AFD and chaotic map-based integrated approach for ECG compression, steganography and encryption in E-healthcare paradigm

The proliferation of tele-healthcare services at an accelerated rate raises concerns over the management,security and privacy of the patient's confidential data (an individual's personal details and medical biography) during its transmission and storage. To resolve these issues, amalgamation of three fundamental techniques of remote healthcare systems, that is, signal compression, data hiding and encryption, are proposed. The proposed approach applies the recently developed adaptive Fourier decomposition (AFD) technique to decompose the electrocardiogram signal in terms of adaptively selected basis functions from the orthogonal rational function that performs a high fidelity compression. Later, chaotic map-based steganography and encryption are proposed on the AFD coefficients to secure the confidential information and the signal itself. The performance of the three processes is evaluated in terms of distortion (both statistical and clinical), compression [(compression ratio (CR) and quality score], steganography [embedding capacity (EC), bit error rate], and encryption (sensitivity, predictivity, correlation coefficient). By imple-menting on 48 records of Massachusetts Institute of Technology-Beth Israel Hospital arrhythmia database and varying N from 15 to 120, the proposed work achieves average CR, EC, and percentage residual difference of 62.39–11.79, 7 � 10 − 3 –6 � 10 − 2 and 3.77–0.32,


| INTRODUCTION
Advancement in communication technology has changed the conventional healthcare systems where experienced medical practitioners provide remote healthcare solutions to homebound patients or patients in emergency without their physical interface. Patient's medical health records, such as physiological signals or medical images, are sent to the experts so that appropriate action can be advised without any delay. Along with this, their personal details, clinical reports and medical history are also communicated for the purpose of identification and methodical diagnosis. Transmission of the patient's vital information over the public network raises the concern for its security and privacy. Moreover, the continuous monitoring of the remote patients requires long-term recordings that pose a huge burden on the transmission and storage devices. Altogether, it is observed that there are three major hitches in growth of e-healthcare-oriented services: (i) management of enormous electronic health records (EHR) during transmission and storage; (ii) privacy of the patient's personal details (name, age, sex, id number, contact number, doctor referred etc.) and their metadata (pathological reports, medical history, medication etc.) and (iii) security of the patient's EHR. All these issues can be resolved by signalprocessing techniques such as compression, steganography and encryption, respectively. Electrocardiogram (ECG), which is considered as a primary signal for assessing cardiac functions, detecting other health ailments [1,2] and a prerequisite during surgeries [3], is marked as a prime signal in the e-healthcare paradigm. Various remarkable approaches have been reported in literature to perform ECG compression in time domain [4], transform domain [5,6], through the feature extraction method [7]; ECG steganography in spatial domain [8], transform domain [9,10], while many approaches perform ECG encryption [11] to avoid any illicit access by eavesdroppers during transmission. But these approaches are designed to address only a single issue. Later, researchers worked on joint approaches but applied only two of these processes such as compression and encryption [12,13], steganography and encryption [14], or compression and steganography [15]. This requires a separate technique to perform the third process which raises the compatibility issue and increases the complexity too. There is a need to develop a technique which is capable of performing the three processes conjointly. This study proposes a novel technique that addresses all the three issues, that is, compression, steganography and encryption which are mandatory to thrive the telehealthcare services.
A recently developed adaptive Fourier decomposition (AFD) approach is used to decompose the ECG signal into AFD coefficientsthatpossesssuccessivelyincreasingnon-negativeanalytic instantaneousfrequencies [16,17].AFDadaptivelyselectthebasis functions from the rational orthogonal system, for example, the Takenaka-Malmaqui1st(TM)systeminaccordancewiththeinput signal thus making it capable to reconstruct the high-fidelity signal withfewcoefficientsonly.Thelocationofbasisfunctionswithinunit diskmakesitfeasibletohidethe patient'sconfidentialinformation comfortablyinthem.Later,encryptionisperformedinthesestego-basisfunctionsaswellasinAFDcoefficientstosecurethecoverECG signal.Chaoticmapswhichgeneraterandomsignalsareemployedin these processes to boost the security. The encrypted AFD coefficients and AFD coefficients are finally encoded with the proposed adaptive bit length encoding (ABLE) technique for transmission. At the receiver end, the reverse of these processes is performedtoreconstructtheoriginalECGsignal.

