Improved Multileader Optimization with Shadow Encryption for Medical Images in IoT Environment

1  Introduction

The current technological advancements and development in Internet of Things (IoT) alleviate the individuals to employ resources across the globe. This rapid advancement in IoT technology facilitates the patients to get consulted by clinicians worldwide. Patient information like X-rays, scan report, clinical test reports etc., can be forwarded via internet while most of the clinical data remain in image format [1]. In general, information communicated via internet is prone to attacks and vulnerabilities. Any misinterpretation or corruption of medical data, due to transitional assaults is non-tolerable. At times, it might threaten the lives of humans as well. This mandates the scheme for safe remote monitoring of medical data [2]. Various image encoding structures have been proposed earlier for ordinary images. However, in case of medical images, such techniques are incapable of preserving the patient’s information. This is because of the strong correlative nature of pixels in healthcare images which tend to have adjacent pixels and other numerous modalities [3]. This complication motivated the authors to develop highly advanced strong encoding schemes [4].

Particularly, medical images such as X-rays, Magnetic Resonance Imaging (MRI), and Computed Tomography (CT) consume huge memory and have high pixel correlation. In case of no security for these images, it can be easily tampered with and attacked through illicit access. Thus, it is essential to develop highly effective health care image encoding techniques [5,6]. The demand for efficient cryptographic solutions for healthcare images need highly advanced system and its seamless application. In order to protect the image from new generations of attacks, encoding solutions must guarantee image confidentiality. Confidentiality can be achieved by encoding the data so that the data becomes unusable and unintelligible, when it is hacked or lost [7–9]. Among the available encoding schemes, symmetric encoding is the best cryptographic solution that provides heavy security for huge volumes of data [10]. In this creative ideology, chaos is considered as an efficient axis of latest cryptographic difficulties that prevail in conventional symmetric encoding systems like Advanced Encryption System (AES) [11,12]. Medical images usually contain confidential information about the patients and it is vulnerable to cyber-attacks when it is transmitted via public networks [13]. Therefore, it is mandatory to secure such images prior to transmission over public networks. The existing encoding approaches cannot provide the highest possible security since healthcare images have varied properties redundancy, higher correlation among pixels, and huge size [14]. In such case, these techniques are incapable of resisting numerous security assaults. In this background, the primary goal of this study is to design an effective image encoding method that can advance the development of complicated secret keys.

The current article introduces Improved Multileader Optimization with Shadow Image Encryption for Medical Image Security (IMLOSIE-MIS) technique for IoT environment. The proposed IMLOSIE-MIS initially generates a set of shadows for every input medical image. Besides, shadow image encryption process takes place with the help of MLO with Homomorphic Encryption (IMLO-HE) technique. Here, the optimal keys are generated using MLO algorithm. On the receiver side, decryption process is initially carried out and shadow image reconstruction process is fulfilled. The experimentation analysis was conducted upon medical images and the results infer the superior performance of IMLOSIE-MIS model over other existing models.

2  Related Works

In literature [15], low-area and high-speed architectures were suggested for Secure IoT (SIT) systems under resource-limited applications. The presented pipeline infrastructure was highly helpful for high-frequency applications whereas the suggested serial architecture is advantageous for low-area necessities that result in hardware cost reduction. The dynamic size of block selection approach rescues a substantial volume of clock cycles during image encryption. Elhoseny et al. [16] inspected the safety of health care images in IoT by using a new cryptographic method with optimization strategies. Most of the times, a patient’s data is saved in hospital cloud server due to security purposes. To increase the security level of decryption and encryption procedures, an optimal key has to be selected with the help of hybrid swarm optimization i.e., Particle Swarm Optimization (PSO), Elliptic Curve Cryptography (ECC) and grasshopper optimization.

In the study conducted earlier [17], a server-client authentic medical image transmission model was suggested. Previously, AES-encoded health care image is secured by embedding One Time Password (OTP) that can be generated via Tent map. Once an OTP is received via Global System for Mobile communication (GSM), the official user confirms the similarity of OTP received on their mobile, and accesses the encoded health care image. The researchers in literature [18] concentrated on disease classification using image processing method. In this study, a secure cloud computing environment was used in which a prolonged zigzag image encoded scheme was tested. This scheme showed high tolerance towards distinct data attacks. Then, a fuzzy Convolutional Neural Network (CNN) system was suggested in this study for potential image classification. The decoded images were utilized for the classification of cancer stages with distinct layers of training. Shafai et al. [19] provided a medical image cryptosystem in which Stacked Autoencoder (SAE) network is used for the generation of two sets of chaotic random matrices. Another set produces a separate series which is used for removal of association between original image and the permuted encoded medical image.

