Secure Patient Data Transfer Using Information Embedding and Hyperchaos

Health 4.0 is an extension of the Industry standard 4.0 which is aimed at the virtualization of health-care services. It employs core technologies and services for integrated management of electronic health records (EHRs), captured through various sensors. The EHR is processed and transmitted to distant experts for better diagnosis and improved healthcare delivery. However, for the successful implementation of Heath 4.0 many challenges do exist. One of the critical issues that needs attention is the security of EHRs in smart health systems. In this work, we have developed a new interpolation scheme capable of providing better quality cover media and supporting reversible EHR embedding. The scheme provides a double layer of security to the EHR by firstly using hyperchaos to encrypt the EHR. The encrypted EHR is reversibly embedded in the cover images produced by the proposed interpolation scheme. The proposed interpolation module has been found to provide better quality interpolated images. The proposed system provides an average peak signal to noise ratio (PSNR) of 52.38 dB for a high payload of 0.75 bits per pixel. In addition to embedding EHR, a fragile watermark (WM) is also encrypted using the hyperchaos embedded into the cover image for tamper detection and authentication of the received EHR. Experimental investigations reveal that our scheme provides improved performance for high contrast medical images (MI) when compared to various techniques for evaluation parameters like imperceptibility, reversibility, payload, and computational complexity. Given the attributes of the scheme, it can be used for enhancing the security of EHR in health 4.0.