| Adaptive Fourier decomposition
AFD is a recently established signal-processing technique and it is a generalisation of the existing Fourier decomposition system. The conventional Fourier series provides information about the range of frequencies present in the signal at a given time period. However, it fails to provide any information about the exact time instant of these frequencies. Many techniques such as discrete wavelet transform (DWT), short-time Fourier transform (STFT) or Gabor transform decomposes the signal, in both frequency and time domain, but these techniques have pre-defined, fixed basis functions which intrinsically limit their performance [17]. AFD solves the aforementioned issues by adaptively selecting the basis function from the rational orthogonal TM system, based on the input signal. The input signal is decomposed into series of non-negative mono components, which clearly defines the time-frequency distribution with most desirable properties. The mono components possessing the maximum energy are selected using the maximal projection principle (MPP) combined with the greedy algorithm. The generalized mathematics involved in AFD is explained below: AFD works in Hardy's space H 2 (D) where ℂ is the complex plane. The rational orthogonal TM system, which AFD uses for its basis function is: M n ðzÞ ¼ M fa 1 ; a 2 ;…a n g ðzÞ M n ðzÞ ¼ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi where a n are the complex parameters in the unit disc D. Let S be the input signal such that S H 2 (D). The AFD decomposition of S is expressed as: where e fag is the normalised reproducing kernel called as evaluators at a and used to generate dictionary: e fag ðzÞ ¼ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi r N symbolises the standard remainder defined as: By using Cauchy's integral formula, ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi The first level decomposition can be expressed as: where a ¼ a 1 is adaptively selected using MPP such that: The standard remainder r 1 after first decomposition is: and is calculated as: where S 2 is also in H 2 space. Furthermore, S 2 can be written as: For (k + 1) th iteration, S kþ1 is calculated from S k using recursion as: where a k is selected with MPP. Thus, in AFD, sifting process is employed with which S kþ1 is computed from S k by finding maximum a k for: Finally, after N iterations: N is the total number of decomposition levels and r n is remainder left after N iterations. For reconstruction, S is backward projected to obtain the reconstructed real valued signal S R as where r 0 is the zero th order Fourier coefficient of S. The difference between S and S R is r and the value of r decreases with increase in the number of decomposition levels N.

| Chaotic sequence generator
Chaotic sequences are the random yet deterministic sequences that are highly sensitive to their initial conditions and control parameters. In the proposed technique, these sequences are employed during steganography to enhance the security of the concealed information. Presently, various mathematical functions exist that are capable to produce these random sequences. However, in this work 1-D coupled chaotic maps (CCM) that exhibits random behaviour over wide range of initial and control parameters are employed [18]. The mathematical equation of CCM for the initial value (u o ), control parameter (v o ) and length (L) is given as Chðu o; v o; lÞ : The chaotic behaviour of the CCM for different values of control parameters (v o ) and initial conditions (u n ) can be observed from the bifurcation diagram in Figure 1. The chaotic behaviour of CCM varies over a complete range of v o between (0,4) for u o in the range (0,1). The randomness properties of CCM were examined in Equation (8) using the National Institute of Standards and Testing statistical test suite [19].
The chaotic sequence generated by Equation (10) has real floating point values. To obtain integer-valued chaotic sequence, Ch is modified by performing the sort operation on it.
The operation arranged the magnitudes of Ch in the ascending order while their original locations are procured to be used as integer-valued chaotic sequence.