Kamal et al. [20] provided a novel encoding system for encoding both color and grey medical images. The potentiality of the suggested technique to encode medical images was assessed with the help of time complexity and security analysis. In literature [21], an Evolutionary system such as memetic system was utilized for encoding the text messages. The encoded data was inserted in medical images with the help of Discrete Wavelet Transform 1 level and 2 levels.

3  The Proposed Model

In current study, a new IMLOSIE-MIS model has been developed to accomplish security in IoT-enabled medical image transmission process. The proposed IMLOSIE-MIS model primarily generates a set of shadow images for the applied medical images. Followed by, IMLO-HE technique is derived to encrypt the shadow images in an effective manner. At the same time, the optimal generation of the keys, involved in HE technique, is performed by MLO algorithm in such a way that the PSNR value gets maximized. Fig. 1 is the block diagram of the proposed IMLOSIE-MIS approach.

Figure 1: Block diagram of IMLOSIE-MIS approach

3.1 Shadow Image Creation Process

In initial stage, the presented IMLOSIE-MIS model primarily generates a set of shadow images for the applied medical images. During shadow creation process, binary secret images are encrypted to ‘n’ shadows from wavelet sub-bands [22]. Each shadow encompasses white and black pixels in the state of clamor and is particularly larger in measurement compared to secret images. Each original pixel of the secret image appears in ‘n’ attuned version and is labelled as ‘shadows’. Fig. 2 illustrates the structure of shadow creation process. Thus, each shadow is a collection of subpixels of RGB images as shown below.

Figure 2: Shadow creation process

Rp=∫1klimk→1,to,no.ofpixelRmn−Subband(1)

Gp=∫1klimk→1,to,no.ofpixelGmn−Subband(2)

Bp=∫1klimk→1,to,no.ofpixelBmn−Subband(3)

In the abovementioned conditions, the matrix location of every band and k denotes the overall number of pixels. At last, two shadow images are attained by shadow creation condition; visual secret share generation process is employed in the production of two shadows in all the images. The individual shadow does not reveal the secret except its size. The original secret is recreated by integrating all the shadows. In encryption technique, each secret pixel becomes two shadows whereas all the shadows belong to respective shadow images.

1.    Choose the secret input images with recompense width and tallness that spread from RGB in order to gray scale the image.

2.    Use wavelet transformation on luminance so as to produce four sub-bands namely HL,LL,HH, and LH, which are determined by high pass, low pass, fat and vertical measurements.

3.    Considering the sub-bands, make two lattices to offer white mystery pixels to the same as black mystery pixels.

4.    Replace the sub-bands by offer. Considering that the ultimate objective is to implement supplant tasks, it should be equal to the size of the sub-bands.

5.    A reverse Discrete Wavelet Transform (DWT) is transported to restore the image.

6.    Initialize variable text style shading and record values.

Visual Secret Sharing system is an effective approach to hide an image. It is completed by segregating the images into meaningless shadows. The secret can be recovered by superposing a thresholding number of shadows without unpredicted computation. The individual shadow does not reveals the secret, except their size while the shadows are transferred to a participant number.

3.2 Shadow Image Encryption Process

Once the shadows are created, IMLO-HE technique is implemented to encrypt the shadow images in an effective manner. In order to encrypt the image or data, Homomorphic Encryption plays a further operation that is represented as a public key cryptosystem. This process contains four functions such as encryption, key generation, decryption, and evaluation, in addition to decryption of the data for evaluation; the evaluation process offers a corresponding result which should be validated to check whether the task is performed appropriately on initial messages. The decision of the plan is subjected to the kind of processes done in the applications, detached from the variable which is used to select the encryption strategy.

In encoding and decoding processes, the related images use a symmetric key while trustworthy security and secrecy are offered in key generation process. Now, key generation technique is continued to choose further parameters for registering both public and private keys.