Introduction
Technology development is changing the world scenario thick and fast. The multimedia and Internet are exponentially changing and revolutionizing all spheres of life. The present generation systems are smarter, self-organized, well-connected, interoperable, decentralized, and flexible. This has been made possible by adapting cyber-physical systems to monitor and control the machines while using the Internet of Things (IoT) for proper management of data and various system components [1][2][3]. It is aimed at progressive virtualization to enable the personalization of healthcare by catering to health care needs electronically. This is known as the Health 4.0 cyber-physical system (HCPS). Personalized healthcare delivery is being achieved by the efficient implementation of core technologies like artificial intelligence, cyber-physical systems, IoT-based architectures, and the evolution of 5G-based networked mobile communication [4][5][6]. The main design principles of Health 4.0 include virtualization, decentralization, real-time capability, modularity, and service orientation. It aims to fully replace conventional healthcare with electronic smart healthcare to allow personalized monitoring and diagnosis [7][8][9][10]. The Indian healthcare industry is growing exponentially and is expected to be at USD280 billion industry by 2020. Though to date, the healthcare industry is fragmented but certain super-specialty chains uses the DE scheme. This method provides enhanced embedding capacity (EC) in addition to the reduced computational complexity compared to some existing techniques [28,29]. The histogram bin shifting based schemes have also been proposed where the lowest or the highest bin is shifted to embed bit '1' or a bit '0' [31][32][33][34][35].
The first IP-based reversible information embedding (IRIE) scheme was proposed by Jung and Yoo with improved capacity and image quality using the method of Neighbor-Mean-Interpolation (NMI) [36]. In [37], another reversible watermarking technique has been presented by Luo et al., where reversibility is accomplished using neighboring pixels. Their scheme properly manages the overflow and underflow conditions near the boundary. Abadi et al. [38] have developed a histogram profile shifting (HS) based method utilizing the pixels at the boundary to improve capacity. An MI-based RDH method has been proposed by Naheed et al. [39]. Lee and Huang [40] applied IP on neighboring pixels (INP) for RDH. Their reported technique improves upon the DH method developed by Jung and Yoo [36]. In their method, the maximum intensity is used to increase the contrast value calculated between adjacent values of the pixels. Furthermore, Lee et al. [41] also offered a scheme using a maximum difference calculated between adjacent pixels. Their reported scheme generates SI of reduced quality. Tang et al. [42] have presented another data embedding scheme with improved payload (PL) capability. Arsalan et al. [43] proposed a method based on the genetic algorithm (GA) and companding schemes that offer low information EC. A reversible secret DH technique for MI is demonstrated by Naheed et al. [44]. They create a balance in the embedding proportions and imperceptibility. However, it is seen that the use of PSO and GA renders high computational complexity with relatively low EC.
Wang et al. [45] describe a reversible HS-based DH technique. The IP error in the wall pixels is used to increase embedding capacity. Abdul Wahed et al. [46] presented an improved local dispersion-based IP scheme and analyzed the embedding performance for medical images. They also presented another IP scheme known as quadratic interpolation for RDH. Their scheme achieves a high EC with better SI-quality [47]. Chang et al. have proposed an RDH method for image IP based on the Sudoku matrix [48]. In this scheme, the distortion of SI is reduced by embedding secret bits in a pair of pixels. Their scheme also achieves a higher EC along with good quality for SI when compared to some of the previous research works. Mathew et al. presented an IP-based RDH for encrypted images [49]. Their illustrated scheme efficiently utilizes the reversible hiding schemes for unencrypted images to maximize the EC. The algorithm outperforms most of the contemporary DH approaches.
Robustness to various signal processing attacks is of vital importance, in addition to the security of embedded data in medical images. Some of the recent work with high robustness to various attacks could be seen in [50,51]. In [50] a method for the secure transmission of medical images has been presented. The scheme uses integer wavelet transform and chaotic sequences. The scheme, though robust, is computationally complex, as it has been implemented in the transform domain. Also, the scheme supports a very small payload of 0.18 bpp. A multilevel discrete wavelet transform and singular value decomposition-based scheme for medical has been reported in [51]. This scheme is highly robust for hiding watermarks, which could be patient information. The transform domain implementation of this scheme makes it computationally complex. The security of the embedded data has been ensured utilizing chaos theory in both cases, as chaotic maps are very efficient for image security [52].
In this work, we have presented an efficient scheme that can hide the encrypted EHR in an interpolated CI developed from a seed image. The proposed IP algorithm generates a better perceptual model of a CI from seed image compared to many contemporary schemes. We use the generated CI for embedding EHR information. For encryption of the EHR hyperchaos (HC) based technique is employed because of the advantages like ergodicity, a high value of randomness, very intense dependence on initial conditions and parameters [53]. HC systems generate strong encryption algorithms because of their highly aperiodic behavior. The chaos-based ciphers provide robust confusion and diffusion of the input data by the process of pixel transposition and value substitution. Pertinent to mention, though 1-D chaos-based algorithms are simple to implement but have disadvantages like less robustness, simpler dynamical nature, and reduced keyspace [54][55][56][57][58][59][60][61][62][63][64]. On the other hand, the HC systems have complex dynamical behavior, better key sensitivity, and larger keyspace. Hyperchaotic systems have two positive Lyapunov exponents and are therefore highly unpredictable. These strengths offered by the HC system have led to the development of many improved algorithms [65].
Kun Zhan et al. have illustrated an HC-based system for image cryptography [66]. Their proposed scheme aims to apply the pseudo-random sequence (PRN) sequence developed by the HC system to the underlying processes of the encryption algorithm. In this paper, we also apply the 4-D HC system to scramble the WM logo embedded to authenticate the transferred content and the EHR of the patient. As the HC system has high key sensitivity and any slight change in the initial conditions will lead to a different sequence being generated, which provides high security to the EHR data.

Proposed Technique
The presented reversible EHR embedding based on an HC encryption and IP algorithm has been discussed in this section. An image IP algorithm is used to achieve the reversibility of the actual up-sampled image. This technique can be efficiently used to construct a CI from the original (ORG) clinical image that can be represented by Equations (1)- (4). The proposed technique is illustrated in the following sub-sections. Section 3.1 describes the detailed process of the IP technique for CI formation. The IP technique can provide high embedding capacity along with the generation of better visual quality. Section 3.2 describes in detail Chen's HC system employed to generate the PRN sequence. A PRN sequence is generated and is utilized to encrypt the EHR data and the authentication logo. Both the EHR and the fragile WM (logo) are implanted into the interpolated CI. The EHR and WM bits are embedded with the help of the spatial domain algorithm called the least significant bit (LSB) substitution technique. Further, Sections 3.3 and 3.4 respectively deal with the process of data embedding and data reconstruction.