½XI� ¼ sort seqðChÞ ð11Þ
Here, X holds the sorted magnitudes of Ch while I contains the original indices of the sorted magnitudes. This is explained with the help of an example.
Let the chaotic sequence obtained from CCM is Here, I is the new chaotic sequence that consists of integer values and is applied in this work for data encryption and steganography.

| ECG database
For the comprehensive analysis of the proposed approach, ECG signals with different abnormalities are acquired from various standard databases available online [20] as well as self-recorded in the laboratory. The details of the databases used are mentioned in Table 1.

| METHODOLOGY
Here, a novel integrated approach is proposed which is capable of performing three basic signal processing techniques: ECG compression, steganography and encryption. These techniques are essential for successful implementation of e-healthcare services. AFD is applied to decompose the ECG signal into complex AFD coefficients and their respective basis functions each of length N, where N is the number of decomposition levels. The ciphered bits of patients'' confidential information are then concealed into the basis functions. The stego-basis functions and AFD coefficients are further encrypted, encoded and transmitted over the communication channel (as shown in Figure 2(a)). The number of decomposition levels or iterations performed to decompose the ECG signal ascertains the precision of the reconstructed signal, compression efficiency and embedding capacity of the signal. The implementation of AFD technique on long-duration input signal overruns the computational capacity of the system. To avoid this issue in the proposed technique. AFD is applied on the non-overlapping ECG blocks of fixed length and the AFD coefficients (M i ) and their basis functions (R i ) obtained from each block are stacked vertically to form two-dimensional (2D) matrices (M) and (R), respectively.

| AFD implementation for compression stage
The proposed technique employs AFD to compress the ECG signal, where ECG is initially divided into nonoverlapping blocks of size L. The real-valued ECG samples (S) in each block are initially projected into the Hardy's space (H 2 (D)) via Hilbert transform to form S D . Furthermore, the process discussed in Section 2.1 is applied to implement AFD but instead of using the greedy algorithmbased MPP, fast fourier transform is applied which reduces the computational complexity of the algorithm [21]. The M i and R i of the length N, obtained from the i th block are stacked vertically to form 2D matrices (M) and (R), respectively. The process of decomposition of ECG signal into limited number of AFD coefficients and basis functions is explained in Algorithm 1. ⌊x⌋ represents greatest integer less than x.

| Ciphering of Confidential Information
Although steganography is performed to provide security to the patient's sensitive information (A), and to enhance the security of the sensitive information further, ciphering is performed prior to steganography. For this, an XOR The ECG signals were recorded from the local population in the Biomedical Signal Processing laboratory at Department of Electronics and Communication Engineering, Dr B R Ambedkar National Institute of Technology, Jalandhar India. The recordings were acquired on lead II using BIOPAC ® MP150 under standard conditions in a quiet room; at comfortable light and temperature levels.

-
operation is performed between the ASCII equivalents of the confidential elements (A) and the integer valued chaotic sequence (I 1 ) where I 1 is generated from G with initial conditions G(u G ;v G ; L c ); L c is the length of the confidential information. The ciphering process is performed as: Ciphering bits (C_bits) = XOR(A, I 1 )

| Steganography stage
The output of compression stage consists of two sets of 2D matrices: AFD coefficients (M) and basis functions (R). The coefficients in M spreads over wide range in complex domain, whereas the values in (R) lies only in unit disc space [-1, 1]. Figure 3 displays the placement of coefficients M (excluding first element of each row) and R in complex plane. The uniformity in magnitudes of the basis functions (R = R R + jR I ) < ±1 makes them suitable to explore for data hiding. The process of hiding data in real (R R ) and imaginary (R I ) components of R is discussed in Algorithm 2. For steganography, both the matrices R R and R I , of size (a),(b) where a and b are their number of rows and columns, respectively, are initially divided vertically into two halves with b/2 columns in each half. Embedding is done alternately in R R and R I on the criteria that if the ciphered bit is 1, then 1 is added to the chaotically selected R R or R I from first half and then swapped with another chaotically selected R R or R I in the second half else if the secret bit is 0, then simple swapping is done between chaotically selected R R or R I from two halves. For random selection of the basis functions, two integer-valued chaotic sequences (I 2 and I 3 ), generated using Equations (10) and (11), are employed.
F I G U R E 2 (a) Compression, steganography and encryption processes at the transmitter side and (b) decryption, extraction of secret bits and decompression at the receiver side. AFD, adaptive Fourier decomposition SONI ET AL.