HpkandHsk

K=cdandω=lcm(r−1,s−1)

For this process, the keys are generated for both decryption and encryption processes in a random manner. These keys should be optimized using IMLO approach. While rest of the process in image security system involves the generation of ideal ideal public and private keys.

3.3 Optimal Key Generation Process

MLO technique is arithmetically modelled for the implementation of optimization problems [23]. The key objective is to exploit information from a member of the population. In such cases, a member of the population exploits the information of diverse leaders to assess the problem-solving space as given herewith.

Xi=[χi1,…,χid,xim](4)

In Eq. (19), Xi refers to i-th member of the population and χid specifies the d-th variable of optimization problem.

Xsort=[X1sort⋮XNsort|Xr1⋮XrN|min(fit)⋮max(fit)](5)

In Eq. (20), XNsort denotes the member with worst fitness value, Xsort specifies the matrix of population, X1sort represents the member with optimum fitness value, XrN signifies the member of the population with worst rank-related fitness value, Xr1 indicates the member of population with best rank-based fitness value, N characterizes the number of members in the population, and fir shows the fitness value. After arranging the population matrix, certain members of the ranked population are selected as leaders.

L={Xlsort,Xlsort∈Xsort,l=1:nL}(6)

xsort=[X1sort⋮XnLsort⋮XNsort]N×m→L=[X1sort⋮Xlsort⋮XnLsort]nl×m(7)

In this expression, L denotes the designated leader member matrix whereas nL indicates the leader numbers. The population in MLO is upgraded. At first, every member of the population is moved into search space based on the location of the leader. Then the leader is determined based on roulette wheel. Every leader is selected for upgrading various parameters.

fitin=fiti−max(fit)Σj=1N(fitj−max(fit))(8)

Pl=fitlnΣj=1nlfitjn(9)

Cl=Pl+Cl−1,C0=0&l=1:nl(10)

Li,cd={ L1=X1sort,0≤r≤C1⋮⋮Lc=Xcsort,Cc−1:.≤r≤C1⋮⋮Lnl=XnLsort,Cnl−1≤r≤Cnl(11)

χi,newd=χid+rand(Li,cd−2×xid)(12)

Xi={Xi,new,fiti,new≤fitiXi,else(13)

Here, fitin specifies the standardized fitness function for i-th population member, Pl represents the possibility of selecting l-th leader to guide the variable, Cl indicates the cumulative possibility of l-th leader, xi,newd indicates the updated value for d-th parameter of i-th dimension, Li,cd symbolizes the d-th parameter of the selected c-th leader to guide d-th variable of i-th population member, and r indicates the arbitrary integer that lies in the interval of [0,1]. Once the updated position is highly pertinent, the member upgrades their position to current status or otherwise they return to their prior position.

xi,newd=xid+2×(1−τT)×(−0.2+rand×0.4)×xid(14)

Xi={Xi,new′fiti,new≤fitiXi,else(15)

In this expression, t represents the tth iteration and T indicates the maximum number of iterations. MLO algorithm is extended to the design of IMLO algorithm based on Nelder Mead (NM) concept [24]. NM model is a simple search methodology for multi-dimensional unrestricted minimization. It consists of expansion, reflection, shrinkage, and contraction. Initially, (n+1) point x1,x2,…,xn+1 is initialized in D-dimensional searching space. Such points are arranged in ascending sequence based on the fitness value. The centroid point X is comprised of initial n point which is evaluated as follows.

x¯=∑i=1nχin(16)

Next, the reflection point χγ is shown below.

xr=x¯+ρ(x¯−xn+1)(17)

In Eq. (17) ρ>0 refers to reflection coefficient, and xn+1 denotes the worst point. Then, the expansion point xe is evaluated by the following expression.

xe=x¯+χ(xr−x¯)(18)

In Eq. (18), x>max{1,ρ} indicates the expansion coefficient. When the fitness of reflection point xγ is worst than the secondary worst xn, it is essential to implement inside or outside contraction. Now, the fitness of point represents the function value. Once the fitness of reflection point xγ is superior to the worst point xn+1, the outside contraction operation is implemented as given below.

x0C=x¯+γ(xr−x¯)(19)

Or else, the inside contraction process gets implemented.

xic=x¯−γ(x¯−xn+1)(20)

where 0<γ<1 indicates the contraction coefficient. When the fitness of outside contraction point x0C is worst compared to reflection point χγ, then the fitness of inside contraction point xic is worst than xn+1. In such case, the shrink operation for i-th point xi is implemented as given below.

vi=x+σ(x−x)(21)

If 0<σ<1 represents the shrink coefficient, then the comprehensive data is applicable for the four abovementioned coefficients.