Improved IP Scheme
The detailed sequence of operations for the presented method with all the underlying operations is shown in Figure 1. highly aperiodic behavior. The chaos-based ciphers provide robust confusion and diffusion of the input data by the process of pixel transposition and value substitution. Pertinent to mention, though 1-D chaos-based algorithms are simple to implement but have disadvantages like less robustness, simpler dynamical nature, and reduced keyspace [54][55][56][57][58][59][60][61][62][63][64]. On the other hand, the HC systems have complex dynamical behavior, better key sensitivity, and larger keyspace. Hyperchaotic systems have two positive Lyapunov exponents and are therefore highly unpredictable. These strengths offered by the HC system have led to the development of many improved algorithms [65].
Kun Zhan et al. have illustrated an HC-based system for image cryptography [66]. Their proposed scheme aims to apply the pseudo-random sequence (PRN) sequence developed by the HC system to the underlying processes of the encryption algorithm. In this paper, we also apply the 4-D HC system to scramble the WM logo embedded to authenticate the transferred content and the EHR of the patient. As the HC system has high key sensitivity and any slight change in the initial conditions will lead to a different sequence being generated, which provides high security to the EHR data.

Proposed Technique
The presented reversible EHR embedding based on an HC encryption and IP algorithm has been discussed in this section. An image IP algorithm is used to achieve the reversibility of the actual up-sampled image. This technique can be efficiently used to construct a CI from the original (ORG) clinical image that can be represented by Equations (1)- (4). The proposed technique is illustrated in the following sub-sections. Section 3.1 describes the detailed process of the IP technique for CI formation. The IP technique can provide high embedding capacity along with the generation of better visual quality. Section 3.2 describes in detail Chen's HC system employed to generate the PRN sequence. A PRN sequence is generated and is utilized to encrypt the EHR data and the authentication logo. Both the EHR and the fragile WM (logo) are implanted into the interpolated CI. The EHR and WM bits are embedded with the help of the spatial domain algorithm called the least significant bit (LSB) substitution technique. Further, Sections 3.4 and 3.5 respectively deal with the process of data embedding and data reconstruction.

Improved IP Scheme
The detailed sequence of operations for the presented method with all the underlying operations is shown in Figure 1.  into the CI-(I1). The process of I1 generation from the original image is illustrated in the subsequent steps:

•
One-pixel I1(k, l) of the CI is equivalent to I(k, l) of the ORG image (I).
The following equations can be utilized to scale up the image: The pixel I1(k, l) is the seed pixel also known as the seed/pivot pixel. With the help of the above-mentioned equations, all the non-seed pixels of the interpolated CI can be generated. The CI has very good quality as compared to the input image (O) which is indicated by the high PSNR value between the two. Figure 2 explains the method by which the resized image is up-sampled to form the CI for embedding. We calculate the pixel values I1(k, l + 1), I1 (k + 1, l), and I1 (k + 1, l + 1) by making use of Equations (1)-(4).
The following equations can be utilized to scale up the image: The pixel I1(k, l) is the seed pixel also known as the seed/pivot pixel. With the help of the above-mentioned equations, all the non-seed pixels of the interpolated CI can be generated. The CI has very good quality as compared to the input image (O) which is indicated by the high PSNR value between the two. Figure 2 explains the method by which the resized image is up-sampled to form the CI for embedding.

Hypechaotic (HC) Encryption
Hyperchaos arises when a high-dimensional non-linear system has more than two positive Lyapunov exponents. HC systems have very complex dynamical behavior and are highly secure and unpredictable. They depend upon many parameters and have a greater number of initial conditions, which increases the key-space of the encryption algorithm. In this paper, we employ the 4-D Chen's HC system to strengthen the protection of the EHR data and WM bits embedded inside the MI. Chen's 4-D HC system has a set of four nonlinear equations indicated in equation set (5) which governs its behavior. The system is in HC state for the value of control parameters as a = 35, b = 3, e = 35, t = 5, 1 = 1, 2 = 0.2, 3 = 0.3. The initial conditions are 1 (1, 1) = 0.12, 2 (1, 1) = 0.23, 3 (1, 1) = 0.34, and 4 (1, 1) = 0.45 [66]. The HC sequence has to be pre-iterated several times to remove the initial adverse effects and the value of time set 'h' is taken as 0.001.
For the HC system, the state variables 1 , 2 , 3 and 4 are calculated for the M × N iterations of the system and two key sequences ( ) and ( ) will be formed, where 'j' represents the index of operation. ( ) and ( ) values are concatenated to develop a sequence which is represented by Equation (6)

