A Hybrid Framework to Secure Medical Signals in E-Healthcare

Abstract

In the era of modern healthcare, medical signals, such as Electrocardiogram (ECG) signals, have become increasingly prevalent, prompting a heightened focus on their security using watermarking techniques. As medical facilities transition towards digitized systems for data acquisition and exchange, safeguarding patient information within Electronic Health Records (EHR) has become paramount. Despite regulatory frameworks like Digital Imaging and Communication in Medicine(DICOM) and Health Insurance Portability and Accountability Act(HIPAA), challenges persist regarding ownership conflicts, data security, and privacy, especially post-data retrieval by authorized entities. This gap underscores the necessity for robust data security solutions in healthcare settings. Watermarking algorithms offer a promising avenue to address these challenges, enabling the invisible embedding of secret marks into medical signals to ensure security, authenticity and integrity of sensitive information. By embedding imperceptible information into the images, watermarking enhances security and prevents unauthorized tampering. However, existing watermarking methods often struggle to reconcile the trade-offs between invisibility, watermark capacity, and robustness, resulting in potentially insecure schemes. Therefore, developing secure watermarking techniques explicitly tailored to medical signals, such as ECG, is crucial to uphold the authenticity and privacy of patient information in modern healthcare systems. This thesis aims to introduce enhanced and innovative methods that improve the key factors in ECG Signal watermarking, including imperceptibility, watermark payload, robustness, and security. The thesis has been partitioned into seven chapters. Chapter 1 explores the basic ideas behind an ECG signal and different information security approaches in E-healthcare systems. Several information security measures include data hiding, watermarking, and cryptography. The process of watermarking using ECG signals is explained thoroughly. An overview of watermarking, generic embedding and extraction procedures, various types of watermarking, their characteristics, and different performance parameters. Subsequently, a comprehensive literature assessment is conducted on diverse medical signal-based watermarking methods, encompassing their advantages and constraints.This is followed by a detailed literiii ature review of significant medical signal-based watermarking approaches, including classical methods,optimization-based, encryption-based, and deep learning-based zerowatermarking. We also discussed performance measurements used to evaluate the watermarking technique. Lastly, we highlighted recent research challenges along with a few potential research directions that could fill in the gaps in these domains for researchers and developers. Chapter 2 presents a non-blind watermarking method for ECG signals. The system utilizes the Contourlet Transform (CT) to extract features effectively and employs Multi-resolution Singular Value Decomposition (MSVD) for reliable embedding of watermarks. To improve reliability and safeguard against unwanted access, the watermark is encrypted via encryption using a combination of a chaotic shift transform and a modified Henon map before it is embedded.The experimental findings of this chapter indicate the visual quality and robustness performance analysis over multiple embedding factors ranging from 0.01 to 0.5. PSNR, PRD, KL distance, and NC attain peak values of 71.2313 dB, 0.133, 0.0197, and 0.9997, respectively. Although the proposed method exhibits greater resilience than prior studies, it suffers from an imbalance between reducing noise in the signal and enhancing its ability to store embedded information. To overcome these issues, Chapter 3 presents methods for enhancing the capacity of data (payload) that can be integrated into the ECG signal. Variational Mode Decomposition (VMD) is utilized to achieve efficient noise reduction, enhancing the overall signal quality. In addition, a technique called watermark fusion is presented, which involves merging two medical images to get a single watermark that is more informative. The combined watermark is concealed inside the ECG signal by using a hybrid of Non-subsampled Contourlet Transform (NSCT), and RDWT-MSVD. Examining results yields superior PSNR and NC values for diverse ptbdb ECG signals, precisely 64.239 and 0.9999, respectively. It utilizes conventional methods such as Pan-Tompkins to preprocess the ECG signal, which has restricted capacity for noise filtering and may not eliminate all noise found in the ECG data. Chapter 4 introduces an improved and more sophisticated version of the methods exiv amined in Chapter 3. The Pan-Tompkins++ algorithm is employed to increase the pre-processing of the ECG signal before embedding the watermark, enhancing the signal quality. A Denoising Convolutional Neural Network (DnCNN) guarantees precise watermark retrieval. This deep learning technique efficiently eliminates noise produced during transmission or processing, enabling a more accurate extraction of the encoded information.Upon analysis, the ideal values for PSNR and NC are 83.4865 (alpha = 0.03) and 0.9995 (alpha = 0.1), respectively. Ensuring security is still a top concern, with a chaotic encryption system protecting the watermark. It uses single watermarking, which can limit the hidden information. Also, the VMD technique is used to denoise the ECG signal, but it cannot denoise the ECG signal with complex noise patterns or overlapping IMFs. Chapter 5 explores the concept of dual watermarking, which resolves the problem of single watermarking and provides an enhanced version of VMD, i.e., MVMD. In the above instance, two different watermarks, usually medical images such as a chest X-ray and a PET image, are inserted into the ECG signal. This approach significantly improves the payload capacity compared to single watermarking approaches, which were not fulfilled in the previous chapter. The approach used to achieve robust embedding involves the integration of MSVD with Dual-Tree Complex Wavelet Transforms (DTCWT) and Hessenberg Decomposition (HD). The optimal PSNR attained is 65.6575 dB, correlating with substantial invisibility. The NC values for the extracted watermarks are 0.999378 and around 1, signifying substantial resilience. Although studies on watermarking for biomedical signals have made significant progress in recent years, most cannot efficiently offer the balanced trade-off between invisibility and robustness. Aiming to achieve the optimal balanced trade-off between invisibility and robustness efficiently, Chapter 6 presents an optimization-based dual watermarking technique to safeguard medical images and signals. The Multivariate Variational Mode Decomposition (MVMD) technique performs strong pre-processing on the ECG signal, resulting in optimal signal quality. The FireFly Optimization algorithm calculates the optimal scaling factor for the ECG data, improving its resilience. Dual watermarks are generated from medical images using NSCT and MSVD, as given in Chapter 5. v The combined watermark is subsequently inserted into the pre-processed ECG signal using RDWT-MSVD.The comprehensive objective and comparative evaluation, which reveals a notable enhancement of 99% relative to current methods, substantiates the enhanced embedding capacity of the proposed strategy. In all previous work, we used the traditional watermarking approach. However, Chapter 7 of the thesis investigates a new method that uses deep learning to enhance the security and authentication of ECG signals. This technique employs zero watermarking, in which the watermark does not include any data. On the contrary, the watermarking method involves altering distinct characteristics within the ECG signal. The Continuous Wavelet Transform (CWT) method transforms the ECG signal into scalogram images. Subsequently, the AlexNet deep learning architecture, which has already been trained, is utilized to extract essential features from the scalogram images. As mentioned earlier, the properties undergo additional processing using NSST and HD algorithms to improve security. The zero watermark is generated by applying a Step Space Filling Curve (SSFC) to scramble the original watermark. This scrambled watermark is then incorporated into the scalogram image by performing an XOR operation with the extracted features. This method provides strong authentication capabilities while maintaining the correctness of the original ECG signal. The last chapter contains a comprehensive summary of the entire thesis, including the main findings and future possibilities. A list of research articles and references follows it. This thesis provides an in-depth study of watermarking methods for safeguarding ECG signals. The article’s content illustrates a shift from CT-MSVD-based techniques to more sophisticated approaches that employ deep learning architectures such as AlexNet. These technological improvements offer strong and secure solutions for protecting the completeness and privacy of digital biomedical signals as it is being transmitted and stored in telemedicine applications.