A new approach of ECG steganography and prediction using deep learning
Introduction
The electrocardiogram (ECG) signal is a measure of activity of human heart, which periodically contracts and relaxes, to maintain flow of blood, within the body [1]. Any abnormality in functionality of heart reflects as undesirable characteristics on the general pattern of ECG. Also, depending upon the age, gender, diseases and lifestyle, ECG characteristics vary person to person. Hence, while recording and evaluating this signal, doctors and physicians should be knowledgeable about the personal information of the respective patient. Thus, while storing and transmitting any ECG record, patient’s personal information is also needed to be stored and sent. However, according to the rules of Health Insurance Portability and Accountability Act (HIPAA) of 1996, which focuses on data privacy; patient’s personal information should be secured from cyber-attack. ECG steganography is an intelligent approach of embedding (encrypting) patient’s information within ECG signal in such a way that, these data cannot be understood by naked eyes, except specified personnel [2], who can decrypt those hidden information using reverse algorithm of encryption. Thus the host ECG signal acts as a ‘cover carrier’, within which patient’s personal information is securely kept, thus ensuring data privacy and authenticity.
While, implementing steganography, it should be kept in mind that after encrypting hidden data (stego-entity), the modified host signal (steganofied signal), should not possess any visible distortion with respect to main signal, as well as data should not be decrypted easily except by the intended users. Time domain approach of steganography e.g. additive approach of data concealing in transformed domain coefficients provided a good detectability rate [3]. ‘Least Significant Bit (LSB)’ technique performs replacement of LSB of host data by secret bits [4], resulting in minimal distortion in host data [5]. Liu et al. proposed [6] an adaptive pixel pair matching (APPM) technique to insert secret bits in pixel value differences of cover image.
Transformed domain techniques have some drawbacks like more computational complexity, higher execution time etc., but they are more popular than spatial domain processes in the view of data security and privacy preservation. Discrete fourier transform (DFT), discrete cosine transform (DCT) etc. are some transformation algorithms that convert host signal into frequency domain coefficients, within which steganography is performed [7]. Discrete wavelet transform (DWT) provided an improved result on steganography over DFT, DCT etc. techniques by concealing confidential data within time-frequency subbands of host data assuring better security. A. Ibaida et al. [8] proposed a scrambling matrix and share key for steganography within various subbands of ECG signal by implementing singular value decomposition (SVD). Dynamic approach of replacement of singular values of SVD, provided an improved result over direct replacement algorithm in [9]. B. Lei et al. [10] incorporated recursive dither modulation (RDM), differential evolution (DE) and quantization steps (QSs) to control watermarking strength on medical images. S. Chen et al. [11] used quantization based digital watermarking on DWT, DCT, DFT coefficients. A. Tareef et al. [12] introduced sparse coding along with DWT-SVD, for ownership authentication and prevention from unauthorized access and false identification. C. Liji et al. [13] utilized integer-to-integer wavelet transform (IWT) and LSB localization technique that provided better result over DWT. To improve the robustness and imperceptibility of steganography by optimizing the scaling factor of SVD on DWT co-efficients, ant colony optimization (ACO) was used in [14].
Curvelet transform (CT) provided a satisfactory result over WT by eliminating the limitation of multiscale representation and defining curved edges in a better manner. Therefore using an adaptive thresholding on CT, S. Jero et al. [15] introduced a new ECG steganography technique. Also, they achieved a better performance using quantization approach [16], by inserting confidential data on curvelet coefficients that have nearly zero value.
R. Thabit et al. [17] used scarlet transformation for converting original image to non-overlapping blocks, and by histogram modification, secret data were embedded by modifying the mean value of subband. C. Yang et al. [18] proposed a reversible lossy steganography algorithm on ECG, followed by coefficient alignment, to preserve the quality of ECG after bit extraction. To classify and predict the disease risk, Na¨ıve Bayesian Classification was implemented in [19] for privacy preserving of patient’s confidential data, so that service provider can detect disease without disclosing medical data. H. Shiu et al. [20] introduced a blind and reversible ECG and EMG steganography, based on hamming code and shared key for information privacy. To overcome concealing capacity of secret data, fast Walsh–Hadamard transform (FWHT) was used in [21]. In recent time, P. Augustyniak et al. proposed a new steganography approach by replacing the baseline noise of 12-lead ECG data, by secret data that best fit the actual noise [22].
Now, while evaluating the ECG, presence of any abnormal beat describes the type of disease, thus assisting in diagnoses. But encryption employed on both abnormal and normal beats, results in distortion of dominant characteristics. Therefore, characterization and detection of abnormal beats are necessary by feature detection, [[23], [24], [25]]. In addition, while implementing steganography, entire beat is modified [10], thus important characteristics points are manipulated. Another disadvantage of existing approaches is that modification performed during encryption are kept permanently, thus some attributes are lost.
Therefore, in this work, at first, beat detection was performed using derivative based approach, which is an effective and time saving algorithm of R-peak detection [[26], [27], [28]], and then data encryption was done within TP-segments, which has very low clinical importance with respect to other segments viz. P-wave, QRS complex, T-wave etc. In order to keep T-wave offset and P-wave onset unaffected, a small section within each TP-segment was modified such that the boundaries of that section may not harm actual onset and offset. A relative abnormality was measured between successive segments, and the abnormal segments were kept unchanged, while encryption and this classification was performed using feature detection. Then data were encrypted using a new approach, in time domain, i.e. within raw ECG signal. Finally, after decrypting patient’s data, those segments, which were manipulated during encryption, were predicted to their original values using LSTM RNN. The internal memory system of LSTM RNN kept tracking the features of previous segments, which helped in predicting the upcoming pattern of succussive segments. In this way, the focus of this work was to keep the main attributes of ECG almost unchanged as well as to predict the modified signal after decryption. Thus the main contributions of this proposed method are, 1) ECG steganography within relatively normal TP-segments, 2) introduction of new data encryption algorithm within time-domain signal, 3) modified segment prediction using LSTM RNN after secret data extraction.
