Effect of lossy data compression on quantitative coronary measurements

With the accepted use of (lossy) data compression at low compression factors (2, 3 and 4 on the Philips DCI), the question was posed whether higher lossy compression ratios can also be used without statistically affecting the results of quantitative coronary arteriography. In this study the influence of two data compression schemes (LOT and JPEG) at three different compression factors (5, 8 and 12) on coronary measurements was assessed with the Automated Coronary Analysis (ACA) package. A series of 30 original acquired digital images were compressed and decompressed at the different factors, and together with the original non-compressed images processed using the ACA package. In these images a total of 37 obstructed coronary segments were analyzed twice to assess the intraobserver variabilities in the obstruction and reference diameters and in the percent diameter stenosis. The results of the first and second measurements in each image were averaged, and from the differences in corresponding images with different compression ratios, the inter-compression variability was obtained. The results show that the intraobserver systematic errors in the absolute diameters are all small (< 0.07 mm), and statistically not significantly different. The intra-observer random errors for the compressed/decompressed series, however, were all larger (up to 0.21 mm) than for the original series(< 0.13 mm). Statistically significant differences in the intra-observer random errors were found for the JPEG compression scheme at a compression ratio of 5 and for the LOT scheme at a compression ratio of 12. The inter-compression systematic errors in the absolute diameter measurements were also small (< 0.07 mm) and not significant, while the random errors were found to be high in the range between 0.23 mm and 0.31 mm. Given the higher intra-observer variabilities for the compressed/decompressed image series as compared to original images, and the fact that all inter-compression variabilities were found to be so high, we must conclude that the higher compression ratios affect the results of QCA in a negative sense. In conclusion, the use of lossy data compression with JPEG or LOT compression schemes al ratios 5, 8 and 12 must be discouraged for QCA.