 7000 ms (GPE) 6 . The simulated spectroscopy signal shown in equation [1] was corrected for time differences by dividing with the square root of TR. The SNR per unit of time for all the calculated TR/T1-ratios were normalized to the maximum signal at α E .
2.2 | Materials 31 P MRSI was performed using an in-house designed 31 P whole body birdcage coil integrated in a 7 T MR system (Philips Healthcare, Best, Netherlands) 22,23 . The body coil, tuned at 120.6 MHz, was powered by a 25 kW amplifier (PID: 53-S26B-128, MKS Technologies, Shenzhen, Republic of China) resulting in a B 1 + field-magnitude of 15μT at the isocentre. Two 1H-TxRx/31P-Rx arrays were constructed for the experiments. Array 1 (A1) contained a 31 P dual-coil receiver (10 x 16 cm 2 , Figure 1A) and two fractionated 1 H dipole antennas (30 cm) used as transceivers, both driven in quadrature mode. Array 2 (A2) contained a 16-channel 31 P body array with eight integrated 1 H dipole antennas, shown in Figure 1B and C 26 .
Spectroscopic imaging data and anatomical proton images for localization were acquired in four patients using one of the two different setups.

| Patients & setup
Four stage III-IV non-small cell lung carcinoma patients (ages: 53-63 years; BMI: 17.7-29.5 Kg/m 2 ) were included in this feasibility study and signed informed consent prior to scanning. Two patients participated after their palliative chemo-and/or radiotherapy sessions and two patients FIGURE 1 A.) image of the 31 P dual-coil receiver from A1 with an apparent curvature to allow close contact with the body. B.) a view of one of the eight elements of the 31 P 16-channel receiver array from A2. Two 31 P receiver coils, overlapped to improve decoupling, are denoted by blue arrows and the 1 H meander dipole antenna for MR imaging is shown by the red arrow. C.) all eight elements of the 31 P 16-channel receiver array positioned around a plastic human mannequin representable for the in vivo setup for MR spectroscopic imaging of the upper torso targeting the lungs participated after the first immunotherapy cycle (see Table 1 for details). Patients were scanned in supine position. Scans of two patients were acquired with the 31 P dual coil Rx (A1) placed on the location closest to the tumor based on previously acquired clinical CT images for tumor localization. The other two patients were scanned with the 31 P Rx array (A2), that was wrapped around the upper part of their torso. The two separate dipole antennas in A1 are positioned on the side and the top of the lung of interest. Maximum tumor dimension ranged from 25 mm to 75 mm and other clinical details per patient are shown in Table 1.

| MR data acquisition
No B 0 shimming was performed nor was the 31 P B 1 + calibration. Phosphorus ( 31 P) spectra were acquired using a 3D 31 P acquisition weighted CSI protocol including elliptical k-space sampling. Excitation was performed using rectangular RF pulses only and the carrier frequency was set to PCr.
The isotropic resolution ranged from 20 to 30 mm and other parameters are summarized in Table 2 24 .

| Data processing
Spectroscopic data from the 3D CSI protocol were processed in Matlab 2018b (The Mathworks Inc., Natick, MA) using an open source in-house designed processing tool (CSIgui v1.1, http://www.csigui.tk, April 2019). 31 P spectroscopy data were averaged and spatially filtered using a 3D hamming window followed by an inverse Fourier transformation to the spatial domain. All free induction decays (FID) were apodized using a 24 Hz gaussian filter and zero filled to 512 samples. Coil data was combined using the whitened singular value decomposition (WSVD) algorithm as reported by Rodgers et al 27 . Zeroth order phase correction was applied automatically, and first order phase correction was applied manually, thereafter. No additional nor aesthetic baseline corrections were performed. Spectra from tumors exceeding the voxel resolution were aligned to the metabolite peak with the highest SNR followed by averaging, excluding voxels with a 50% or less partial tumor tissue volume on available MR images. The SNR of metabolites was calculated using equation [2] with S max , the real part of the maximum signal intensity and the noise defined as the absolute standard deviation of the last 50 samples points of the spectrum.

