Image texture analysis quantification of concentric ring artifacts in quality assurance scans

Background: In the course of evaluating transaxial reconstructions of SPECT phantom data to assess camera capability, the appearance of conspicuous concentric ring artifacts will trigger retuning conventional rotating SPECT detectors, as they are evidence of inadequate uniformity corrections. However, these evaluations are visual, not quantitative, & no criteria have been established for what constitutes a sufficiently severe artifact to necessitate retuning. Our study was undertaken to determine the degree to which observers agree about the presence & severity of SPECT concentric ring artifacts, & to test whether any texture analysis metrics correspond to significant artifacts.

Methods: All test data were acquired as part of routine quarterly quality assurance using standardized SPECT phantoms that included sections of solid spheres, solid rods, & uniform volumes, loaded with 740 MBq ^99mTc for 30-34x10⁶ counts, reconstructed by filtered backprojection with a Hanning (cutoff = 1.0 cm^-1) post-filter & Chang attenuation correction (µ = 0.11 cm^-1). From a collection of over 200 acquisitions performed on 12 different SPECT systems, 40 were identified as either having evidence of concentric ring artifacts, or were acquired in order to assess whether artifacts were resolved following camera retuning performed within 1 wk of obtaining an unacceptably non-uniform result. Transaxial reconstructions were reviewed at one outside institution to obtain independent readings, & underwent texture analysis at another outside institution. Manual volumes of interest were created in the uniform sections of each set of phantom data, within which counts were tabulated, from which were computed 71 texture analysis metrics, including “radial contrast,” derived from the radial profile of summed slices transformed into polar coordinates, & “radial signal-to-noise ratio (SNR).” Two experienced medical physicists independently graded severity of artifacts on a 5-point scale (0 = “no artifact” to 4= “severe artifact requiring camera retuning”), & 1 observer rescored images blinded to his initial scores. Readers had no knowledge of the texture analysis metrics values.

Results: Among the 40 phantoms, artifacts were considered sufficiently severe to require camera retuning in 10 rods sections, 17 sphere sections, & 16 uniform sections. In uniform sections, there was “good agreement” for inter-observer & intra-rater assessments (κ = 0.66, Fisher exact p < 0.0001 & κ = 0.61, Fisher exact p = 0.001, respectively). While several texture analysis metrics agreed significantly (p < 0.05) by ROC analysis with visual detection of significant artifacts in uniform sections, the most strongly correlated metric was radial contrast > 4.75% (ROC AUC accuracy = 88±5%, sensitivity = 83%, specificity = 83%, p < 0.0001), for which values were significantly different for the phantoms without artifacts versus those with severe artifacts (3.8±1.0% versus 5.5±1.0%, p < 0.0001), with increasing magnitude of the metric significantly correlated with increasingly severe artifacts (rank correlation ρ = 0.72, p < 0.0001). Radial SNR was equally associated with the visual impression of significant concentric ring artifacts (ROC AUC accuracy = 87±5%, sensitivity = 83%, specificity = 83%, p < 0.0001), with lower values for phantoms without artifacts than with artifacts (2.1±0.7 versus 3.3±0.7, p < 0.0001), & significant correlation with increasing artifact severity (rank correlation ρ = 0.71, p < 0.0001).

Conclusions: There is good agreement among observers, with reproducible results, as to the presence of circular ring artifacts in uniform sections of SPECT quality assurance scans. The fact that some texture analysis metrics agree well with visual impressions in uniform sections suggest that further work is warranted to enable texture analysis applications to non-uniform phantom areas containing spheres & rods, where artifacts also can appear.