Hypechaotic (HC) Encryption
Hyperchaos arises when a high-dimensional non-linear system has more than two positive Lyapunov exponents. HC systems have very complex dynamical behavior and are highly secure and unpredictable. They depend upon many parameters and have a greater number of initial conditions, which increases the key-space of the encryption algorithm. In this paper, we employ the 4-D Chen's HC system to strengthen the protection of the EHR data and WM bits embedded inside the MI. Chen's 4-D HC system has a set of four non-linear equations indicated in equation set (5) which governs its behavior. The system is in HC state for the value of control parameters as a = 35, b = 3, e = 35, t = 5, l 1 = 1, l 2 = 0.2, l 3 = 0.3. The initial conditions are s 1 (1, 1) = 0.12, s 2 (1, 1) = 0.23, s 3 (1, 1) = 0.34, and s 4 (1, 1) = 0.45 [66]. The HC sequence has to be pre-iterated several times to remove the initial adverse effects and the value of time set 'h' is taken as 0.001.  The PRN sequence 'k' is formed by joining the S j sequences as shown in Equation (7): The values of the sequence 'k' are XORed with the binary watermark and the EHR data which is to be embedded. The receiver needs to provide the same set of input keys which include the initial values and control parameters of the HC system to recover the authentication logo and the encrypted EHR data.

Data Embedding
We illustrate the method of implanting the Electronic Health Record (EHR) and the WM. As an initial step, the cover image pixels are distinguished as either seed pixels (SP) and non-seed pixels or data containing pixels (DCP). To assist in the proper reversibility only the non-seed pixels have been utilized for encrypted EHR embedding and no embedding is performed using seed pixels. Furthermore, an encrypted fragile WM has also been put in the cover media. The data vector that is thus implanted in the CI comprises of EHR bits and watermark bits (WM) and has been obtained by concatenating the individual vectors as: Data vector = [WM: EHR] The embedding of a fragile watermark helps detect any possible interference with the EHR. The proposed method is described as Algorithm 1. Algorithm 1. Image IP, EHR, and WM Encryption using HC, and LSB embedding of data bits in DCP Input: Grayscale M × N Secret Image, a hyperchaotic map with the initial values s1, s2, s3, s4 and control parameters a, b, e, t, l1, l2, l3. Output: Stego Image (SI) of size M × N BEGIN

Data Extraction
In this section, we illustrate the reverse function of the embedding process. The secret bits are extricated from the LSB of the non-seed pixels of the SI. The bits extricated from the cover media include the EHR and the WM. The drawn-out bits are distinguished into WM and EHR content. Subsequently, decryption is performed by the XOR operation between the encrypted WM bits and the HC-PRN sequence to generate the WM logo. The WM bits thus received are reshaped into a logo of size 128 × 128. If it is observed that the WM is affected in any manner, the EHR is not extracted and a request for retransmission is initiated. However, if the WM is correctly decrypted, then the EHR data bits are extracted and decrypted using the HC-PRN sequence. The seed pixels from each block are acquired to reconstruct the original image. The data extraction method is described as Algorithm 2. s1 , s2 , s3 , s4 ← key as an initial condition for hyperchaos 2.
end for 9.
Obtain PRN sequence 10. Extricate the LSB bits from the non-seed pixels of the SI to form the encrypted WM and encrypted EHR data bit vectors 11.
for rounds ← 1: length (extracted data vector) 12. Original data vector = extracted data vector ⊕ k 13. End 14. Obtain the WM and EHR from this data vector 15. Reshape the original WM to check if it has tampered 16. Reconstruct the CI using the seed pixels. 17. END