IMLO algorithm defines a fitness function for optimal key generation process which defines the maximization of PSNR. PSNR defines the proportion between the possible assessment of signal and the energy of mutilating clamor that influences its nature of portrayal. Both reference and original images should be of similar class and size.

Fitness=Max(PSNR)(22)

PSNR=1s∑i=1shares10log10⁡(2552/MSE)(23)

The PSNR condition ‘s’, as a number in DWT, is used as shadows. In Peak Signal to Noise Ratio (PSNR), the square of peak values in the images should be assumed (in case of 8-bit images, the peak value is 255) and divided by Mean Square Error.

3.4 Encryption and Decryption Process

During encryption procedure analysis, an ideal public key is used for encoding all the pixels of the images. It is regarded as message bit m while the figure should be computed to notify cipher data action. With the help of confidential key Hsk, the client encrypts the original image Ip and create Hk−opt(Ip) together with public key, Kpk. This ciphered image Ic is sent to the server. Hpk=(k,i) and Ksk=(c,d) Enc(I,Hsk) select the arbitrary variables and cipher data c=I.rkmodk2 is calculated. During decryption process, the ciphered image is reviewed since it contains the encryption pixel addressed by (c,d) and the secret vector, s. Decryption method consumes two veils, particularly, the confidential mask <as> while even the masks from steady progression. The message bit (pixel esteem) m in cipher image and another confidential parameter should be decoded. The generated dec(f(Hsk−opt)) is decrypted by the used based on Ks and the following primary result is obtained.

DecImage=L(cαmodkopti2)L(iαmodkopt2)modk(24)

4  Experimental Validation

The current section validates security performances of the proposed IMLOSIE-MIS method utilizing a set of benchmark medical images. The outcomes were assessed under both presence and absence of attacks. Tab. 1 and Fig. 3 is a brief overview on the results achieved by IMLOSIE-MIS model without attacks under different measures such as Mean Square Error (MSE), PSNR, Bit Error Rate (BER), and Structural Similarity Index (SSI). The table indicates that the proposed IMLOSIE-MIS model effectively encrypted the input images. Besides, IMLOSIE-MIS model attained high PSNR and low MSE values. For instance, in image 1, the proposed IMLOSIE-MIS model achieved a high PSNR of 68.306 dB and a low MSE of 0.098. Also, in image 3, IMLOSIE-MIS technique accomplished a high PSNR of 69.441 dB and a low MSE of 0.086. Besides, on image 6, the proposed IMLOSIE-MIS method attained a high PSNR of 66.547dB and a low MSE of 0.120.

Figure 3: Results of the analysis of IMLOSIE-MIS technique without attacks (a) MSE, (b) PSNR, (c) BER, and (d) SSI

Tab. 2 and Fig. 4 details about the results accomplished by IMLOSIE-MIS method in the presence of attacks. The table denotes that the proposed IMLOSIE-MIS technique effectively encrypted the input images. Also, IMLOSIE-MIS methodology acquired high PSNR and low MSE values. For example, in image 1, IMLOSIE-MIS approach attained a high PSNR of 67.543 dB and a low MSE of 0.107. Additionally, in image 3, the proposed IMLOSIE-MIS algorithm achieved a high PSNR of 69.340 dB and a low MSE of 0.087. Moreover, in image 6, the proposed IMLOSIE-MIS technique achieved a high PSNR of 65.333 dB and a low MSE of 0.138.

Figure 4: Results of the analysis of IMLOSIE-MIS technique under attacks (a) MSE, (b) PSNR, (c) BER, and (d) SSI

Tab. 3 and Fig. 5 shows the comparative MSE inspection results achieved by IMLOSIE-MIS method and other existing models. The results infer that the proposed IMLOSIE-MIS methodology produced less MSE values under all the images. For instance, on image 1, IMLOSIE-MIS technique obtained a minimal MSE of 0.098, whereas hybrid, Grasshopper Optimization (GO), PSO, Rivest, Shamir, Adleman (RSA), and AES models produced maximum MSE values such as 0.203, 0.375, 0.515, 0.695, and 0.857 respectively. Also, in image 3, the proposed IMLOSIE-MIS approach reached a minimal MSE of 0.086, whereas the hybrid, GO, PSO, RSA, and AES methodologies produced the highest MSE values such as 0.234, 0.361, 0.526, 0.66, and 0.795 correspondingly. In addition, on image 6, IMLOSIE-MIS technique attained a minimal MSE of 0.120, whereas hybrid, GO, PSO, RSA, and AES methods resulted in maximal MSE values such as 0.287, 0.425, 0.533, 0.679, and 0.853 correspondingly.

