ABSTRACT
We defined, implemented and tested two new methods for irreversible compression of angiocardiographic still images: brightness error limitation (BEL) and pseudo-gradient adaptive brightness and contrast error limitation (PABCEL). The scan path used to compress the digital images is based on the Peano-Hilbert plane-filling curve. The compression methods limit, for each pixel, the brightness errors introduced when approximating the original image (i.e. the difference between the values of corresponding pixels as grey levels). Additional limitations are imposed to the contrast error observed when considering along the scan path consecutive pixels of both the original and the reconstructed image. After previous testing on angiocardiographic images selected as clinically significant from 35 mm films, we enlarged our experiment to a set of 38 coronary angiograms digitally acquired. BEL and PABCEL methods were experimented according to several values of the implied thresholds. Up to a compression ratio of 9:1 for the BEL method and 10:1 for the PABCEL method, no deterioration of the reconstructed images were detected by human observers. After a visual evaluation, we performed a quantitative evaluation. The visualization of pseudo-colour difference images showed the capability of BEL and PABCEL for preserving the most significant clinical details of the original images. For comparison, we applied the JPEG (joint photographic experts group) image-compression standard to the same set of images. In this case, pseudo-colour difference images showed a homogeneous distribution of errors on the image surface. Quantitative compression results obtained by testing the different methods are comparable, but, unlike JPEG, BEL and PABCEL methods allow the user to keep under his direct control the maximum error allowed at each single pixel of the original image. These different behaviors are confirmed by the values obtained for the considered numerical quality quantifiers.
Subject(s)
Angiocardiography , Algorithms , Coronary Disease/diagnostic imaging , Humans , Software DesignABSTRACT
2D-echocardiography is a safe and inexpensive method for accurate left ventricular volume quantification. However, accuracy of measurements is limited by the requirement of geometric assumptions and volume computation algorithms. Reproducibility is diminished by erroneous image plane positioning, which is due to the necessity of standardized reference images. The degree of variation can influence clinical decision making. 3D-echocardiography has overcome those problems. Volume quantitation is performed without the need of geometric assumptions, computation algorithms or reference images. The shape of a ventricle is defined by acquired data, not by calculated data. Since the recent development of sophisticated hardware- and software techniques the method can be utilized under clinical circumstances. For data acquisition tomographic planes are acquired either in a random or in a sequential order. For random acquisition the transducer is positioned at various acoustic windows at the thoracic wall and spatial orientation of the corresponding plane is localized by a spark gap device. By this way various, non-parallel cross-sectional images of the left ventricle are acquired. The spatial position of the planes is displayed by lines in a parasternal longitudinal view, each representing a "line of intersection". When tomographic planes are acquired in a sequential manner, mostly rotational scanning is utilized. Here, multiple apical views of the left ventricle are acquired during ECG- and respiration gated rotation of the transducer over a span of 180 degrees.(ABSTRACT TRUNCATED AT 250 WORDS)