Experimental Results
The MATLAB R2017a platform has been utilized for carrying out the experimental investigations for different gray-scale natural and MI. Both types of images that have been used for testing have dimensions 512 × 512. We have conducted experiments using a 64-bit Windows 10 Operating system with an i5 processor, 8 GB RAM, and 2.40 GHz clock speed. The different images, as well as a logo as WM, have been represented in Figure 3. The binary WM employed for authentication purposes has a 128 × 128 size. The technique is evaluated for a payload of 196,608 bits or 0.75 bits per pixel (bpp). The scheme reports an average encryption speed close to 1Mbps. The image quality metrics applied for evaluation of the scheme include normalized cross-correlation (NCC), peak signal to noise Ratio (PSNR), and structural similarity measure index (SSIM) [16][17][18][19][20]. In-depth analysis has been performed, which includes imperceptibility analysis, computational complexity analysis, and fragility analysis. Furthermore, a detailed comparison of many contemporary techniques has been described. Furthermore, the fragility analysis carried out reveals that the WM is fragile to all possible attacks and can easily detect tampering of data. Ratio (PSNR), and structural similarity measure index (SSIM) [16][17][18][19][20]. Inhas been performed, which includes imperceptibility analysis, computatio analysis, and fragility analysis. Furthermore, a detailed comparison of ma rary techniques has been described. Furthermore, the fragility analysis carr that the WM is fragile to all possible attacks and can easily detect tamperin

Imperceptibility Analysis
The main attribute of the presented image interpolation method is it generating perceptually better-quality CI. The CIs generated by this sche high quality as indicated in Figure 4 and the evaluation quality metrics wh PSNR1 and SSIM1 have been presented in Table 1. The scheme performs e the natural as well as the MI.

Imperceptibility Analysis
The main attribute of the presented image interpolation method is its capability of generating perceptually better-quality CI. The CIs generated by this scheme are of very high quality as indicated in Figure 4 and the evaluation quality metrics which include the PSNR1 and SSIM1 have been presented in Table 1. The scheme performs equally well for the natural as well as the MI.

Computational Complexity Analysis
The analysis of computation complexity for CI generation time is illustrated for various test images in Table 2. A detailed comparison of the CI generation time of our technique with two important interpolation schemes NMI [36] and INP [40] is presented below. It could be seen that despite providing better quality CI, our scheme does not require any extra computational effort. The proposed interpolation method takes lesser time compared to the state-of-art-NMI and INP schemes as is evident. This makes the developed system a potential candidate for real-time imaging systems.

Payload and Reversibility Analysis
The payload is a critical underlying requirement of a data hiding system. The perceptual quality of the SI is evaluated using subjective and quality metrics like PSNR and it is a direct function of payload. In our proposed system we have embedded two types of data in the CI obtained using a novel IP method: EHR and watermark. Both types have been embedded using the LSB approach. The visual quality of the SI yielded by the embedding method is evaluated based on objective as well as subjective analysis. Figure 5 presents various original and Stego-MIs with a payload value equal to 0.75 bpp. The generated SI's quality is much superior, along with high payload value. Furthermore, to validate the described RDH method a comparison has been made with contemporary works mentioned in [44] as shown in Figures 6 and 7.     The PSNR2 and SSIM2 value for SI and the said payload is illustrated in Table 3.  Table 3 results depict that our system is capable of providing high visual quality images despite embedding an average payload of 0.75 bpp reversibly. This data embedding capacity is more than four times compared to 0.18 bpp [24]. Also, our scheme being implemented in the spatial domain is computationally less complex compared to [24].
In the comparison of the schemes for EC and PSNR-evaluation parameters, it is evident that our scheme performs better than many contemporary methods. The PSNR is comparable and the security offered to the WM and EHR is improved.

Fragility Analysis for Noise Attacks, Filtering Attacks, and Compression Attacks
The various watermarking images obtained have been attacked using different attacks as depicted in Table 4. This has been done to evaluate the fragility analysis of the proposed scheme. The value of avg. PSNR for MI is 52.3756 dB whereas the avg. SSIM is 0.9849. This indicates that the scheme can yield high-quality images with a considerably high payload.
The PSNR2 and SSIM2 value for SI and the said payload is illustrated in Table 3.  Table 3 results depict that our system is capable of providing high visual quality images despite embedding an average payload of 0.75 bpp reversibly. This data embedding capacity is more than four times compared to 0.18 bpp [24]. Also, our scheme being implemented in the spatial domain is computationally less complex compared to [24].
In the comparison of the schemes for EC and PSNR-evaluation parameters, it is evident that our scheme performs better than many contemporary methods. The PSNR is comparable and the security offered to the WM and EHR is improved.

