Generalized SDNR analysis based on signal and noise power. (August 2019)

• Record Type: Journal Article
• Authors: Monnin, P.; Gnesin, S.; Verdun, F.R.; Marshall, N.W.
• Is Part Of: Physica medica. Volume 64(2019)
• Page Start: 10
• Page End: 15
• Publication Date: 2019-08
• Subjects: SDNR -- Signal-to-noise ratio -- Image quality -- Quality controls; Medical physics -- Periodicals; Biophysics -- Periodicals; Biophysics -- Periodicals; Imagerie médicale -- Périodiques; Radiothérapie -- Périodiques; Rayons X -- Sécurité -- Mesures -- Périodiques; Physique -- Périodiques; Médecine -- Périodiques; 610.153
• DOI: 10.1016/j.ejmp.2019.06.005
• Languages: English
• ISSNs: 1120-1797

Abstract

Highlights: This work presents a new generalized methodology for assessing the SDNR. The proposed method is valuable for objects of any shape, size and boundary condition. The new method shows increased accuracy for small objects and is less sensitive to image noise. Abstract: The standard approach to signal difference-to-noise ratio (SDNR) analysis requires a region of interest (ROI) positioned within the object to measure signal-difference, restricting this metric to flat-topped objects with large, sharply delineated areas. This work develops a generalized expression for SDNR (SDNRg ) calculated from a ROI encompassing the object. Signal power, defined as the deviation of pixel values from the mean background due to the object, is used instead of signal-difference. Comparison was first made by simulating ideal flat-topped discs with sharp edges and diameters between 1 and 80 pixels, into a uniformly noisy background using a known signal-difference. For discs covering more than 20 pixels, SDNRg and standard SDNR (SDNRst ) were within 3%, while for discs of less than 20 pixels, SDNRg was within 26% of the truth compared to 58% for SDNRst . Generalized and standard SDNR were compared for radiography images of three different phantoms with microcalcification-like objects (MTM-100 phantom), hemispheric objects of different thicknesses with a Gaussian intensity distribution and mammography quality control (QC) images. Applied to Gaussian details, SDNRg was between 20% and 45% higher than SDNRst, depending on object thickness, while for the QC images, SDNRg was with 1.7% of the standard SDNR. Compared to the standard SDNR, SDNRg is applicable to non-uniform signals, where an explicit contrast measurement is not suitable, and has improved accuracy when assessing SDNR of small objects.