Reversible and imperceptible watermarking approach for ensuring the integrity and authenticity of brain MR images

The digital medical workflow has many circumstances in which the image data can be manipulated both within the secured Hospital Information Systems (HIS) and outside, as images are viewed, extracted and exchanged. This potentially grows ethical and legal concerns regarding modifying images details that are crucial in medical examinations. Digital watermarking is recognised as a robust technique for enhancing trust within medical imaging by detecting alterations applied to medical images. Despite its efficiency, digital watermarking has not been widely used in medical imaging. Existing watermarking approaches often suffer from validation of their appropriateness to medical domains. Particularly, several research gaps have been identified: (i) essential requirements for the watermarking of medical images are not well defined; (ii) no standard approach can be found in the literature to evaluate the imperceptibility of watermarked images; and (iii) no study has been conducted before to test digital watermarking in a medical imaging workflow. This research aims to investigate digital watermarking to designing, analysing and applying it to medical images to confirm manipulations can be detected and tracked. In addressing these gaps, a number of original contributions have been presented. A new reversible and imperceptible watermarking approach is presented to detect manipulations of brain Magnetic Resonance (MR) images based on Difference Expansion (DE) technique. Experimental results show that the proposed method, whilst fully reversible, can also realise a watermarked image with low degradation for reasonable and controllable embedding capacity. This is fulfilled by encoding the data into smooth regions (blocks that have least differences between their pixels values) inside the Region of Interest (ROI) part of medical images and also through the elimination of the large location map (location of pixels used for encoding the data) required at extraction to retrieve the encoded data. This compares favourably to outcomes reported under current state-of-art techniques in terms of visual image quality of watermarked images. This was also evaluated through conducting a novel visual assessment based on relative Visual Grading Analysis (relative VGA) to define a perceptual threshold in which modifications become noticeable to radiographers. The proposed approach is then integrated into medical systems to verify its validity and applicability in a real application scenario of medical imaging where medical images are generated, exchanged and archived. This enhanced security measure, therefore, enables the detection of image manipulations, by an imperceptible and reversible watermarking approach, that may establish increased trust in the digital medical imaging workflow