The rest of this paper is organized as follows, Section 2 describes the method used for data encryption and decryption, Section 3 shows the experimental result of this algorithm, implemented on various ECG signals, and compares it with other techniques, and finally Section 4 concludes this paper.
Section snippets
Materials and methods
The main focus of this work was to keep important features within ECG almost unchanged, by modifying only TP-segments after encryption and to predict those segments using a trained LSTM RNN, after decryption. Fig. 1 shows the general structure of ECG beats and it is clear that the TP-segment (iso-symmetric line) carries very low characteristics feature within ECG. A ‘shared key’ is used during both encryption and decryption that contains four elements expressed as [K1, K2, K3, and K4]. The
Results and discussion
This steganography algorithm was implemented on physionet [36] records, among which 48 were of mitdb record and 547 were of ptbdb record and 90 records in European ST-T database. As TP-segment detection was performed using a heuristically derived Eq. (2), sensitivity analysis was performed upon 222 records (among them, total 130, 40 and 52 records were from ptbdb, mitdb and European ST-T database respectively, including training data mentioned in Section 2.1.2) for T-wave offset and P-wave
Conclusion
In this work, a new ECG steganography technique of hiding patient’s confidential information is proposed. Unlike the existing techniques, which utilize the entire ECG beat for steganography; this algorithm utilizes only the TP-segment, which has very low clinical importance. Thus in this technique, important characteristics features of a beat were almost unaffected during encryption as well as decryption. In addition, in this work, a LSTM RNN network was used, to predict the modified samples of
Funding sources
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
CRediT authorship contribution statement
Soumyendu Banerjee: Writing - original draft, Data curation, Methodology, Software, Validation. Girish Kumar Singh: Conceptualization, Visualization, Investigation, Supervision, Writing - review & editing.
Declaration of Competing Interest
The authors report no declarations of interest.
References (36)
- et al.
Experimental studies of LSB watermarking with different noise
Proc. Comp. Sci.
(2015) - et al.
Reversible watermarking scheme for medical image based on differential evolution
Exp. Sys. Appl.
(2014) - et al.
A highly secure oblivious sparse coding-based watermarking system for ownership verification
Exp. Sys. Appl.
(2015) - et al.
Integer-to-integer wavelet transform based ECG steganography for securing patient confidential information
Proc. Tech.
(2016) - et al.
Robust reversible watermarking scheme using Slantlet transform matrix
J. Sys. Soft.
(2014) - et al.
Preserving privacy of online digital physiological signals using blind and reversible steganography
Comp. Meth. Prog. Biom.
(2017) - et al.
Delineation of ECG characteristic features using multiresolution wavelet analysis method
Meas. J. Int. Meas. Confed.
(2012) - et al.
Recommendations for the standardization and interpretation of the electrocardiogram
Heart Rhythm
(2007) - et al.
An automatic r and t peak detection method based on the combination of hierarchical clustering and discrete wavelet transform
IEEE J. Biomed. Heal. Inform.
(2020) - et al.
ECG Acquisition and Automated Remote Processing
(2014)
Steganography in Digital Media: Principles, Algorithms, and Applications
An additive approach to transform-domain information hiding and optimum detection structure
IEEE Trans. Multi.
Detection of hiding in the least significant bit
IEEE Trans. Sig. Proc.
A secure steganography for privacy protection in healthcare system
J. Med. Sys.
Perceptual DFT watermarking with improved detection and robustness to geometrical distortions
IEEE Trans. Inform. Foren. Secu.
Wavelet-based ECG steganography for protecting patient confidential information in point-of-care systems
IEEE Trans. Biomed. Eng.
Discrete wavelet transform and singular value decomposition based ECG steganography for secured patient information transmission
J. Med. Sys.
Hiding patients confidential data in the ECG signal viaa transform-domain quantization scheme
J. Med. Sys.
Cited by (24)
ROSEmark: Robust semi-blind ECG watermarking scheme using SWT-DCT framework
2022, Digital Signal Processing: A Review JournalAgent-based beat-by-beat compression of 12-lead electrocardiogram signal using adaptive Fourier decomposition
2022, Biomedical Signal Processing and ControlA new real-time lossless data compression algorithm for ECG and PPG signals
2022, Biomedical Signal Processing and ControlCitation Excerpt :In tele-monitoring applications, Mukhopadhyay et al. proposed ASCII character coding, to compress PPG data [20]. In recent works, steganography [21] along with PPG compression using ASCII coding and SVD was proposed in [22]. This paper achieved high CR, but at the cost of comparatively higher PRD.
An automatic tumour growth prediction based segmentation using full resolution convolutional network for brain tumour
2022, Biomedical Signal Processing and ControlCitation Excerpt :Semi-automatic approaches are deformable models and they need user interaction to segment the brain tumour [10]. Because MRI data is a highly noisy signal, it must constantly be filtered to remove high frequency disturbances before it can be processed [11–13]. The automatic brain tumour segmentation approaches need hand crafted features.
Effective compression and classification of ECG arrhythmia by singular value decomposition
2021, Biomedical Engineering AdvancesDeep neural network based missing data prediction of electrocardiogram signal using multiagent reinforcement learning
2021, Biomedical Signal Processing and Control