ABSTRACT

Digital image watermarking methods are used in medical domain for integrating Electronic Patient Records with medical images. If a watermarking system is to be used for a particular application, there must a standard mechanism for the evaluation of the system. A benchmark for a watermarking system gives a standard mechanism for the evaluation and comparison of watermarking methods. There are no universally accepted performance measures for every watermarking system. Apart from the existing benchmark methods, medical image watermarking exclusively requires evaluation certain factors. This paper discusses the need for a benchmark exclusively for medical image watermarking and the major parameters that arise while designing such a benchmark.

(Note: This Paper was presented in ICSIP., Signalspot Please Download the paper

for proper formatting, images,equations and symbols)

KEY WORDS

Medical image watermarking, Electronic patient record (EPR), Region of Interest (ROI), visual quality, robustness and capacity

1. Introduction

1. Medical image watermarking

Medical information record of a patient is a complex of clinical examinations, diagnosis annotations, prescriptions, histological and other findings, and images in various modalities. In the digital format they are centered in the EPR (Electronic Patient Record). Popularity of internet has become a boon to patients and low capital hospitals to utilize the facility to communicate with the clinicians for the diagnosis of the medical images. The medical images of different modalities can be sent to the clinicians residing any corner of the globe for the diagnosis [4]. The digital handling of EPR on network requires a systematic content validation, which is aimed at quality control: actuality and reliability. Such a scheme preserves confidentiality of the patient data; avoiding detachment of the EPR data from the image.

Another application of medical image watermarking is the integration and storage of medical data with images. Millions of medical images are being produced in various radiology departments of hospitals and research institutes around the world. These are invaluable source of information for medical students, practicing doctor and researchers [7]. A lot of efforts are being made to integrate these images and corresponding information such as first information report about the patient and detailed diagnosis report of the image. Medical staff requires a central medical image service with greater capacity and longer-term storage of digital images for study and education than is generally available. The Hospital Information System contains Integrated Medical Image Database and Retrieval System (INIS) that enables doctors to browse patient images at any time. The INIS allowed CT, MRI, CR, endoscopic and ECG images to be integrated into an image database server with the DICOM standard. Such a system also would need a high level of security, which can be ensured by using digital watermarking.

1.2 Benchmarking

A benchmark gives a standard mechanism for the evaluation and comparison of watermarking methods used for a particular application. Benchmarking procedure involves examining a set of mutually dependent performance factors of the watermarking system. But there are no universally accepted performance measures applicable for every watermarking system [6]. Medical image watermarking requires evaluation of some additional parameters in addition to existing performance measures - visual quality, robustness, capacity and security. This calls for a benchmark exclusively for medical image watermarking methods. The performance measures must include Region of Interest in the medical image as another parameter. The robustness of the system must be checked against all the possible transmission and storage attacks. Rather than performing the evaluation on images of different formats, the medical image format can be confined to the DICOM standard. The EPR diffusion into medical images requires more attention to the capacity of the data hiding without affecting visual quality of the image. The evaluation of imperceptibility of the mark must consider the properties of Human Visual System. The security of the system is dependent on the watermarking key and the performance evaluation of the system must be done by varying the embedding strength and different type of keys. The delay encountered during embedding and recovery of the watermark is also an important factor in telemedicine applications.

2. Requirements of Watermarking in Medical Images

An important factor to be considered while watermarking medical images is that medical images contain Region of Interest (ROI). I.e. regions, which are diagnostically important and the insertion of watermark should not affect the ROI because the distortion in the ROI will affect the diagnosis. It is usual that a medical image is diagnosed before storing the image in the long-term storage, so the significant part of the image called Region of Interest is already determined.

Digital watermarking can imperceptibly embeds messages without changing image size or format. When applied for medical images, the watermarked image can still conform to the DICOM format. Other patient details such as patient ID, age etc. can be inserted into the DICOM header and the diagnosis information can be hidden into the medical image.

One of the major objectives of using medical image watermarking is the memory saving. By attaching the data in the corresponding images will save so much of the storage space. The memory for storage can be saved in HIS by embedding the EPR in the image. So maximum data must be embedded into the medical image. So more importance must be given to the capacity. Rather than testing a large number of images, the evaluation must be performed by varying the embedding strength into a DICOM image. Since the main applications of EPR hiding is in telemedicine and telediagnosis, the robustness measures must deal with the processing operations when transmitted over the web. The watermarked image is kept in the long-term storage of the HIS. So the evaluation of robustness must include the storage noise.

Also in telemedicine applications, large delay during embedding and recovery of the EPR is not acceptable. So time is considered as another parameter. Also, since the aim is to hide the EPR, more focus must be given to imperceptibility and the benchmark should exploit the features of human visual system for the evaluation of imperceptibility. The evaluation should be based on local changes in the watermarked medical image.

Since such a system must perform true extraction of the data, the system must be a blind watermarking system. Insertion includes EPR (watermark) according to the key information. Retrieval must be invertible and only with the presence of watermarked image and key. Present benchmarks perform the tests by inserting the same message into various images. But the proposed system must insert different messages since capacity of embedding is also important. ROI must also be a parameter. Robustness measure should be applied after removing the storage and transmission attacks. Watermark insertion and extraction are according to the key and security level is dependent on the key material and therefore to increase the security level, some amount of randomization must be given to the watermark key.

2.1 Benchmark Parameters

ROI

Region of Interests is diagnostically important parts of the medical image. The ROI is identified by the clinician and the watermark must inserted into regions other than ROI. The parameter used to define the ROI is its percentage area compared to the total area of the medical image.

