Wavelet-based, inhomogeneous, near-lossless compression of ultrasound images of the heart

With hospitals switching from film to digital storage of medical images, there is an ever increasing need for efficient image compression techniques. This paper presents a new approach to segmented coding of ultrasound images of the heart for inhomogeneous spatial reconstruction quality. The authors' method employs a geometric modelling technique to detect the useful image area from the background. Based on these results, they make use of a hybrid segmentation technique to isolate the anatomical structures of the heart in the presence of noise. Segmented structures are assigned importance according to their diagnostic content and compression is performed using the wavelet transform and entropy coding. The authors select the most important wavelet coefficients across scales. This involves the rejection of noise related coefficients and emphasizing on those carrying the bulk of the diagnostic information in the original image.