Multimodal Fusion-Based Image Hiding Algorithm for Secure Healthcare System

The development of artificial intelligence plays a significant role of multimedia applications, especially in the healthcare domain. However, it has brought about the problem of sensitive information leakage. To address these challenges, an interesting multimodal fusion-based robust image hiding algorithm is proposed in this article. First, fused image is considered as mark image generated from MRI and CT images using nonsubsampled shearlet transform. Second, we employed principal component analysis to compute the appropriate coefficients of cover image for embedding purpose. Third, fused mark image is Arnold cat map encoded to address the security issue hidden mark media. Finally, the fusion of fractional dual-tree complex wavelet transform and randomized singular value decomposition is utilized to conceal encrypted fused mark media inside host image. Our findings show that the proposed algorithm outperforms some of the recent techniques in terms of high robustness and invisibility.

The development of artificial intelligence plays a significant role of multimedia applications, especially in the healthcare domain. However, it has brought about the problem of sensitive information leakage. To address these challenges, an interesting multimodal fusion-based robust image hiding algorithm is proposed in this article. First, fused image is considered as mark image generated from MRI and CT images using nonsubsampled shearlet transform. Second, we employed principal component analysis to compute the appropriate coefficients of cover image for embedding purpose. Third, fused mark image is Arnold cat map encoded to address the security issue hidden mark media. Finally, the fusion of fractional dual-tree complex wavelet transform and randomized singular value decomposition is utilized to conceal encrypted fused mark media inside host image. Our findings show that the proposed algorithm outperforms some of the recent techniques in terms of high robustness and invisibility.
T he development of the AI plays a significant role of multimedia applications, especially in the healthcare domain. 1 Due to the global COVID-19 pandemic time, electronic patient record contains sensitive information, and these records are shared with healthcare workers, patients, and in some scenarios, researchers for diagnostic, business analytic, and study purposes. 1,2 Presently, the use of the Internet of Things (IoT) in several applications not only improves operational efficiency for professionals and organizations but also provides service convenience for support staff and their families. 3 However, the security of multimedia data in the form of images is a prerequisite for IoT-based applications. At the same time, they are also more and more stored in an externalized way, as in the cloud or data warehouse. To this end, these electronic data must be protected against illegal access and fraudulent usage.
Digital watermarking is the most-used copyright and content authentication scheme, as because of its robustness and invisibility. Watermarking schemes hide secret information to achieve covert transmission, identity authentication, and copyright protection. 3,4 Robustness, invisibility, and embedding capacity are the major characteristics of any watermarking scheme. 4 However, there are several issues that must be addressed, such as the security of digital records, achieving a good balance between major characteristics of digital mark with low cost. 4 We present a fusion-based image-hiding scheme for color images, which can ensure high security and robustness with good invisibility concurrently. The main contributions are as follows.
› Advantages of multimodal image fusion: This article utilizes the mechanism of the nonsubsampled shearlet transform (NSST)-based fusion 5 to fuse the multimodal images. Compared to the individual image, the fused image has rich information.The proposed algorithm considers fused mark image to assure the accurate diagnostic for healthcare applications.
› Encryption of fused image before embedding: Fused mark image is Arnold cat map 6 encoded to improve the security of the hidden mark. It is applied due to its rich properties of simplicity and periodicity. contains the majority of the information. 7 It determines the suitable component for embedding purpose without degradation of the host media quality.
› Embedding of fused image through fractional dual-tree complex wavelet transform (Fr-DTCWT) and randomized singular value decomposition (RSVD): The fusion of Fr-DTCWT and RSVD is utilized to perform an imperceptible marking of the encrypted fused image within the host media, which offers the high robustness and imperceptibility. Fr-DTCWT 6 contains several properties, such as multidirection, shift invariance, and low cost compared with wavelet-based transform. In addition, RSVD has less expensive storage and computation costs than SVD. 8 › Improved robustness: The simulated and experimental results demonstrate the advantages of our algorithm in comparison with those of current watermarking algorithms.
The rest of this article is organized as follows. The relevant techniques are presented in the "Literature Review" section. The proposed algorithm details are introduced in the "Proposed Algorithm" section. The extensive experimental results are reported and analysed in the "Results Analysis" section. Finally, the "Conclusion" section concludes this article.