Figure 5: MSE analysis of IMLOSIE-MIS approach under distinct images

A comparative PSNR inspection was conducted between IMLOSIE-MIS model and other models under distinct images and the results are shown in Tab. 4 and Fig. 6. The experimental values indicate that the proposed IMLOSIE-MIS model achieved a superior performance with maximum PSNR values. For instance, in image 1, the proposed IMLOSIE-MIS model achieved a high PSNR of 68.306 dB, whereas the hybrid, GO, PSO, RSA, and AES models gained low PSNR values such as 61.981, 56.650, 53.895, 51.291, and 49.471 dB respectively. Concurrently, in image 3, the proposed IMLOSIE-MIS technique produced an improved PSNR of 69.441 dB, whereas the hybrid, GO, PSO, RSA, and AES methods accomplished low PSNR values such as 60.746, 56.981, 53.711, 51.740, and 50.123 dB correspondingly. Eventually, on image 6, the proposed IMLOSIE-MIS method resulted in improved PSNR of 68.306 dB, whereas the hybrid, GO, PSO, RSA, and AES approaches reached low PSNR values such as 58.973, 55.563, 53.596, 51.493, and 49.512 dB correspondingly.

Figure 6: PSNR analysis results of IMLOSIE-MIS approach under distinct images

A comparative SSI investigation was conducted between the proposed IMLOSIE-MIS approach and other existing methodologies under distinct images and the results are shown in Tab. 5 and Fig. 7. The experimental values denote that the proposed IMLOSIE-MIS technique outperformed all other methods and achieved a superior performance with maximal SSI values. For instance, on image 1, IMLOSIE-MIS method resulted in improved SSI of 1.000, whereas the hybrid, GO, PSO, RSA, and AES models gained low SSI values such as 0.993, 0.940, 0.936, 0.845, and 0.752 correspondingly.

Figure 7: SSI analysis results of IMLOSIE-MIS approach under distinct images

Simultaneously, on image 3, the proposed IMLOSIE-MIS methodology achieved an improved SSI of 0.999, whereas hybrid, GO, PSO, RSA, and AES techniques accomplished low SSI values such as 0.998, 0.963, 0.970, 0.817, and 0.728 correspondingly. Finally, on image 6, IMLOSIE-MIS method resulted in enhanced SSI of 1.000, whereas the hybrid, GO, PSO, RSA, and AES models attained low SSI values such as 0.999, 0.973, 0.966 B, 0.894, and 0.752 correspondingly.

5  Conclusion

In this study, a new IMLOSIE-MIS technique has been developed to accomplish security in IoT-enabled medical image transmission process. The proposed IMLOSIE-MIS model primarily generates a set of shadow images for the applied medical images. Then, IMLO-HE technique is implemented to encode the shadow images in an effective manner. At the same time, the optimal generation of the keys involved in HE technique is accomplished by MLO algorithm in such a way that the PSNR values get maximized. At the receiver point, decryption process is initially carried out along with shadow image reconstruction process. The experimentation analysis was carried out on the medical images and the results exhibit that the proposed IMLOSIE-MIS model achieved better results than the existing techniques. The comparison analysis results demonstrate that IMLOSIE-MIS model is robust and offers high security in IoT-enabled healthcare environment. In future, disease diagnosis module can be included in the proposed model to enable medical image classification process.

Funding Statement: The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for funding this work through Small Groups Project under Grant Number (241/43). Princess Nourah Bint Abdulrahman University Researchers Supporting Project number (PNURSP2022R319), Princess Nourah Bint Abdulrahman University, Riyadh, Saudi Arabia. The authors would like to thank the Deanship of Scientific Research at Umm Al-Qura University for supporting this work by Grant Code: (22UQU4340237DSR30).

Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the present study.