Enhanced secure transmission of data in WBAN with predictive model for health care applications

Abstract

Wireless Body Area Network (WBAN) has been emerged as a recent development in technology, which provides convenience to monitor patient health remotely. E-healthcare management has largely promoted by WBAN. It eases the life style of patient and refrain them from visiting hospital regularly. At present, E-healthcare is facing the issues of secure information transmission over public network and assistance to hospital staff for disease identification based on information collected from wearable sensors. In this paper, to preserve confidentiality of messages, knapsack cryptosystem based solution have proposed. To authenticate transmission of data in wireless body area network, we have proposed an improved batch verification based digital signature scheme that improve the computational efficiency of underlying algorithm. Computing one-way hash for generation of digital signature is not required in proposed algorithm and hence provide a lightweight security mechanism suitable for resource constraint wearable sensors. A mechanism has proposed which facilitate the doctors to identify the disease probability by a probabilistic method of logistic regression. Further, a tree based batch validation approach has also proposed for vulnerable network. To validate the security of proposed scheme, various attacks have verified that the scheme has been capable to resist. Performance improvement in digital signature verification scheme has demonstrated and validated using implementation results.

Keywords

Wireless body area network logistic regression prediction model batch verification knapsack crypto system security privacy and authentication recovery based digital signature

1. Introduction

Recent technological advancements have enabled a new variant of Wireless Sensor Networks, the Wireless Body Area Network (WBAN); which is provisioning remote monitoring of patients carrying wearable sensors. The networks are able to transmit the health related information to the processing body wirelessly, which makes such networks to suffer with the security threats and may in turn cost to the ultimate beneficiary viz. the patients. Hereby, data authentication and confidentiality become the primary concerns in the WBAN.

There are two working scenarios for the concerned application of patient monitoring using WBAN-indoor monitoring i.e. when patient is inside the hospital premise and outdoor monitoring where patient is outside the premise. Each scenario requires different level of security treatments. In search of probable solutions, cryptographic possibilities have investigated in detail and it has found that the knapsack crypto system could be applied on indoor patient monitoring and the scheme of digital signature may be suitable solution for outdoor scenario.

Knapsack crypto system is derive for a NP-complete problem-knapsack problem where there are n integers and a given weight called $k$ . The problem consists of determining integers whose sum is equal to the given weight $k$ . The solution employs a super increasing vector to encipher the message, which cannot be decrypted in the absence of private legitimate parameters specific to the legal receiver by an intruder. Moreover, the decryption process has explained later in the proposed scheme section in the article. The scheme when implemented for the indoor patient [18] monitoring using WBAN ensures confidentiality of message; however, since both the server and the base station are situated in the same hospital premise, authentication is no more required in this particular scenario.

On the other hand, outdoor patient monitoring requires more secure and high performing algorithm for which a number of recovery based digital signature schemes have been investigated [18]. The recovery based digital signature schemes do not require separate transmission of message as the message is assorted into the signature itself. Transmitted message is first recovered from the digital signature and then is validated using sender’s public key. The scheme of verifying each signature independently results into a huge increase in computation time. Such demerit of independent signature verification prompts to device a recovery based batch verifying digital signature scheme without any significant increase in communication overhead and in latency. A scheme is being proposed here is the later section which ensures perfect resilience to the packet loss while keeping the above cited parameters in consideration. The proposed one is combined with knapsack crypto system in order to provide a solution towards authentication and confidentiality of message for outdoor patient monitoring using WBAN.

Fore prediction of disease, a number of techniques have been investigated [19] and logistic regression has been applied here. Linear regression is a statistical method used to estimate relation among variables for predictive analysis. It is used to analyze several variables where one variable is dependent upon other variables. Simplest linear regression has been defined using the equation with one independent and one dependent variable as follows:

$\displaystyle Y=\alpha*x+\beta$ (1)

where $Y$ is estimated dependent variable, $\alpha$ is constant, $x$ is an independent variable and $\beta$ is regression coefficients.

Logistic regression applies sigmoid function over linear regression to compute probability in a variable. This method has been applied with security mechanism to assist the hospital staff for disease identification.

