Data collection
Data were collected across three sessions from February to May 2022. four sets of T1-weighted images were acquired on a 1.5 and 3T scanner (the scanners are from different manufacturers) in each session, two with the manufacturer’s proprietary ANR off and two with ANR on. All other user-controlled acquisition parameters remained unchanged. Data were acquired in accordance with the ACR phantom scanning instructions [10], with the phantom (Fig. 1) positioned inside a head coil. key acquisition parameters are shown in Table 1. 20 and 32 channel head coils were used at 1.5T and 3T respectively.
Table 1
Key acquisition parameters
| Pulse sequence | TR (ms) | TE | FOV | total slices | Slice thickness | Slice gap | NEX | Matrix | Acq time |
| Spin Echo | 200 (1.5T), 500 (3T) | 20 ms | 25 cm | 11 | 5 mm | 5 mm | 1 | 256*256 | 2:10(3T), 2:12(1.5T) |
Image Analysis
DICOM files were exported from the acquisition workstations and processed using ImageJ v1.47n [11]. Custom macros were used to extract data from the images, the function of each is described below.
CNR data was extracted from the four low-contrast object phantom slices. These slices contain spokes of low contrast disks, average counts in ROIs placed in eight of the most easily seen disks and two accompanying background ROIs were extracted with a macro that prompted the user to manually place a point in the centre of each disk and background region before optimising the final ROI location to maximise CNR (finding highest average signal in each disk and lowest average signal in each local background region). An example of a low contrast disk slice and resultant ROIs is shown in Fig. 2a-b.
SNR, ghosting, and intensity uniformity data were all extracted from a water-only slice near the centre of the phantom, utilising a 200cm2 circular area in the centre, as this region should have relatively uniform signal intensity within a head coil [10].
For SNR and IU data collection, an automated macro was used to place a series of ~ 1cm2 area tessellated hexagons inside the central 200cm2 phantom region, each of which yielded an SNR and IU value. These regions are shown in Fig. 2c.
An automated ghosting data extraction macro found the mean value within the central 200cm2 phantom region and from each of a circular array of background ROIs placed around the outside of the phantom, equidistant from its centre (Fig. 2d).
Data Processing
CNR: The method used by the Royal Australian and New Zealand College of Radiologists for signal difference to noise ratio[12] was used to calculate CNR for each low contrast disk/background using the mean (µ) and standard deviation (SD) as show in Eq. 1.
\(CNR=\frac{({\mu }_{disk}-{\mu }_{bckgrnd})}{\left(\sqrt{{SD}_{disk}^{2}+{SD}_{bckgrnd}^{2}}\right)/2}\) (Eq. 1)
SNR: The ACR MRI QC Manual NEMA method [13] was employed for each of the contiguous hexagonal array of ROIs. The within-session repeat acquisition was subtracted from the first acquisition to create a “difference image” for noise quantification. The mean signal of each hexagonal each water slice ROI and the SD of the corresponding difference image ROI were then used to calculate an array of SNR values for each session according to Eq. 2.
\(SNR=\sqrt{2}\frac{{\mu }_{water slice ROI}}{{SD}_{difference ROI}}\) (Eq. 2)
Ghosting: The ghosting % in each background ROI was found by dividing the mean value of the central water ROI by the mean of the background ROI and multiplying by 100.
Intensity Uniformity: The mean signal of the central water region was used to convert each hexagonal ROI to an absolute % signal variation according to Eq. 3.
\(IU deviation \%=\left|\left(\frac{{\mu }_{hexagonal ROI}}{{\mu }_{central ROI}}*100\right)-100\right|\) (Eq. 3)