A scan-specific quality control acquisition for clinical whole-body (WB) MRI protocols

Objective. Image quality in whole-body MRI (WB-MRI) may be degraded by faulty radiofrequency (RF) coil elements or mispositioning of the coil arrays. Phantom-based quality control (QC) is used to identify broken RF coil elements but the frequency of these acquisitions is limited by scanner and staff availability. This work aimed to develop a scan-specific QC acquisition and processing pipeline to detect broken RF coil elements, which is sufficiently rapid to be added to the clinical WB-MRI protocol. The purpose of this is to improve the quality of WB-MRI by reducing the number of patient examinations conducted with suboptimal equipment. Approach. A rapid acquisition (14 s additional acquisition time per imaging station) was developed that identifies broken RF coil elements by acquiring images from each individual coil element and using the integral body coil. This acquisition was added to one centre’s clinical WB-MRI protocol for one year (892 examinations) to evaluate the effect of this scan-specific QC. To demonstrate applicability in multi-centre imaging trials, the technique was also implemented on scanners from three manufacturers. Main results. Over the course of the study RF coil elements were flagged as potentially broken on five occasions, with the faults confirmed in four of those cases. The method had a precision of 80% and a recall of 100% for detecting faulty RF coil elements. The coil array positioning measurements were consistent across scanners and have been used to define the expected variation in signal. Significance. The technique demonstrated here can identify faulty RF coil elements and positioning errors and is a practical addition to the clinical WB-MRI protocol. This approach was fully implemented on systems from two manufacturers and partially implemented on a third. It has potential to reduce the number of clinical examinations conducted with suboptimal hardware and improve image quality across multi-centre studies.


Background
Most modern MRI radiofrequency (RF) receiver coil arrays utilise several groups of independent coil elements, facilitating higher signal-to-noise ratio (SNR) and acquisition acceleration techniques (Griswold et al 2002, Lustig et al 2008, Gruber et al 2018).The performance of any one of these RF coil elements may degrade over time, leading to the detection of little or no signal for that coil element.In phased-array imaging intensity uniformity correction algorithms are used to compensate for this to an extent; however, there is still an effect on local SNR (Charles-Edwards et al 2016).While this SNR reduction may not always be noticeable to the naked eye, subtle pathologies imaged in the region of a broken RF coil element could go undetected as a result, particularly for acquisitions that have inherently low SNR such as diffusion-weighted MRI.
Whole-body MRI (WB-MRI) utilises multiple RF coil arrays to achieve full body coverage by acquiring images in several stations, with groups of RF coil elements activated based on their proximity to the imaging volume.WB-MRI protocols consisting of axial diffusion-weighted imaging (DWI) and Dixon imaging, and sagittal spine imaging are recommended in oncology applications (Padhani et al 2017, Messiou et al 2019, Petralia et al 2021), where WB-MRI has demonstrated value in diagnosis, staging, assessment of therapeutic response and screening (Morone et al 2017).
A typical WB-MRI acquisition could include up to five RF receiver coil arrays with a combined total of more than 100 coil elements.The number of RF coil elements involved means that a WB-MRI study has an increased likelihood of encountering one or more broken elements, which result in the detection of low or zero signal.Additionally, the inherently low signal in DWI with high b-values (typically 900 s mm −2 or higher in WB-MRI) makes it less able to withstand the loss in SNR resulting from broken elements, with potentially significant effects on the image quality.
Phantom-based quality control (QC) is used routinely in clinical MRI departments and scanner manufacturers typically provide an automated QC acquisition and analysis routine that measures the SNR from each element individually to identify any that are recording low signal (Kwok 2022).Practical considerations limit the frequency with which these measurements can be made in a busy clinical department however, and while some RF coil arrays may be tested every day, there may be an interval of several months between QC tests for certain coil arrays.The examination of images from individual RF coil elements is recommended in national guidance to investigate image quality issues (Charles-Edwards et al 2016) but these images are not routinely reconstructed in clinical practice in the interests of streamlining data storage and simplifying image reporting.
The infrequency of testing means that faulty RF coil elements are often first identified when the scanning radiographer or reporting radiologist observes poor image quality.The reduction in SNR may not, however, be sufficiently large to elicit a response from the reader but this does not mean that subtle contrast differences are not being obscured by the degradation in image quality.Many patient examinations could therefore be conducted using suboptimal equipment before being detected by phantom-based QC.By acquiring QC images during clinical examinations, image quality can be monitored with a frequency that matches the clinical use of the RF coil arrays (Peltonen et al 2018, Tracey et al 2023).
In addition to image quality issues associated with scanner hardware, WB-MRI is a challenging technique to perform (Winfield et al 2021) and has limited adoption outside of specialist centres (Hillengass et al 2019).One of the challenges is setting up the RF coil arrays and patient in a way that achieves full coverage with adequate SNR, minimising regional variations in signal.Regional variation due to coil array misplacement may be clearly visible by eye but it would be useful to have a quantitative measure, acquired during the clinical examination, to alert the operator to positioning errors.
The aim of this work was to develop a rapid QC acquisition to detect faulty RF receiver coil elements and to identify errors in RF coil array placement during patient set-up.The acquisition time should be sufficiently short to be a practical addition to the WB-MRI clinical protocol, allowing for on-going monitoring of image quality during clinical examinations.
In addition to having value in routine clinical use, it is proposed that this approach could be a valuable tool in multi-centre WB-MRI studies, a crucial step in the translation of the technique into more widespread clinical practice (O'Connor et al 2017).To be applicable in a multi-centre setting, the method needs to be manufactureragnostic, so work was also undertaken to implement equivalent acquisition and processing pipelines for systems from three manufacturers.