Imperceptibility

Currently popular metrics for the evaluation of imperceptibility of the watermark are SNR and PSNR, which are based on Mean Squared Error between the cover image and watermarked image.The popular benchmarks – Stirmark and Checkmark - use a single performance measure (PSNR) [2]. These two measures are independent of images and ignore the fact that judging the perceptual quality of image significantly depends on the human observers. So the evaluation of imperceptibility of the watermark must taken into consideration the properties of Human Visual System. According to perceptive model of human vision, signals that have similar components use the same channels from the eye to the cortex. It appears that such signals interact and are subject to non-linear effects. Masking is one of those effects. It occurs when the detection threshold, i.e. the minimum level below which a signal cannot be seen, is increased because of presence of anothersignal. The three performance measures based on the properties of HVS are Weighted PSNR (WPSNR), Just Noticeable Difference (JND) and Watson metric.

WPSNR

WPSNR is Based on the fact that human eye is less sensitive to changes in textured areas than in smooth areas. WPSNR uses an additional parameter called Noise visibility Function (NVF), a texture masking function, as a weight factor. NVF values near zero indicate flat regions, where the watermark should be attenuated, while NVF values near zero indicate texture or edge regions, where the watermark should be amplified. In this way, the watermark is embedded to resist estimation-based attacks derived from image denoising [5].

The contrast sensitivity function (CSF) describes the sensitivity of the HVS to different spatial and temporal frequencies that are present in the visual stimulus. Some image quality metrics include a stage that weights the signal according to this function.

JND

Just Noticeable Difference is referred to as visibility thresholds that are defined as functions of the amplitude of luminance edge in which perturbation is increased until it becomes just discernible. The JND thresholds are image dependent and JND reflects the texture masking property of HVS; the noise is more visible in flat or textureless areas and less visible in regions with edges and textures.

Watson Metric

Watson Metric computes the perceptual error. The perceptual error is computed using the Watson model which takes into account three factors: contrast sensitivity, luminance masking and contrast masking. In Watson model, an 8 X 8 perceptual threshold matrix of DCT transform is generated. The watermarked image is compared block by block with original image using perceptual threshold matrix as reference. The perceptual error is computed as the average of all blocks error.

SSIM

In addition to the above described parameters, another perceptual metric used to model the degradation of medical images when embedding watermark payloads is the Structural Similarity Measure (SSIM). The metric is ideal for testing for similarities in medical images because it focuses on local rather than global image similarity [1]. The SSIM metric compares two images in an overlapping block-wise fashion, using a circular symmetric Gaussian weighting function to reduce blocking artifacts. Each pair of corresponding blocks is compared for luminance, contrast and structural similarity, with the results combined over all blocks to give a similarity measure in the range (0,1). The luminance comparison is a function of corresponding blocks’ mean intensity, and the contrast comparison is a function of corresponding blocks’ standard deviation. The structural comparison is computed as the correlation coefficient of the two blocks [3]. The product of the luminance, contrast and structural comparison functions is then taken as the combined similarity.

Robustness

In medical image watermarking, the extraction of data from the watermarked image is performed only in the presence of the key. Such a blind watermarking scheme uses Bit Error Rate (BER) between the original watermark and extracted watermark as a measure of robustness.

Attacks

Rather than intentional attacks, benchmark must evaluate the performance of the system under storage and transmission attacks. The noise due to long-term storage of the image can be modeled as speckle noise and noise during transmission can be modeled as Gaussian noise [5].

Security

Security of the watermarking system is related to the key used for embedding and recovery of the watermark. Therefore, some amount of randomization must be given to the key.

2.2 Benchmark Architecture

Figure 1 Block Diagram

The diagnosis report is saved in a text file. It is then encrypted into ASCII form and the letters are converted into 7-bit code. Then it is watermarked into areas other than ROI of the medical image using LSB techniques. Before calculating the WPSNR between the cover image and watermarked image, the pixel values in both the images must be converted into the range between (0,1). The perceptual quality values for the watermarked image for varying ROIs are tabulated below,

3. Conclusions

The various factors that can be used for evaluating the performance of medical image watermarking schemes are presented in this paper. The benchmark architecture and dependence of visual quality of the system with the newly introduced parameter ROI is also given.

References

[1] Anthony J Maeder, Birgit M Planitz, Medical Image Watermarking for Multiple Modalities, Proceedings of the 34th Applied Imagery and Pattern Recognition Workshop (AIPR05)

[2] V. Solachidis, A. Tefas, N. Nikolaidis, S. Tsekeridou, A. Nikolaidis, I.Pitas, A benchmarking protocol for watermarking methods, 2001 IEEE Int. Conf. on Image Processing (ICIP’01), pp. 1023-1026, Thessaloniki, Greece, 7-10 October, 2001

[3] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli. Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing, 13(4): 600-612, 2004.

[4] Rajendra Acharya U., P. Subhanna Bhat, Sathish Kumar, Lim Choo Min, “Transmission and storage of medical images with patient information” Journal of Computers in Biology and Medicine, Vol.33, 2003, pp.303-310.

[5] Sviatolsav Voloshynovskiy, Shelby Pereira, and Thierry Pun, Attacks on Digital Watermarks: Classification, Estimation-Based Attacks, and Benchmarks, IEEE Communications Magazine, August 2OO1

[6]M. Kutter and F. A. P. Petitcolas, A fair benchmark for image watermarking systems, Electronic Imaging ’99. Security and Watermarking of Multimedia Contents, vol. 3657, Sans Jose, CA, USA, 25{27 January 1999. The International Society for Optical Engineering.

[7] Pengyu Cao, Masao Hashiba, Kouhei Akazawa, Tomoko Yamakawa, Takayuki Matsuto, An integrated medical image database and retrieval system using a web application server, International Journal of Medical Informatics (2003) 71, 51-55

Attached Files:

• dip012861.doc (1219 KB)