Fragility Analysis for Noise Attacks, Filtering Attacks, and Compression Attacks
The various watermarking images obtained have been attacked using different attacks as depicted in Table 4. This has been done to evaluate the fragility analysis of the proposed scheme. Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image Sensors 2021, 21, x FOR PEER REVIEW 13 of 20 Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others.   Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.

Extracted Logo
Sensors 2021, 21, x FOR PEER REVIEW 13 of 20 Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.

Attacked Image
Sensors 2021, 21, x FOR PEER REVIEW 13 of 20 Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.

Extracted Logo
Sensors 2021, 21, x FOR PEER REVIEW 13 of 20 Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.

Attacked Image
Sensors 2021, 21, x FOR PEER REVIEW 13 of 20 Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks.  Table 4. Subjective quality images after various attacks.

Effect of Gaussian Noise (0.001) on the Extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo

Effect of jpeg 50 compression on the extracted Watermark for Authentication
Attacked Image

Extracted Logo
Attacked Image Extracted Logo Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks. Table 4 shows the salt and pepper noise with noise density 0.01 can produce recognizable WM. The effect of other attacks like the Gaussian noise (0.0001), Median Filtering (MF) attacks, and JPEG 50 compression are also shown in Table 4. It is concluded that the fragile WM can detect many types of tampering attempts like noise attack, filtering attack, and compression attacks of various quality factors among many others. Table 5 represents the value of NCC for different attacked watermarks. The NCC values presented in Table 5 for various attacks reflect that our system is capable of detecting any attack. This is evident that for even a low strength attack, we are unable to recover the authentication WM.

Comparison of Proposed Scheme with Contemporary Methods
We have compared our scheme with several state of art techniques. The comparison of our scheme in terms of payload, BPP, and SSIM values are shown in Table 6 with reported works in [25,27,37,39,40,44]. The results depicted in Table 6 show that the proposed scheme provides high-quality SIs despite high embedding payload. Furthermore, a comparison of the scheme has been made with [24][25][26]28,33,34,36,37,39,44,55] for PSNR and payload for natural images MI-K and MI-M as shown in Table 7. Tables 6 and 7 that our scheme provides better quality SIs, irrespective of the fact whether the cover media is a medical image or a natural image. In addition to this, unlike most of the schemes under comparison, like [24][25][26]33,34,37,39,45], our scheme makes use of a fragile WM for ensuring the early tamper detection at the receiver. This feature coupled with reversibility makes the proposed scheme a good candidate for securing patient records in e-health.

Key-Space and Statistical Analysis
The proposed technique utilizes Chen's hyperchaotic system which utilizes four initial values s1, s2, s3, s4, and seven control parameters a, b, e, t, l1, l2, l3. Thus a total number of eleven parameters need to be set to define the actual keyspace. We have chosen the precision levels of all the eleven quantities as 10 −10 . This results in a total key space of 10 110 (~2 360 ) which is extremely large and as such the proposed scheme is robust to the brute-force attack. A comparison of Key space with state of the art has been shown in Table 8. It is evident that our scheme provides better key space compared to state-of-the-art. Additionally, statistical analysis is of vital importance to determine strength of encrypted data. The effectiveness of the hyperchaotic system used in this work has been reported in [67,68] by successfully subjecting it to various tests of Statistical Test suite (NIST SP 800-22).

Encryption Speed
Speed of data encryption is an important parameter. Chen's 4-D HC system used for chaotic sequence generation has a set of four non-linear equations that use eleven control parameters. The overall speed of the encryption depends on computational time of chaotic sequence followed by its use for data encryption. Table 9 provides the speed analysis of proposed system compared to state of art. As observed, our scheme outperforms those under comparison.

Histogram Analysis
When data is embedded in a cover image, it changes its various parameters. One of the important requirements of a data embedding system that keeps statistical attacks at bay is that stego-image histogram should not significantly alter when data are embedded in it. Ideally, the cover and stego-image histograms should be identical. The more the two histograms are identical, the less information a histogram of the stego-image reveals about data embedded into it. We have tested our scheme for histogram analysis at a payload of 0.75 bpp. The results for some of the images have been depicted in Figure 8. The histograms of the original and stego-images are highly identical, indicating that the proposed scheme is less susceptible to statistical attacks.