Technique development
Development work was conducted with a 1.5 T scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany).The RF coil array configuration matched that used clinically and consisted of a 20-element head and neck coil array, two 18-element body arrays (identified as A and B), a 32-element spine array and, for patients with myeloma, a 36-element peripheral angiography array.For clarity, here 'RF coil array' refers to the entire structure (e.g. the head and neck coil array), which is made up of several 'RF coil elements' that each detect signal.This is distinct from the 'integral body coil', which is built into the scanner and used primarily as an RF transmitter in usual practice.
A QC acquisition was devised as a proton density weighted single-slice coronal gradient echo sequence with six contiguous stations, TR = 50 ms, TE = 3.37 ms (the lowest in-phase value), slice thickness = 10 mm, image resolution = 4.7 mm × 4.7 mm and FOV = 300 mm × 498 mm.Images for each station were acquired using the automatically selected RF coil elements, which may come from multiple RF coil arrays.Each individual element image was reconstructed and saved as well as the image reconstructed from all active elements.The integral body coil was used to acquire reference images of signal intensity that are independent of local receiver coil positioning.
The method yields three sets of images: one image per station reconstructed using signal from all automatically selected RF coil elements, one image per station reconstructed using signal from the integral body coil, and a variable number of images per station reconstructed using signal from each individual RF coil element.Total acquisition time for the additional QC sequences (one per imaging station) was 82 s (approximately 14 s per station), which is negligible in a WB-MRI examination that has a usual duration of 45-60 min.
To further define the method and analysis pipeline, ten healthy volunteers (male/female: 6/4, median age = 31 years, range = 25-44 years) were scanned using the QC acquisition.Volunteer studies were approved by a research ethics committee (ClinicalTrials.govidentifier: NCT05118555) and volunteers gave written informed consent.
The effects of imperfect positioning were simulated by repeating the acquisitions with the two 18-element body coil arrays rearranged as follows: (1) 2 cm overlap between the two body coil arrays (the recommended clinical set-up) (2) No overlap or gap (3) 2 cm gap (4) 5 cm gap (5) Lower body array placed on padding to raise away from the volunteer's body.
Images were processed offline using custom-built Matlab (R2019a, MathWorks, Natick, MA, USA) scripts, using the two pipelines described in figure 1.To identify broken RF coil elements the signal-to-background ratio (SBR) was calculated for each individual coil element image (SBR element ) and the sum-of-squares combined coil elements image (SBR combined ).SBR is defined as the ratio of the mean signal occurring from within the volume of the patient (defined by a binary segmentation of the patient outline on the combined element image) to the standard deviation of the signal depicted outside of the volume of the patient.Although less accurate than other methods as it assumes that noise conforms to a Rician distribution (Goerner and Clarke 2011), SBR is used as a proxy for SNR due to the simplicity of calculation.An element was flagged as potentially broken if the ratio of SBR element to SBR combined fell below a threshold value of 0.001.Experiments were made using various threshold values and, although not mathematically optimised, the empirically determined threshold of 0.001 was successful in identifying all cases where faulty coil elements were subsequently verified.
To identify RF coil array positioning errors, both sets of images were scaled and smoothed before the combined element image was subtracted from the integral body coil image to generate a difference image (figure 1(bii)).The mean signal was calculated for each row of the difference images to produce a 1D profile of signal difference for the entire length of the body (figure 1(biii)).An expected profile and range were defined using the acquisitions with the ideal set-up and the percentage of image rows falling outside of this expected range was used to quantify the amount of variation in the profile for an individual examination, where greater variation indicates regions of lower signal.The processing pipeline took approximately five seconds to complete for a single examination.

