Assuring authenticity of digital mammograms by image watermarking

Wide availability of medical images in digital form introduces risk of changes to the image data, whether by deliberate or accidental means. Some level of protection is afforded by use of encapsulating software such as PACS and other image storage systems, and by making use of file format information (e.g. DICOM header fields). However a greater degree of assurance can be obtained by embedding a subtle hidden pattern within the actual image data, an approach termed image watermarking. The effect of adding the watermark must be below the perceptual threshold of a skilled observer, and must not measurably affect the performance of readers. This paper uses an approach developed specifically for medical image watermarking, and describes efforts to validate its effects on digital mammograms according to the above criteria. The operating region based on embedding strength values for two watermarking methods (DCT and DWT) in the vicinity of the perceptual threshold was characterized using two established perceptual quality metrics (PSNR and SVDP Sum). Subjective testing was then undertaken to compare these metric values with actual observer perception in a 2AFC experiment, and a clear perceptual threshold effect was found.