Histogram Analysis
When data is embedded in a cover image, it changes its various parameters. One of the important requirements of a data embedding system that keeps statistical attacks at bay is that stego-image histogram should not significantly alter when data are embedded in it. Ideally, the cover and stego-image histograms should be identical. The more the two histograms are identical, the less information a histogram of the stego-image reveals about data embedded into it. We have tested our scheme for histogram analysis at a payload of 0.75 bpp. The results for some of the images have been depicted in Figure 8. The histograms of the original and stego-images are highly identical, indicating that the proposed scheme is less susceptible to statistical attacks.

Image
Original Image Stego-Image

Discussion
In this section, we present a discussion regarding the capability of the proposed method for its application in Health 4.0. The main aim of this work is to develop a scheme having the ability to enhance the security of EHR. In a smart-health system, while transferring EHR to the receiver, many parameters are of significant importance. These include the security of EHR, lossless nature of diagnostic data, authentication of the received content, and less computational complexity involved.
Towards this end, we have developed a dual-layer security framework for EHR utilizing a hyperchaotic algorithm and a newly developed interpolation scheme. The developed interpolation scheme ensures reversibility, hence avoids any loss of cover media information, facilitating better diagnosis. It was observed that our interpolation scheme performs better on both the fronts of the visual quality of stego-images and computational complexity. This has been observed by evaluating various parameters including imperceptibility, computational complexity, reversibility, and fragility.
The stego-module used to embed EHR has been evaluated using various objective parameters, which shows that the scheme provides an average PSNR of more than 52 dB for a payload of 0.75 bpp while maintaining reversibility. A comparison with state-of-art techniques reveals that the scheme is efficient enough to handle a higher payload while maintaining imperceptivity. For an embedding rate of 196,608 bits, an average PSNR of 52.3866 dB has been obtained for this method which is higher by 3.37 dB from the scheme in [39,44], 11.19 dB higher than the scheme present in [27]. The keyspace analysis shows that our system is highly robust to brute force attacks, while as histogram analysis also reveals that the proposed scheme is robust to statistical attacks.
The use of hyperchaotic encryption and the fact that data has been embedded in the spatial domain result in relatively lesser computational complexity compared to transform domain embedding. The watermark embedded within the image has been used to check the tampered regions (if any) at the receiving end. Fragility analysis shows that the proposed scheme can detect the presence of the tampers for various possible attacks which is a very important parameter while designing an e-health data security scheme.

Conclusions
Security of the EHR and authentication of the received content is significantly important in e-health services. This paper presents a dual-layer security mechanism for EHR security. We make use of steganography and cryptography to develop an efficient and secure system. Firstly a new interpolation method is developed and analyzed. It has been shown that the proposed interpolation method is capable of producing better CIs compared to some well-known methods like INP and NMI. The data to be embedded in the CIs generated by the proposed interpolation scheme is encrypted using a hyperchaotic system. This results in a highly secure system as a result of a very high keyspace of 10 110 . The embedded data includes EHR and a 128 × 128 authentication logo. To ensure reversibility of the cover images, no embedding is done at seed pixels, leaving them un-altered. The proposed scheme provides an average PSNR of more than 52 dB for a payload of 0.75 bpp. A comparison with various state-of-art techniques shows that our scheme is efficient enough to handle a higher payload while maintaining imperceptivity and reversibility. We have carried out the authentication analysis of our scheme to various signal processing attacks, and it shows that it can successfully detect various alterations. Thus the authenticity of the received content could be verified utilizing the proposed scheme. Also, the histogram analysis shows that it provides stego-images with a closely matched histogram to that of CIs. This minimizes any chances of statistical attacks. The scheme has been implemented in the spatial domain and, as such, is computationally efficient. Given all the features of the proposed scheme, it can be utilized for the security and authentication of EHR in Health 4.0. In the future, we aim to improve the data hiding capacity and computational complexity of our scheme by utilizing new chaotic maps.