Clinical evaluation
Patient studies were approved by the Institutional Review Board.The requirement for written consent was waived by the institution.
The QC acquisition was added to the clinical WB-MRI protocol for all patients scanned at a single institution between 9th April 2022 and 5th April 2023.This included six scanners across two sites (two 1.5 T MAGNETOM Aeras, two 1.5 T MAGNETOM Solas and two 3 T MAGNETOM Vidas, all Siemens Healthcare, Erlangen, Germany).Patients who were referred for WB-MRI examinations were included (at our institution WB-MRI is conducted in patients with myeloma, metastatic prostate and breast cancers, and a small number of patients with other indications).Clinical imaging was conducted according to the relevant clinical guidelines (Padhani et al 2017, Messiou et al 2019), with the additional QC acquisition appended to the end of the protocol.The total number of included patient examinations was 892.
All QC images were transferred offline and processed as described in figure 1 to identify RF coil elements recording low signal and coil array positioning errors.For an individual exam a coil element may be recording low signal for legitimate reasons unrelated to hardware failure (i.e. a small volume of tissue in the region of the coil element's sensitivity) so a single instance of low signal was not considered justification for immediate investigation.Instead, failures were monitored over the course of several consecutive exams, with further investigation required if a given element recorded low signal for: • Three consecutive exams • Four out of five consecutive exams • Five out of eight consecutive exams When any of these criteria were met, the manufacturer's automated phantom QC acquisition was used to determine whether the coil array needed to be replaced.

Other manufacturers
Equivalent acquisitions were implemented on a 1.5 T Signa Artist (GE Healthcare, Waukesha, WI, USA) and a 3 T Ingenia (Philips Healthcare, Best, Netherlands).Some minor modifications to acquisition parameters were necessary, as summarised in table 1. Sample images were acquired from healthy volunteers on each system and an equivalent processing pipeline was developed to that described in figure 1.