segregated real and imaginary parts into a complex form
The use of chaotic sequences in selection of R for embedding and swapping adds unpredictability in the stegobasis functions that strengthens the security of the hidden information. This method executes reversible steganography as the basis functions restore their original values after extraction of secret bits. The impact of embedding the ciphered bits in basis functions is also well observed in Figure 3. The original basis functions are organised near the unit circle, whereas the stego-values are scattered all over the unit disk.

| Encryption process
Next step is to safeguard the cover ECG signal from any tampering by the eavesdroppers. For this, confusion is added in stego-basis functions and AFD coefficients by swapping elements of randomly selected rows and columns of one-half of the matrices with the elements of their second-half. Two F I G U R E 3 Illustration of impact of steganography and encryption on adaptive Fourier decomposition (AFD) coefficients and ECG reconstruction with and without extraction of secret bits and decryption when observed on record 100 of MIT-BIH arrhythmia database at N = 75 chaotic sequences are used in the random selection of the rows and columns. The encryption process is explained in Algorithm 3. The impact of encryption on Rs and M can be visualised in Figure 3. Although the elements of Rs and en_Rs form a similar pattern, their locations are entirely different. It is also well illustrated in Figure 3 that the ECG signal reconstructed with encrypted values (en_Rs and en_M) is completely distorted while the ECG signal recovered after decryption and extraction of secret bits is similar to the original signal. Generate two chaotic sequences K1 and G with initial conditions

| Encoding stage
The encrypted AFD coefficients (en_M) and the basis functions (en_Rs) consist of complex floating-point numbers and it is a challenge to encode them into a suitable binary format for transmission. Here, encoding is performed in two steps: 3.4.1 | Quantization process en_Rs and en_M are in complex format and are initially segregated into their real and imaginary parts for encoding.
The obtained floating point numbers are then truncated and quantised to convert them into integer values. This conversion of complex coefficients to binary integers is explained in Figure 4. The real and imaginary parts of en_Rs and en_M are separated as R R , R I and M R , M I, respectively. Their signs are segregated as sign_R R , sign_R I , sign_M R, and sign_M I and the absolute magnitudes are converted into 1D as Bits required to encode the largest number of the sequence, that is, log 2 (2515) is 12; therefore, total bits used to encode the whole sequence is 12 � 5 = 60 bits.
The The highest number in M RE and M RM is 25 and 15, respectively, and minimum number of bits required to encode them are 5 and 4, respectively. Hence, the total bits used to encode the sequence are (5 � 5 + 4 � 5) = 45 bits which are sufficiently small to improve the CR. Here, d1 = 3 and d2 = 2 are considered.
The selection of the block length (bk) directly affects the length of the encoded bits and eventually impacts the compression ratio (CR). If bk is small, the input sequence will be divided into more number of blocks, hence more number of headers will be inserted in the modified sequence. This will directly increase the length of the encoded bits. On the other side, higher bk increases the number of sequence elements to be encoded with the max_no of bits which again shoot up the length of encoded bits and reduces the CR. Figure 5 displays the impact of bk on CR when measured on all 48 records of MIT-BIH arrhythmia database of 5 min. Figure 5 clearly illustrates the correlation of bk and CR. When bk is small (<10), the CR is less (<16), with the highest CR observed at the bk of 54. Interestingly, with further increase in bk, a reduction in CR is observed. Therefore, this trend highlights the fact that the selection of block size must be optimum to fetch maximum CR. The decoding of this ABLE technique is performed following the inverse process.

