The application of wavelets to retinal image compression and its effect on automatic microaneurysm analysis

Compression of radiological images is an effective mechanism for storage and transmission. The use of such images for teleradiology is of increasing importance, with one of the main reasons being the ability to call upon remotely located diagnostic experts. A possibility of future application is the transmission of images to centres that have developed automated diagnostic systems. Whilst many researchers have addressed the problem of how the degradation of image quality with compression ratio affects observer-based diagnostic accuracy, in this work we examine how software performance is altered by image compression. Ground-truth is the labelling of microaneurysms in fluorescein angiograms of the retina, an automatic image analysis task that has already been rigorously compared to expert opinion using uncompressed images. Wavelet and JPEG compression are found to produce opposite trends in detection. With an understanding of the analysis and compression algorithms, a simple model can describe this behaviour. This suggests that software, which is designed to be used reliably on compressed images, will need to be adaptive to compression methodology as well as to compression ratio.