LITERATURE REVIEW
For the last few years, several watermarking mechanisms have been proposed for telehealth scenarios. For example, Fares et al. 9 introduced a blind watermarking framework for color images in transform domain. Different Fourier transform have been applied on each component of host image, and then mark data are concealed into the appropriate coefficient of host image, which maintains the visual quality. It offers better resistance against wide range of attacks. Su and Chen 10 described Hessenberg-based watermarking approach for providing the copyright protection of multimedia object. Prior the embedding process, Arnold transform was applied on mark, and then scrambled mark image concealed into Hessenberg matrix of host image. This scheme has less resistance against geometric attacks. Chen et al. 11 suggested QR decomposition-based data hiding scheme for multimedia data. First, mark image is scrambled using Arnold transform to ensure the additional security of this scheme. Further, QR decomposition is performed to determine the optimal position for embedding purpose. This scheme is less resistance against scaling attacks and Gaussian noise. Prabha and Sam 12 introduced a Walsh-Hadamard transform (WHT)-based watermarking scheme, which can protect multimedia information against copyright issue. First, the cover image is transformed using WHT and then mark data are concealed into the cover image. It delivers better robustness against most attacks. To achieve the copyright protection of multimedia data, Shen et al. 13 proposed an optimization-based data hiding scheme for color image. The adaptive multiple embedding factor has been employed to determine optimal position for embedding purpose. Further, the integration of particle swarm and grey wolf optimization has been utilized to finds suitable factor for embedding purpose. Liu et al. 14 presented an algorithm, which aims to provide the solution of the copyright issue. Prior to embedding procedure, mark image is encrypted using affine transform, which offers high security. Al-Otum 15 proposed a robust and secure watermarking framework using wavelet packet transform. Prior to embedding, authors applied the combination of chaotic and Arnold transform on mark image, and then encrypted mark image is concealed into cover image to achieve high security. However, the payload value is low. From the previous discussion, there exist three problems: 1) most of the existing schemes are failed to make a good balance between major characteristics of watermark; 2) most of the watermarking schemes have limited security; and 3) most of the abovementioned references are embedding single mark instead fused mark, which leads to an inaccurate diagnostic for healthcare applications. Algorithm 1. Multimodal image fusion approach 1: Input: Img 1 , Img 2 2: Output: Fus img 3: Begin 4: Img 1 imread (Img 1 ) 5:

PROPOSED ALGORITHM
The proposed algorithm consists of three major parts: the fusion of multimodal medical images, the dimensionality reduction of the host image, and then embedding and extraction of the fused mark. The flowchart diagram of the complete proposed method is shown in Figure 1. Notation and its explanation used in the algorithms are given in Table 1.

Fusion of Multimodal Medical Images
In this article, we utilize NSST-based fusion to fuse the CT and MRI images to generate a fused mark before embedding. The algorithmic steps for the generation of fused image are shown in Algorithm 1. for j = 1 : y do 11: Wat ext (i, j) (S 2 (i,j)-S(i,j)) /

Dimensionality Reduction Using PCA
In the preprocessing stage, the cover image is transformed into RGB component, and then reshape each component of the cover image to obtain the modified matrix. Further, PCA is utilized on the modified matrix to determine the PCA img , which is a suitable component for embedding purpose. The simplified procedure of dimensionality reduction is described in Algorithm 2.