| Receiver side
The process followed at the receiver side is illustrated in Figure 2 (b). It consists of the following steps: 1. Decode the received binary sequence to obtain the encrypted stego-basis functions and encrypted-AFD coefficients. 2. Generate chaotic sequences K, Q1, Q2 and G using same initial conditions K(u k ;v k ; a=2) and Q1(u 1Q ,v 1Q ; aÞ and Q2( u 2Q ,v 2Q ; b=2) and G(u G ;v G ; L c ). 3. Decrypt en_Rs and en_M. 4. Extract secret bits from the chaotically selected locations and reverse swapping. 5. Take inverse AFD and reconstruct 1D decompressed ECG.

| RESULT AND DISCUSSION
The major focus of the signal-processing techniques, when implemented on the medical data, is to preserve its diagnostic features. As discussed earlier, the proposed technique performs three major operations: compression, steganography and encryption. All these operations are crucial to securely communicate the patient's medical as well as personal information in a remote-healthcare paradigm. However, it is also important to study the impact of these techniques on the diagnostic features of ECG signal. Figure 6 displays the first 3000 samples of (a) original (b) encrypted ECG and (c) decrypted and decompressed ECG signals of record 100 of MIT-BIH arrhythmia database at N = 30, 75 and 120. The visual inspection validates the similarity of the original signal with the decrypted and decompressed signal. On the other hand, the encrypted signals are observed as distorted. The impact of N on the quality of the reconstructed signal is also visible with better performance at N = 120 as compared with N = 30. The analysis of the signals through visual inspection is feasible for short durations only. To observe the ECG signal of long durations, it is preferred to opt for computer-aided analysis. The performance of the proposed technique is evaluated mathematically in terms of statistical and clinical parameters. The statistical measures like percentage residual difference (PRD), normalised PRD (PRDN), mean PRD (PRD1024), signal to noise ratio (SNR), peak signal to noise ratio (PSNR) and Kullback-Leibler Divergence (KL-Divergence) [10,22] are used to measure the variation in the two signals, but it distributes the error equally over the complete signal. Clinically acceptable measures such as wavelet-based  weighted PRD (WWPRD) [22] and wavelet energy-based diagnostic distortion (WEDD) [23] computes the local distortion in the processed signal; hence significant in clinical distortion. In addition, the process specific parameters such as CR and quality score for compression [13], embedded capacity (EC) and bit error rate (BER) for steganography are also investigated. Here, EC is computed as the ratio of number of bits embedded to the number of ECG samples in original ECG signal. As discussed earlier, the reconstruction of the ECG signal with the encrypted values disturb the signal morphology ( Figure 6 (b)). The amount of distortion caused in the ciphered signal is evaluated in terms of Selectivity (Se%), positive predictivity(Pe%) and correlation coefficient (CC).  Table 2. As the number of ECG samples in the fixed time duration increases with the sampling frequency, the number of bits embedded also increases. As summarised in Table 2s, the number of bits embedded in self-recorded data with the sampling frequency of 500 Hz is the highest, that is, 5550 compared with the bits embedded in other databases with lower sampling frequencies.

| Impact of AFD decomposition level (N)
The optimal selection of decomposition levels (N) directly influences the performance of all the three processes implemented here. Table 3 demonstrates the impact of N on the performance metrics when measured on all 48 records of the MIT-BIH arrhythmia database for the 5-min duration. It is observed that distortion (PRD, WWPRD and WEDD) has an inverse correlation with N contrary to signal quality (SNR and PSNR) which varies directly with N. CR and EC are the prime attributes of compression and steganography algorithms, respectively; and in this combined approach, both behave in contrast relation with N, that is, EC increases but CR decreases with increase in N.
Henceforth, a trade-off is required between EC and CR such that in cases where high CR is a requisite, N is kept low; whereas N is high for higher EC and better reconstruction quality. It is observed in Table 3 that QS of 38.42 is maximum at N = 75, therefore here, N = 75 is considered for experimentation. The proposed technique is efficient to perform compression, steganography and encryption processes individually as well as jointly by modifying certain parameters. In the subsequent sections the performance of each process is analysed separately.

