1 of 4 Tracing Medical Images Using Multi-Band Watermarks Mingyan Li, Sreeram Narayanan and Radha Poovendran Department of Electrical Engineering, University of Washington email: radha@ee.washington.edu Abstract— The enforcement of Health Insurance Portability and Accountability Act (HIPAA) in April 2003 highlights the determination of federal government to protect the privacy of patients’ records. However, we identify that there is a gap between current security solutions for the privacy protection of medical images and the HIPAA security guidelines: there is no measure to prevent and trace the authorized users who distribute medical images illegally. We evaluate the suitability of some most widely used watermarking techniques for tracing medical images, and demonstrate a need of developing new watermarking schemes that withstand standard medical image processing while maintaining high image quality for diagnostic purpose. We propose a multi-band watermarking scheme, that is robust to low pass filtering and high pass filtering, and yields high perceptual quality. We also present open research problems. Keywords— medical image security, watermarking, pri- vacy, HIPAA I. I NTRODUCTION Privacy protection of medical images has always been an important issue in the management of patients’ medical records. As part of Health Insurance Portability and Account- ability Act (HIPAA), the standards to protect health data privacy issued by the federal government took effect on April 14th 2003 [1]. The HIPAA mandates hospitals, doctors, and other health providers to ensure “confidentiality and integrity of individually identifiable health information, past, present or future.” As digital technology pervades our society, a vast amount of medical images now exists in electronic format for easy stor- age and maintenance. Ubiquitous wired and wireless networks make it possible to access and share data among professional personnel from anywhere, to promote high quality care for patients. The convenience of data access and distribution also poses a great challenge on privacy protection for patients’ information. Constant efforts have been made to provide security solutions [2], [3], [4], [5] to ensure (i) medical image transmission cannot be accessed by unauthorized parties (confidentiality), (ii) received images are not modified during transmission (integrity), and (iii) images are from correct sources to the claimed receivers (authentication). Continuously updated Digital Imaging and COMmunication in medicine (DICOM) standards provide guidelines to ensure authentica- tion, integrity and confidentiality of medical images [2]. However, security measures in DICOM and the research on medical image security [3], [4], [5] do not guarantee privacy of patient data at the recipient end. At present, even if the original data is privacy preserving, the recipient can violate it by giving the images away to unentitled parties. Thus the current privacy mechanisms are not adequate to meet all the HIPAA requirements [1]. In this paper, we make the first attempt to study the problem of tracing illegal distribution of medical images. We use watermarking techniques to trace the authorized person who first distributed the image, on retrieval of images from unauthorized parties. A weaker version of the tracing problem is identifying the ownership of the image itself. Spread Spectrum (SS) scheme [6], Image-Adaptive DCT (IA-DCT) scheme [7], IA Wavelet (IA-W) scheme [7] are among the most widely used watermarking schemes. These techniques survive processing common to multimedia. How- ever, we find that they are unsuitable for medical image, where processing is performed in order to extract information specific to the physician. These processing may be in the form of frequency selective operations like, High Pass Filtering (HPF) and Low Pass Filtering (LPF). Therefore, new watermarking schemes that can withstand these processes and preserve high perceptual quality for diagnostics are required. In this paper, we propose a novel Multi-Band Embedding (MBE) scheme. The algorithm embeds watermarks at different frequency bands in the original image, such that if the pro- cessing removes some components, the rest is detected. The MBE survives HPF and LPF, while preserving high perceptual quality. The rest of this paper is organized as follows: In the next section, we explain how watermarking is performed and detected, and discuss a few common watermarking techniques; Suitability of watermarking to medical images will be ex- plored in Section III and a new watermarking scheme will be proposed; Section IV gives the results of various attacks and their effects on the watermarking schemes; We conclude with possible future directions in Section V. II. WATERMARKING Watermarking is a technique to embed identification codes, called watermarks, into cover media or host media. Water- marking is used for the protection of intellectual property, data integrity, and data authentication [6]. Under the assumption of a unique watermark per image per user, watermarking can be used as fingerprinting. A watermarking system consists of two components: a watermark embedder, and a watermark detector. Watermark embedding is performed in either spatial domain or transform domain. Embedding watermarks in transform domain is shown to be more robust than that in spatial domain [6]. Fig. 1 illustrates a transform domain watermarking system, where a watermark is inserted into the transformed image at positions determined by the embedder. At the de- tection module, original images as well as the watermarks are assumed to be available. To examine the presence of the watermark the detector extracts a watermark and calculates the