| RESULTS
Simulations resulted in a maximum decrease of 8% in SNR per unit of time within the used TR/T 1 range for + and − 30% deviating flip angles, as can be seen in Figure 2. In addition, the α E + 30% variation showed a lower decrease in SNR per unit of time compared to the α E -30% variation. A similar trend is seen for a α E ± 50% variation showing a maximum decrease of 23% in SNR per unit of time within the same TR/T1 range for the α -50% variation. According to the B 1 maps available for the 31 P body coil, we could expect a maximum of 30% deviation in flip angle in the in-vivo measurements using equal power settings between subjects and, in addition, a maximum decrease in SNR per unit of time of less than 6% is seen for the TR/T 1 -ratios range that corresponds to the 31 P metabolites of interest (0.009; 0.13) and the proposed protocol TR (60 ms) 23 .
All patients were imaged within an hour of scan time with one of the two setups. Positioning the 1 H transmit coils for patient #1 was limited due to a stent in the superior vena cava (SVC) located close to the tumor. No other patient related difficulties were experienced during the scan sessions. Images obtained with the dipole antenna in A1 were adequate for tumor localization and planning ( Figure 3A), when tumor location was known from previous CT images ( Figure 3B).
Images and tumor localization using A2 were improved compared to A1 as is depicted in Figure 3A Figure 3E have higher SNR compared to the anterior side.
Obtained 31 P lung carcinoma spectra were acquired with high SNR for PCr (9.5) and the ATP resonances (>4.7) using A1, the 31 P dual-coil Rx ( Figure 4A) and with high SNR ranging from 3.9 (PME) to 13.2 (α-ATP) using A2, the 31 P 16-channel Rx array ( Figure 4B). It allowed discrimination of PME, Pi, PDE, PCr in all patients, the three ATP resonances and UDPG in all subjects except for patient #1 and NADPH in patient #4. Moreover, the SNR of the phospholipid-and energy-metabolites was found higher with A2 compared to A1. The lack of B 0 shimming and partial volume effects over the large field of view is visible in the spectra with measured linewidths ranging from 0.20 ppm to 1.1 ppm after apodization when using either coil setup.

| DISCUSSION
3D 31 P spectroscopic imaging was successfully obtained in four lung carcinoma patients with either the 31 P dual-coil receiver or the 16-channel receiver array in combination with the integrated 31 P body coil at 7 tesla. Both Rx setups allowed the acquisition of phosphorous metabolic information from the lung carcinoma via a non-invasive method, while targeting the full organ for evaluation. The increased spectral resolution at the ultrahigh magnetic field strength of 7 T allowed differentiation of multiple phosphorous metabolites related to cell membrane and energy metabolism. A minimal decrease in SNR per unit of time was apparent from the simulations performed to study the effect of a +/−30% deviation from the Ernst angle due to the lack of individual body coil power calibrations in this patient population. This minimal SNR loss of at maximum 8% allowed for 3D fast spectroscopic imaging with short TR and low flip angle excitation without time-consuming calibration steps during the scan session.