| Compression analysis
The key concern in lossy ECG compression techniques is to conciliate between the CR and the reconstruction quality. An average CR of 62.39 to 11.79 and PRD of 3.77 to 0.32 as N varies from 15 to 120 illustrating the linear and robust CR-PRD relationship (see Table 3). The compression is nearly lossless at higher values of N. A comparison with the recently reported ECG compression techniques in the transform domain is shown in Table 4. All the techniques were evaluated on the MIT-BIH arrhythmia database. The proposed technique exhibits comparable CR, PRD and QS despite implementing combined techniques of compression steganography and encryption.

| Steganography analysis
In the proposed approach, steganography is performed by modifying and swapping both real and imaginary components of the complex basis functions. The amount of bits embedded depends on the number of basis functions obtained from decomposition which ultimately depends on N. For higher EC, N is kept high. The readings in Table 3 depict the increase in number of bits embedded and correspondingly the EC with increase in N. It is observed that bits embedded =756 at N = 15 are raised to 6.480 Kb at N = 120. To further improve the EC, the window size (L) of the ECG signal can be reduced for the same N. This increases the number of ECG segments and decomposing each segment into N coefficients, raises the number of basis functions and hence the EC. Moreover, the proposed approach performs reversible steganography which retrieves same basis functions after extraction of secret bits, hence does not cause any distortion. The novelty of the steganography process in the proposed approach is that the increase in N increases the EC as well as improves the quality of the reconstructed signal. This is contrary to the existing steganography techniques in which the signal quality degrades with increase in EC. A comparison of the proposed technique with the existing steganography techniques is carried outin Table 5.
Despite performing compression with CR = 11.79, EC = 0.06 at SNR = 50.19 is achieved by the proposed technique at N = 120 which is better as compared with EC = 7.8 � 10 −3 and SNR = 32 obtained in [9]. The EC, PRD, PSNR and KLdivergence of 0.06, 0.32, 53.43 and 5.52 � 10 −6 , respectively, obtained by the proposed technique are also better than 0.058, Abbreviations: CR,compression ratio; PRD, percentage residual difference; SNR, signal to noise ratio; QS, quality score.

T A B L E 3
The impact of N on the average performance metrics measured on all 48 records of the MIT-BIH arrhythmia database of 5 min

| Encryption analysis
Encryption in case of physiological signals is reflected and evaluated in terms of the loss of its diagnostic features. The amount of distortion occurred in ECG morphology due to encryption can be observed in Figure 7 where Figure 7(a) illustrates the comparison between the original ECG signal and the encrypted ECG signal obtained when decompressed with encrypted-AFD coefficients and basis functions while Figure 7 (b) compares the original signal with decrypted and decompressed ECG signal.

| Security analysis
To quantify the impact of encryption on the ECG signal as well as to verify the strength of the proposed algorithm against stego-attacks, various parameters are studied.
4.5.1 | Sensitivity (Se%) and positive predictivity (Pe%) In ECG, R-peaks are pivotal in diagnosis of various heart ailments, therefore the extent of distortion occurred in the R-peaks of the reconstructed ECG is computed from the R-peaks count in terms of Se% and Pe%. Se% gives the percentage of correctly detected heartbeats while Pe% determines the percentage of real true heartbeats and defined as [1,27]: where TP (true positive) is the number of correctly detected Rpeaks; FN (false negative) is the number of undetected R-peaks and FP (false positive) is the number of falsely detected Rpeaks. Here, the R-peaks are detected using the Pan-Tompkins algorithm [28].

| Correlation coefficient
The correlation coefficient (CC) is another security measure that computes the similarity between the original signal and reconstructed signal [22,27] and ideally the correlation coefficient of the encrypted signal should be zero.