Results
Figure 2 illustrates the approximate positioning of the RF coil arrays alongside sample QC images and the individual element images for a single station when all RF coil elements are functioning.The healthy volunteer acquisitions were used to define the processing pipeline and set of criteria described in the previous section.
During the first five volunteer acquisitions, a coil element in one of the 18-element body arrays was observed to be recording persistently low signal.Further investigation confirmed the fault and the coil array was replaced.
Example signal difference profiles are shown for two set-ups in figure 3. The profiles acquired with the ideal set-up were used to define the expected range of variation (mean ± 2 * std.) for the subsequent clinical evaluation.For volunteer examinations conducted after the coil array replacement (i.e. with fully-functional equipment), the percentage of image rows exceeding the expected range in variation is shown in figure 3.
Figure 4 summarises the detection of faulty coil elements across all 892 patient examinations on all scanners in the clinical evaluation.Faulty coil elements were flagged on five occasions over the course of the study and were investigated using the manufacturer's QC routine.Faults were confirmed in four of these cases and the RF coil array was replaced by the manufacturer.In the case where the coil array was not replaced, an intermittent fault was recognised but did not occur with sufficient consistency for replacement.Figure 5 demonstrates the effect of a faulty RF coil element on QC images and on the clinical DWI.
Faulty elements were also flagged on several occasions for the lower coil elements of the spine array; however, this occurred in examinations covering from skull vertex to upper thighs, which did not use the peripheral angiography coil array.Patients' legs were therefore raised away from the table for comfort and situated outside of the imaging slice, leading to false positives.As the cause of these was deduced and can be easily avoided in future by reducing the number of stations for these patient groups, they are not included in the statistical breakdown of RF coil array failures.Routine phantom QC continued throughout the duration of the study and did not detect any coil array failures that were not identified by the QC acquisitions, indicating good sensitivity.Using the definitions of precision and recall defined for object detection (Godil et al 2014), these results give the technique a precision of 80% and a recall of 100%.The true precision may in fact be higher than this as the intermittent fault was classified as a false positive because the coil array was not replaced, although a fault does appear to be present.The inter-scanner variation in mean difference image line profile was minimal, as illustrated in figure 6.The data from all scanners was therefore used to calculate a rolling mean line profile and limits of agreement throughout the evaluation.Across all scanners the mean percentage of image rows that fell outside of the limits of agreement was 3.4% (std.= 5.5%).There were 41% of exams with 0 rows outside of the limits of agreement, with 72% of exams having less than 5% of rows outside the limits and 88% having less than 10%.Line profiles and images from example patients are also shown in figure 6.
Images acquired from all three manufacturers' systems are shown in figure 7 to illustrate the equivalence of the acquisitions for both the images acquired with all active coil elements and those acquired with the integral body coil.The processing pipeline was adapted to be manufacturer-agnostic and was able to plot line profiles and identify positioning errors using data from all three manufacturers.For Philips scanners, any RF coil element which will not significantly contribute to SNR for the planned geometry is identified by a prescan and then deactivated, to preserve image quality by removing coil elements which will not improve the final image.As the method here required individual coil element images to be acquired for all elements, this means that this processing pipeline is not able to detect faulty RF coil elements for Philips scanners.

