Ultrasound and angio image compression by cosine and wavelet transforms

Abstract

The investigation results for improving lossy compression techniques for ultrasound and angio images are presented. The goal was to determine where the compression process could be improved for the medical application, and to make efforts to improve it. It is proved that the wavelet transform outperforms the discrete cosine transform applied to ultrasound and angio images. A lot of wavelet classes were tried for choosing the best one suited for corresponding image classes, which were characterised by a content complexity criterion. The analysis of international image compression standards was carried out. Special attention was paid to an algorithmical and high level service structure of a new still image compression standard JPEG2000. Its open architecture enables including some wavelet classes which we would like to suggest for medical images. A set of recommendations for acceptable compression ratio for different medical image modalities was developed. It was carried out on the base of compression study performed by the group of angiologists and cardiologists.

Introduction

Images including medical atlases have been playing a crucial role for medical applications. Increasing computer speed, memory capacity, and transmission channels throughput stimulate the development of more sophisticated medical applications which are based on images as a main source of information. Data compression techniques can save data storage and transmission time. ‘Image compression is the art and science for reducing (compressing) the number of bits, required to describe an image’ [1].

There are lossless and lossy image compression techniques. The lossless compression techniques are based on differential pulse code modulation, chain or run-length coding. The lossless compression technique has a low compression ratio (ca. 2:1). An idea to improve it almost twice by operating with a useful field of view, which covers approximately about 40–60% of the image (Fig. 1b and c) was suggested in [2].

The lossy compression technique is based on transforms, which decorrelate pixel values. It can be any kind of transform with given basis functions. For the linear transform, these functions are Karhunen–Loeve, polynomials, trigonometric functions and wavelets. In Section 2, the discrete cosine and wavelet transforms are analysed. The goal was to choose a transform, which achieves the best compression quality for the same compression ratio. The international still image compression standards are reviewed in Section 3. The experimental results of medical image compression with the discrete cosine transform (JPEG) and the wavelet transform with different wavelets are presented and discussed in Section 4. A compression study was performed to evaluate the compression schemes, based on the wavelet transform. Statistical results of this study and evaluation of an acceptable compression ratio in clinical environment are discussed in Section 5.