Content-based compression of mammograms with JPEG2000

This thesis presents three strategies for the content-based compression (CBC) of mammograms. The first strategy is based on a previous CBC approach [1], while the last two strategies utilizing a modified version of the JPEG2000 standard.

Unlike the traditional compression techniques, CBC is comprised of segmentation and compression processes. In this two-step process, the clinically important structures are first identified via a fractal-based segmentation method. Then, a compression strategy is applied in such a way that the extracted structures from the first step are compressed losslessly while the remaining regions are lossily compressed. The first two strategies introduced in this thesis achieve lossless and lossy compression with separate compression engines. On the other hand, the last strategy achieves CBC with a single IPEG2000 compression engine using the max-shift ROI (Regions-of-Interest) coding method.

Preliminary results show that the fractal-based segmentation method covers, on average, over 90% of calcifications. For mammograms with masses, subjective observation shows that over 80% of images will have at least the mass boundaries covered. Also, the two newly proposed CBC strategies can achieve an average compression ratio of 14:1 with PSNR of more than 41dB, while completely preserving the clinically important regions for digitized mammograms. For digital mammograms, the achievable compression ratio is even higher (-20:1) with PSNR greater than 50 dB.