Discussion
The results of this study show that examination-specific QC can detect broken RF coil elements and may enable more frequent monitoring of hardware performance than phantom-based QC.Hardware issues and positioning errors contribute to poor image quality in WB-MRI and more consistent imaging performance may be achieved by implementing a method for real-time monitoring of image quality using data acquired during each clinical examination.
This evaluation detected four RF coil element faults over the course of a single year in our institution, which leads to an indicative order-of-magnitude estimate of approximately one failure every 220 WB-MRI measurements.The clinical impact of broken RF coil elements is difficult to evaluate.There are circumstances where the resulting reduction in the SNR of clinical images is substantial enough to prompt the reading radiologist to raise a concern; however, the effect can be less pronounced.In these cases, the drop in SNR may go unnoticed but subtle pathology could be obscured.The approach presented here allows for ongoing monitoring Example QC images for all six stations acquired using all active elements in all RF coil arrays and using the integral body coil, displayed alongside a schematic showing the approximate location of the RF coil arrays relative to the patient (HN-head and neck, S-spine, BA/BB-body matrix A/B, PA-peripheral angiography).(b) Individual RF coil element images for all active coil elements in station 2 (the location of station 2 is indicated in (A).Coil elements within a coil array are divided into groups and activated by group for a given station.E.g. body matrix A has 18 elements divided into 3 groups of 6 elements.For station 2, the top two groups have been activated (BA11-16 and BA21-26) but the third group is too far from the signal-generating volume and is inactive.The RF coil elements are labelled according to the nomenclature of the manufacturer (i.e. the labels found in the DICOM header for images acquired using these elements).
of RF coil array performance and demonstrated good sensitivity for detecting broken RF coil elements in clinical practice.
There was one occasion during this evaluation when a coil was identified as faulty by the method but could not be replaced as it did not consistently fail the manufacturer's QC routine.These intermittent failures will be highlighted by this method to allow for further investigation, although it may be difficult to act on these kinds of failures without persistent failure of the manufacturer's coil array QA routine.
Some RF coil array safety checks are performed during the scanner adjustment/preparation stage before scanning, preventing the scan from continuing if a problem is found.This evaluation did not find any cases where these tests failed, even in the cases where broken RF coil elements and image quality issues were later detected.This suggests that the RF coil arrays with broken coil elements were passing the safety checks even when image quality was affected, illustrating the value of this method beyond the existing safety checks.
The total additional imaging time due to the QC acquisitions was approximately 20 h across all patients on all scanners (for the 12 month clinical evaluation).The same amount of time, shared equally across scanners, would only be sufficient to run the manufacturer's QA routine twice per scanner for all involved coil arrays during the same 12 month period.This would leave periods of several months where the performance of the coil arrays was unverified, with an unknown effect on clinical imaging.While the focus of this work was on WB-MRI, the RF coil arrays are also used for other examinations and suboptimal performance may have an effect across the spectrum of clinical imaging.WB-MRI examinations include many of the most frequently used coil arrays at our centre, although equivalent examination-specific QC protocols could be developed to include RF coil arrays that are not used in WB-MRI.
In the case of RF coil array positioning errors, it is valuable to have a quantitative measurement that allows the cause of poor image quality to be diagnosed and addressed.This could be particularly useful in the context of multi-centre studies involving sites with limited experience of WB-MRI, allowing feedback to be provided to achieve higher image quality for subsequent scans.The integral body coil has a fixed position and can therefore be used as a control to identify mispositioning of the receiver coils.This would be a significant aid to data quality assurance in multi-centre clinical trials.
For this dataset, the line profiles of individual patients were generally in good agreement with the mean line profile calculated across all examinations.Our centre has extensive experience of acquiring WB-MRI data so it is reasonable to assume that the majority of exams were conducted with the correct positioning of coil arrays.These data can be used to define an acceptable threshold for the minimum number of image rows that fall outside of the expected variation in difference profile (e.g.10%), with exams failing QC checks if they do not meet this requirement.
The demonstration of approximately equivalent approaches on systems from two other manufacturers is critical if this method is to be useful in multi-centre trials, which must include scanners from a range of manufacturers if they are to effectively facilitate translation into wider clinical practice.
A limitation of this work is that our year-long clinical evaluation only included scanners from a single manufacturer.Although implementation of the technique has been demonstrated for other manufacturers, the ability to detect faults has not been comparably evaluated.The approach taken on Philips scanners to preemptively identify RF coil elements that are not contributing to SNR means that this approach is not suitable for detecting broken coil elements on those systems.The workflow described here also requires research licenses for Philips and GE scanners to create the acquisition sequences, although not for Siemens.In addition, there was one occasion where a coil element was identified as persistently failing by this method but did not consistently fail the manufacturer's QC routine and was therefore not replaced.
The QC images were acquired at the end of examinations and analysed retrospectively in this study as the objective was to assess the value of the QC method.The ideal implementation for this approach however would be for in-line fault finding.If the processing pipeline could be established to run on the MR scanner or a directly connected device and produce outputs in a reasonable timeframe, the QC images could be acquired at the beginning of examinations and not the end.This would mean that potential problems could be brought to the Each column is a single exam and each row is a particular coil element, grouped by coil array (HN-head and neck, S-spine, BA/BB-body matrix A/B, PA-peripheral angiography).Colour scheme: black-the entire coil array was unused for that exam, Blue-the coil element passed the QC test, Green -the coil element fell below the threshold for that particular exam, Yellow-the coil element fell below the threshold and is part of a group of exams that were collectively highlighted as a potential broken coil element (highlighted by dotted red ovals).attention of the operator at a time when it is still possible to correct for them, either by swapping for a different RF coil array or by patient repositioning.
Current WB-DWI examinations utilise multiple repeat acquisitions to improve SNR.There is a trend towards accelerated acquisitions utilising machine learning based noise filtering, which requires fewer acquisitions with lower SNR to produce the final image in a shorter acquisition time (Zormpas-Petridis et al 2021, Tajima et al 2022).Such techniques will require increased confidence in the performance of the RF coil arrays to produce reliable reconstructions, necessitating accurate and frequent QC.

Conclusion
This work has demonstrated a rapid and practical QC acquisition that can detect broken RF coil elements and positioning errors in WB-MRI examinations.The approach has been evaluated through implementation in the clinical protocol for all WB-MRI examinations over a period of one year in one institution, identifying four broken RF coil elements that may have gone undetected for some time otherwise.The additional acquisition time is sufficiently short to make this a practical addition to the clinical protocol, allowing continuous  monitoring of RF coil array performance outside of routine phantom-based QA.Similar approaches have been implemented on scanners from three manufacturers, demonstrating the versatility of the method.
As well as providing a method of retrospective quality control, this technique will be developed to provide real-time feedback on RF coil array performance, allowing issues to be addressed in advance of clinical imaging.This has the potential to improve the quality of WB-MRI in multi-centre trials and reduce the number of clinical examinations conducted with sub-optimal equipment.
The clinical evaluation was approved by the institutional review board (The Royal Marsden Hospital NHS Foundation Trust Committee for Clinical Research).The requirement for written consent was waived by the institution.

