PT - JOURNAL ARTICLE AU - Kenneth Nichols AU - Christopher Palestro TI - Image texture analysis to verify detection of sub-centimeter <sup>18</sup>F-FDG lesions DP - 2019 May 01 TA - Journal of Nuclear Medicine PG - 62--62 VI - 60 IP - supplement 1 4099 - http://jnm.snmjournals.org/content/60/supplement_1/62.short 4100 - http://jnm.snmjournals.org/content/60/supplement_1/62.full SO - J Nucl Med2019 May 01; 60 AB - 62Purpose: In evaluating 18F-FDG PET &amp; PET/CT scans physicians rely on SUV values &amp; also on subjective visual impressions of the presence of small lesions. The criteria for the latter may vary among observers. Our investigation used physical phantom simulations to determine whether any image texture analysis metric reliably corresponds to visual criteria used to identify lesions, &amp; whether any metric differentiates counts of background regions from those of sub-centimeter simulated lesions. Methods: Routinely collected quarterly quality assurance test data were processed retrospectively for 13 different 18F-FDG PET scans performed of standardized phantoms for 4 different PET/CT systems. Phantoms included 4 cylinders ranging in size from 8 mm - 25 mm embedded in a cylindrical water bath. Cylinders were loaded with an activity concentration of 6 kBq/mL while background was loaded with 3 kBq/mL, to be consistent with counts acquired for typical whole body PET protocols for a 70 Kg man injected with 370 GBq 18F-FDG. Algorithms were written to automatically isolate regions containing cylinders, sample counts in cylinder regions &amp; within similarly-sized multiple background regions, &amp; to compute several classes of image metrics: quantile curve metrics derived from plotting quantiles of counts in cylinder regions against those in background regions, image texture analysis gray-level co-occurrence matrix metrics, image contrast metrics derived from polynomial fits of counts-versus-radii curves, &amp; count histogram metrics that included signal-to-background (SBR) measured as SBR = ((mean cylinder counts - mean background counts)/mean standard deviation of multiple background count samples). For qualitative image scores, a physicist, who had no knowledge of quantified texture analysis metrics values, independently graded cylinder visibility on a 5-level scale (0 = definitely not visible to 4=definitely visible) &amp; assigned dichotomous visibility scores. Results: The 3 largest cylinders were visible in 100% of cases with mean visibility score of 3.4±1.3, while the smallest 8-mm cylinder was visible in 38% of cases with significantly lower mean visibility score of 1.0±1.0 (p &lt; 0.0001). By ROC analysis the metric with the greatest area under curve (AUC) for dichotomous cylinder visibility of all cylinders was count quantile plot slope with AUC = 94±3%, sensitivity = 93% &amp; specificity = 90%, for which the threshold for discrimination was count quantile plot slope &gt; 1.69. The metric that agreed most closely with the observer’s dichotomous visibility readings for the 8-mm cylinders alone was image contrast with AUC = 70±16%; SBR was more accurate at discriminating background regions from those containing 8-mm cylinders with AUC, sensitivity &amp; specificity all = 100%, for which the threshold for discrimination was SBR &gt; 2%. Conclusion: Count quantile plots connect visual impression of simulated lesion visibility to quantified image metrics &amp; count histogram metrics aid in verifying the presence of sub-centimeter simulated lesions. Therefore, image texture analysis metrics are potentially useful for 18F-FDG PET/CT studies.