Embedding and Extraction Procedure of Fused Mark
Prior to embedding, fused mark image is Arnold cat map encoded to improve the security of the hidden mark. After that, the combination of Fr-DTCWT and RSVD is utilized to transform PCA img for embedding purpose. Further, the embedding operation is carried out by using the singular value of Enc img with the help of factor, "a." Furthermore, the inverse of RSVD, Fr-DTCWT, and PCA operation is employed to obtain the marked image, Wat img . In the extraction procedure, fusion of Fr-DTCWT, and RSVD is performed on Wat img . After that, inverse RSVD is applied to recover the encrypted fused mark image,Enc img . Finally, Arnold cat map is performed on Encf img to recover fused mark image, W img . Algorithms 3 and 4 show the embedding and extraction procedure, respectively.

RESULTS ANALYSIS
Experiments were performed using MATLAB R2019a to access the performance of the proposed algorithm. Different multimodal medical images of 256Â256 pixel, which is considered as mark image, and cover image of 512Â512 pixel are chosen from Openi, 16 Kaggle datasets, 17 respectively. To evaluate the performance of the proposed method, peak signal to noise ratio (PSNR) and structural index similarity (SSIM) are utilized between the cover and marked image to compute the visual quality and normalized coefficient (NC) is determined to measure robustness between the original mark and extracted mark image. 18 The number of pixels change rate (NPCR) and unified average changing intensity (UACI) are utilized to determine the strength of the encryption technique. The performance is examined at different gain factors in terms of PSNR, SSIM, and NC with the outcomes, as shown in Table 2. As observed from Table 2, PSNR for all images is more than 59.45 dB and the SSIM value is close to the ideal value 1, indicating the good visual quality of marked images. Further, NC has an ideal value of 1, indicating good robustness of extracted mark image. The distorted marked images are obtained against the different attacks, with the outcomes, as shown in Figure 2. From this figure, it can be noticed that the visual quality of distorted marked images is acceptable against listed attacks.
Further, NC has ideal values 1, indicating good robustness of extracted mark image against attacks, with the outcomes, as shown in Figure 3. Comparative analysis of the proposed algorithm with some existing works of Chen et al., 11 Singh et al., 19 and Mahto et al.'s 20 is carried out in Table 3. The NC scores of the proposed algorithm are higher than the schemes in Chen et al., 11 Singh et al., 19 and Mahto et al.'s 20 work under considered attacks. For example, when Gaussian noise with density ¼ 0.01, the NC score of the proposed algorithm is 0.9792 greater than that of the Chen et al. 11 and Mahto et al.'s 20 work. However, when sharpening noise with density ¼ 0.1, the NC score of the proposed algorithm is greater than that of the Singh et al. 19 and Mahto et al.'s 20 work. Further, the comparative analysis of the proposed algorithm with some existing works of, 10 Chen et al., 11 Singh  et al., 19 and Mahto et al. 20 in terms of some more parameters is carried out in Table 4. For example, when the payload ¼ 0.2500 b/pixel, PSNR and SSIM score of the proposed algorithm is 63.98 and 1.00 dB, respectively, greater than that of Su and Chen's, 10 Chen et al.'s, 11 Singh et al.'s, 19  The average value of NPCR and UACI score is obtained as 0.9960 and 0.3346, respectively, which

Attacks
Noise density

Parameters Proposed algorithm
Su and Chen 10 Chen et al. 11 Singh el al. 19 Mahto  10 Chen et al.'s, 11 Singh et al.'s, 19 and Mahto et al.'s 20 work. Furthermore, Figure 4 displays the robustness performance against the mentioned hybrid attacks. From this figure, it can be observed that NC score is greater than 0.9682 against most of the attacks, which shows the high robustness feature of the proposed method.

CONCLUSION
A robust watermarking algorithm is suggested for the secure transmission of medical images for healthcare applications. Initially, the principal component of the cover image is computed by using PCA for embedding purpose to assure high invisibility. Prior to embedding, Arnold cat map is employed on fused image, which can enhance the security of the watermarking system. Several standard tests were performed on the mentioned dataset to determine the benefit of the proposed technique. Our simulation findings show that the proposed algorithm outperforms some of the recent techniques in terms of high robustness and invisibility. Improvement in robustness and security through deep learning and blockchain can be seen as future work.