| Key space analysis
Another parameter to measure the security strength of an algorithm is its key space. The brute force attacks can be easily avoided with larger key space. In the proposed algorithm, the key consists of the initial conditions and control parameters of four chaotic sequences that are used during implementation of various processes as well as the values of N, L and bk. If the initial values and control parameters are set to 14 decimal places then according to the IEEE 754 standard [29], 64 bits are required to encode each parameter. The values of N, L and bk requires 8, 12 and 7 bits, respectively. Therefore, the total length of the key will be (64 � 8) + 8 + 12 + 7 = 539 bits which is sufficiently large as compared with the minimum key space of 128 bits required to avoid brute force attack [18].

| Differential attacks
Number of pixel change rate (NPCR) is an important parameter used to measure the resistance of encryption algorithm against differential attacks [18]. The theoretical range of NPCR is 100. To analyse the strength of the proposed algorithm against differential attacks, NPCR is computed on first 10,000 samples of records 100, 117 and 119 of MIT-BIH arrhythmia at N = 75 and varying values of L. The results obtained are displayed in Table 7 and it is found that NPCR is 75% at L = 2000 while it is 100% for higher values of L.

| Computational complexity
The complexity of the proposed technique is measured in terms of the time taken by the algorithm to execute the complete algorithm including the three processes: ECG compression, steganography, encryption and then encoding. The time elapsed in performing these operations on record 100 of MIT-BIH arrhythmia database of 5 min duration at N = 75 are 18.5043, 6.338, 0.0238, and 0.8861s, respectively. The simulation of the proposed technique is done in Matlab R2017a on the system having configuration of Intel i5-4210U CPU@1.70GHz.

| Comparison of the proposed technique with the existing techniques performing joint approaches
The proficiency of the proposed technique performing combined operations of compression, steganography and encryption on the ECG signal is compared with the recently published state-of-the-art techniques. As observed from the survey, very limited researchers implemented integrated approaches on biomedical signals. As per the author's knowledge, no work is reported in literature that incorporates all the three processes together on ECG signal. However, the recently published techniques that executed at least two processes are compared with the proposed work in Table 8. It is obvious from the table that, despite performing the three processes, the proposed technique exhibits remarkable results as compared with the techniques performing only two operations. At nearly the same CR as in [13], low PRD = 0.48 is achieved by the proposed technique at N = 69 despite embedding the secret bits in comparison to PRD = 1.06 obtained in the existing technique. In [14], EC = 3.2 � 10 −3 is quite low at high PRD = 4 obtained from ECG signal of 30 min as compared with relatively high EC of 6 � 10 −2 at negligible PRD% = 0.32 achieved by the proposed technique. When EC = 0.8, PRD of 9 attained in [15] is quite high at CR = 5 as compared to PRD = 0.23 achieved by the proposed technique at high CR of 7.81 when N is set to 145. The EC obtained with this technique is high but the author did not discuss about BER occurred during extraction. Further in [30], EC and CR achieved are 7.8 � 10 −3 and 1.96, respectively, that are noticeably low in comparison with the results displayed by the proposed technique on the same database. Despite performing steganography, nearly the same CR is obtained at N = 22 in the proposed work as compared with the results obtained in [12].

| CONCLUSION
This novel work proposed a blend of three operations: compression, steganography and encryption that are indispensable to strengthen the tele-healthcare services. The recently developed AFD technique demonstrated to successfully decompose the ECG signal into the limited number of complex coefficients and concealing the confidential data into these coefficients perform steganography and encryption of the ECG signal. Furthermore, the proposed ABLE method seems to be very promising in encoding the complex AFD coefficients for transmission of medical data. The chaotic sequences incorporated during steganography and encryption also supplements the security. The performance of the proposed technique is evaluated on standard and self-recorded databases including both normal as well as abnormal records. An average CR, EC, PSNR and PRD achieved varies as 62.39-11.79, 7 � 10 −3 -6 � 10 −2 , 32.63-53.43 and 3.7-0.32, respectively, when decomposition levels (N) vary from 15 to 120. The convincing results obtained from the proposed technique compared with the currently available techniques indicate a major milestone achieved for data compression along with privacy and reliable transfer of medical records across different geographical locations. This approach has the potential to become a reliable precursor for the transformation of medical information exchange in the fast changing digital world.