[OA043] A new signal-to-noise ratio (SNR) measurement method based on low-pass filtering for magnetic resonance imaging (MRI) quality control (QC) purposes

Purposes Signal-to-noise ratio (SNR) is a prominent metric to assess scanners performances. SNR measurement in MRI is challenging because of the difficulty to quantify noise level from images. A gold-standard consist in pixel-by-pixel variance calculation on several images (u003e100) scanned in identical conditions. This method is accurate but cannot be practically implemented because of time constraints. Another gold-standard consists in scanning images without radiofrequency excitation hence collecting noise-only. This latter is also unpractical because it requires a specific and generally not-granted access to the hardware. Consequently, there is a need for designing alternative methods for accurate and precise noise quantification. This work aims to introduce a new noise level quantification method for SNR measurement. Methods Our method ( Noise filt ) quantifies noise level by combining these operations: (1) subtraction of two images subsequently scanned in identical conditions, (2) Low-pass filtering to extract and suppress low-frequency remaining information, (3) ( Noise filt  = standard deviation of pixels. Using a dataset of 30 exams of a head phantom scanned on a 3T MR unit with an 8-channel phased-array brain coil, we compared our method with three others from the literature: noise level quantified using one image-background region-of-interest (ROI) ( Noise bgd ), several background ROIs ( Noise bgd+ ), and from the subtraction of two images ( Noise diff ) [6]. By means of Bland–Altman analysis, we assessed the agreement between all these methods and the noise-only standard method ( Noise noRF ). We also assessed their correlation with Noise noRF (Pearson’s correlation coefficient r , p Results Bland–Altman analysis showed that Noise filt agrees more with the standard than the three other methods. Correlation of our Noise filt method with the gold-standard ( r filt =0.80, p r bgd  = 0.37, r bgd+  = 0.61, r diff  = −0.04). Conclusion We demonstrated that our method produces accurate measurement, is simple to implement, and is usable for instance for quality control (QC) purposes. This new metric is being tested with additional scans in QC monitoring of different MR scanners. Download high-res image (87KB) Download full-size image Download high-res image (159KB) Download full-size image