FIGURE 2
Simulation of the SNR per unit of time for the 3D 31 P spectroscopic imaging at Ernst angle (α E ) and with a 30% and 50% deviation for the TR/T 1 ratio ranging from 10 −6 to 0.3. The SNR per unit of time at α E is marked by the solid black line, the increased and decreased angles for both the 30% (red) and 50% (blue) deviations are displayed as dashed and dash-dotted lines respectively Increasing the number of receiver coils improved the field of view coverage of 31 P MRS images expanding the available metabolic information over a larger field of view. This agrees with previous demonstrations in literature 28 . In addition, the SNR increase gained with the 16-channel receiver array used in patient #3 and #4 not only allowed discrimination of PME, Pi, GPE, GPC, PCr, ATP (with α-, ß-& γ-resonances) and uridine-diphosphate glucose (UDPG) as with the dual-coil receiver but also nicotinamide-adenine dinucleotide phosphate (NADPH) in patient #4. UPDG is a known liver metabolite and indicates minor liver signal contamination, however SNR was insignificant (SNR < 3) 5 . NADPH (SNR > 3) however, though also found in the liver, is a cofactor involved with anabolic reactions, already linked to tumor tissue 29 . In addition, the highest SNR of the dual-coil receiver was measured for PCr ( Figure 4A, Patient #1) which is not directly associated with tumor tissue, but rather muscle tissue 5 . This can be explained by signal contamination from chest muscle signals contained in neighbouring voxels that bled in the tumor FIGURE 4 Spectra of lung tumor tissue for all four patients acquired with the 31 P chemical shift imaging protocol using a) A1, the 31 P dual coil Rx and B) A2, the 31 P 16channel Rx array. Phosphomonoesters (PME), phosphodiesters (PDE), glycerophosphoethanolamine (GPE) plus glycerophosphocholine (GPC), inorganic phosphate (pi), phosphocreatine (PCr) and the α-, βand γ-ATP resonances are labelled where applicable. The number of tumor voxels used for averaging is denoted by N in the right top corner of the spectrum except for single voxel spectra. Notice the increase in PDE with respect to PME in patient #4 that might indicate tumor response to immunotherapy FIGURE 3 A) Coronal MR image including labels for the tumor, neck and lungs plus B) a coronal CT image with PET scan overlay, both from patient #1 and used for tumor localization. C) Single transverse slice of the 3D spectroscopic imaging data from patient #1 with the tumor voxels indicated by the red rectangle. D) Transverse and coronal CT images from patient #4 for tumor localization and planning. E) the MR image from patient #4 with an overlay of a single slice of the 3D spectroscopic imaging data. Tumor voxel is highlighted by the yellow rectangle voxel location due to the small field of view of this patient, which excluded the full body circumference, in combination with the relatively large voxel size. Additional averaging of the 20 voxels also increased signal contamination in this patient but was required to regain the SNR of the spectrum. Spectra acquired with the 16ch Rx still show signal contamination, as can be seen by the remaining PCr peaks, though to a lesser extent than the first patient and even in opposite phase ( Figure 4B, patient #3). The lung and tumor morphology itself may already minimize signal contamination from neighboring voxels as tissue density in healthy lung tissue is, compared to other areas in the body, extremely low. Further protocol development could minimize signal contamination within a short time frame by increasing spatial voxel resolution or reducing point spread by more complex k-space weightings and filtering 24 . Another strategy could be the use of selective pulses to fully eliminate specific tissue signal such as the one coming from the muscles.
The top part of the torso, especially at the collar bones, limits proper positioning of the top elements of the receiver array. The eight rigid elements of the array lack the body-shaped curvature, disfavouring coil loading and resulting in a suboptimal receive fields for these coils. This can be seen by the increased SNR at the posterior side of the patient compared to the anterior voxels shown in Figure 3C. Additional suboptimal coil combinations using the WSVD algorithm could also disfavour the SNR gradient over the spectroscopic image.
Moreover, in the presence of large susceptibility differences, such as the lung itself, implants and the moving heart, spectral quality is surprisingly adequate for the distinction of the metabolites, even without B 0 shimming. Resulting B 0 homogeneity was adequate with a spectral linewidth ranging between 0.2 and 1.1 ppm. This B 0 non-uniformity is much less critical than for 1 H MRS as the spectral separation between metabolites (i.e. PME versus PDE) is substantial (i.e. >3.5 ppm) at 7 T. However, it should still be noted that the B0 field uniformity can be highly variable both spatially as temporally. In our previous study we simulated spatiotemporal magnetic field uniformity, which at worst case conditions (i.e. at diaphragm comparing fully inhaled versus exhaled condition) could be up to 3 ppm 30 . Either avoiding inclusion of subjects with tumors in locations of such severe non uniformities or using new means of local shim coils that can mitigate these distortions may be required 31 . In addition, shimming procedures could be performed within a breath-hold and when combined with gating it is expected to improve linewidth, possibly increasing sensitivity to allow detections of the individual PME and PDE 32 .
In this study we have altered the flip angle between subjects. As prior knowledge about signal levels was unknown, we started by focusing on ATP, therefore setting the flip angle to 20°.After confirming observation of phospholipids, the angle was set to 9°(i.e. Ernst angle for PME and PDE). Finally, we completed the protocol by small over-tipping to also consider SNR of other metabolites that all have a shorter T 1 . Note that the spectra are obtained with T 1 weighting, so altered peak ratios can be caused by concentration differences, but also by alterations in T 1 . To extract the T 1 dependence, subject specific T 1 knowledge could be obtained by acquiring the same scan twice albeit with a different flip angle.

| CONCLUSION
We conclude that 31 P MRSI in lung carcinoma is feasible at 7 T. Employing large receiver arrays that can cover the whole torso, improves the field of view coverage allowing full organ 31 P-MRSI acquisition. With only minor signal contamination to overcome, 31 P MRSI shows great potential as tumor biomarker for treatment response monitoring in lung cancer.