Effect of Lossy Compression on the
Visibility of Low Contrast Features in Mammography

P M Conmy¹, G J S Parkin², A R Cowen¹, A G Davies³, S M Kengyelics¹

¹FAXiL, Wellcome Wing, The General Infirmary at Leeds, Leeds, UK.

² Dept. of Diagnostic X-ray, The General Infirmary at Leeds, Leeds, UK.

³ Research School of Medicine, University of Leeds, Leeds, UK.

Objectives

To assess the suitability of wavelet compression techniques for digital mammography.

Introduction

Mammography is a particularly demanding field of radiology requiring high resolution, low noise, images for effective diagnosis. Digital acquisition of mammography images at the General Infirmary at Leeds is currently performed with a high resolution Computed Radiography (CR) system, which is based on photostimulable storage phosphor technology. The advantage of such a system, compared with conventional screen-film technology, is that digital image processing can be carried out after the acquisition stage. Image processing can optimise the presentation of images and thus increase their diagnostic efficacy. However, each high resolution mammography image produced is over 8 mega-bytes in size and on average four acquisitions are required for each patient. Retaining these images would, in time, inevitably lead to massive amounts of data requiring high capacity long term storage devices and effective network management systems. In addition, the demand on these systems is soon  likely to increase with the introduction of acquisition devices having even higher sampling rates and greater bit depths.

Data compression techniques reduce the storage capacity required per image and can be used to increase the speed of image transmission over a computer network. “Lossless” compression allows the original image to be restored intact, but only small compression ratios are achievable. “Lossy” compression only allows an approximation of the original image to be restored, but achieves much greater compression ratios. However, even relatively low compression ratios can lead to the degradation of clinically relevant information.

Wavelet techniques offer clinically acceptable image quality even for lossy compression, at much higher ratios compared with conventional techniques such as JPEG compression. This study gauges the suitability of wavelet compression techniques for digital mammography and assesses the achievable compression ratios.

Material and Methods

For this study an AC3 Computed Radiography system was used to acquire several images of a mammography test object (Leeds TOR[MAM]). The test object contains simulated micro-calcifications and low contrast features presented within both featureless and simulated tissue areas. Regions Of Interest (ROI’s) were extracted and varying levels of compression were applied. The extracted areas were presented to multiple viewers for a Two Alternative Forced Choice (2AFC) study on hardcopy films. The viewer is presented with two regions from an acquired image, side by side, one containing only noise and the other containing a feature with noise. The viewer is required to judge which region contains the feature.

Results

Results show that for the visibility of low contrast and micro-calcification details, wavelet compression techniques offer a viable alternative to other conventional techniques and promise the advantage of higher compression ratios.

Conclusion

This study demonstrates the viability of wavelet techniques for the compression of mammography images with compression ratios hitherto unavailable with conventional techniques.

Corresponding Author

P M Conmy,

FAXiL,

Wellcome Wing

The General Infirmary at Leeds

Leeds,

LS1 3EX, UK

Phone: +44-113-3923542

Fax: +44-113-2420926

e-mail: philippa(at)faxil.leeds.ac.uk

---

Oral presentation at EuroPACS'98, Barcelona, Spain

---