Figure 1 .
Figure 1.Graphical description of both strands of the processing pipeline.(a) Detection of broken RF coil elements by calculating the signal to background ratio (SBR) of the combined element image (i) and each individual element image (ii).When the ratio of these values (iii) persistently falls below an empirically determined threshold value, the RF coil element is flagged as faulty (iv).(b) Detection of RF coil array positioning errors by smoothing and rescaling the combined element and integral body coil images (i).A difference image is calculated (ii) and a 1D profile of signal difference (iii) is used to identify regions where the signal in the combined element image is low (iv).

Figure 2 .
Figure 2. (a)Example QC images for all six stations acquired using all active elements in all RF coil arrays and using the integral body coil, displayed alongside a schematic showing the approximate location of the RF coil arrays relative to the patient (HN-head and neck, S-spine, BA/BB-body matrix A/B, PA-peripheral angiography).(b) Individual RF coil element images for all active coil elements in station 2 (the location of station 2 is indicated in (A).Coil elements within a coil array are divided into groups and activated by group for a given station.E.g. body matrix A has 18 elements divided into 3 groups of 6 elements.For station 2, the top two groups have been activated (BA11-16 and BA21-26) but the third group is too far from the signal-generating volume and is inactive.The RF coil elements are labelled according to the nomenclature of the manufacturer (i.e. the labels found in the DICOM header for images acquired using these elements).

Figure 3 .
Figure 3. (a) Mean difference profiles for each of the volunteers scanned following the replacement of the RF coil array in the ideal coil array set-up, plotted alongside expected signal variation (calculated as the mean ± 2 * std.).(b) Mean difference profiles for each of the volunteers scanned following the replacement of the RF coil array with a 5 cm gap between the 18-element body coil arrays.The approximate location of the coil array gap is indicated by the red arrow.(C) A graph showing the percentage of image rows that fell outside of this expected range for each of the set-ups in these volunteers.

Figure 4 .
Figure4.Illustration of faulty coil element detection across all exams on all scanners.Each column is a single exam and each row is a particular coil element, grouped by coil array (HN-head and neck, S-spine, BA/BB-body matrix A/B, PA-peripheral angiography).Colour scheme: black-the entire coil array was unused for that exam, Blue-the coil element passed the QC test, Green -the coil element fell below the threshold for that particular exam, Yellow-the coil element fell below the threshold and is part of a group of exams that were collectively highlighted as a potential broken coil element (highlighted by dotted red ovals).

Figure 5 .
Figure 5. (a) Images from two rows of individual RF coil elements across body coil arrays A and B (BA and BB), demonstrating low signal from element BA34.The faulty coil element has a limited effect on the combined elements image, where the approximate location of BA34 is indicated by the red circle.(b) Clinical diffusion-weighted (DW) images (b-value = 900 s mm −2 ) acquired five weeks (row 1) and one week (row 2) prior to fault detection, and during the exams where the fault was detected (row 3).Images are shown from matching slice numbers and therefore illustrate approximately the same position relative to the faulty coil element.The deterioration in image quality between these exams was highlighted by a radiologist during a blinded assessment.

Figure 6 .
Figure 6.(a) The mean difference image line profile for each scanner, illustrating the low inter-scanner variation.Mean difference profiles represent the difference in signal between the integral body coil image and the combined element image for each row of the image and are calculated using the method described in figure 2. (b) The mean line profile across all scanners (black) and limits of agreement from the mean (red dotted), plotted alongside three example profiles from individual examinations.(c) Composed combined element images from the three examples shown in B. Regions of greater signal difference are indicated in matching colours on the profiles (b) and images (c).

Table 1 .
Table of parameters for the QC acquisitions as implemented across the three manufacturers' systems.
* For the development work in volunteers, TE = 3.4 ms.