An improvement in batch digital signature verification scheme has also been proposed if batch verification fails. When batch validation fails then instead of verifying each message independently, a tree based approach has been proposed to make the scheme efficient even if few malicious messages are received by a receiver.

2. Literature review

Wireless Body Area Networks has evolved as a major variant of WSN, which allows remote monitoring of patients by the hospital staff. The architecture of WBAN can be seen as a hierarchical one, comprising of three different tiers viz. Tier-1, Tier-2, and Tier-3 [1]. Tier-1 locates the patients with sensors mounted on their respective bodies; Tier-2 contains relay nodes and deals with network connectivity; whereas, Tier-3 embodies the hospital cloud and the server.

Due to its intrinsic nature of being wireless and ad-hoc, WBAN suffers with a number of security issues viz. authentication, confidentiality, network availability. In addition, thereby the network is exposed to several security threats e.g. unauthorized access, disclosure of message, message modification, routing attack etc. [2]. Solutions available to such threats include hardware encryption, tinysec, biometric authentication, Bluetooth security etc. Besides such solutions, secure key management algorithms, cryptography based key management viz. ECC, RSA and patient’s psychological parameters based key management have also been observed as worthy solutions towards establishment of secret key between the communicating entities in WBAN. In our work, public key cryptography is used to ensure confidentiality, authentication and avoidance of other attacks.

Knapsack crypto system [15] is a well-known public key cryptography scheme which was primarily derived for the knapsack problem. It adds security features to the sensitive information by transferring the message using sender’s public key which only receiver can decrypt using its private key. Knapsack crypto system is implemented for indoor patient monitoring in our proposed work.

Shieh et al. [3] proposed a recovery based digital signature scheme suitable for resource-constrained devices, which does not have hash functions to compute the digital signature. The message is implanted within the digital signature during the respective phase of signature creation at sender side. The receiver recovers the message from the received digital signature, which is then validated by using the public key of sender. However, the original scheme does not protect against the message theft, as the message can easily be recovered using transmitted parameters and sender’s public key. Moreover, numerous attacks have been identified in [4, 5, 6, 7] due to forgery attack, cryptanalysis, and impersonation. In the scheme proposed by Kang and Tang [8], resistance had made against all the known attacks and it is based on the basic concepts of signing and independent message verification. It is obvious that the message verification cost will be high at the receiver end due to this independent treatment of messages.

Several schemes in [9, 10] have proposed towards decreasing the verification cost of the messages. However, all such schemes generate the digital signature on a block of messages but at the expense of increasing communication overhead or susceptibility to packet loss. Loss of messages due to packet loss is inevitable as it fails the signature verification for the entire block. Later on, a scheme [11] is devised which is resistant to packet loss to a certain extent as hash of message are concatenated with multiple message. Therefore, there exists a trade-off between the packet loss and redundancy of hashes, which incurs to communication overhead. A variant of RSA, batch RSA [12] proposed by Bellare et al. does not have any dependency among the messages and hence, does not imposes any communication overhead; but, it also provides no protection against the repudiation attack [10].

The healthcare sector have collected huge amount of data related to disease in recent years [16]. Grounded upon the data, [16] have proposed principle component analysis based scheme to find minimum number of attributes required to enhance precision of machine learning algorithms. A number of supervised learning algorithms have been discussed to predict heart disease based upon selected components.

A disease prediction models based upon convolution neural network have been proposed in [17] which uses structured and unstructured data from hospital. It shows the impact of regional disease on prediction algorithms. Author has also focused on the issue of low quality of data which leads to weak prediction of disease.

3. Proposed work

In the proposed work, machine learning based logistic regression technique has been used to compute the probability of disease based on health information like EEG values, ECG values collected by body sensors. This probability will provide assistance to medical staff in predicting the disease based upon a model. The encrypted data, along with this probability, will be transmitted through the network as per following scenarios.

3.1 Scenario 1: When the patient is inside the hospital

Once the patient admitted to the hospital, he would diagnosed by the doctor; thereafter, a number of sensors would attached with the patient’s body. Since the patient is within the hospital premise, installation of any complex and time consuming encryption system is not required as the data will not be routed outwards i.e. to the Internet. Moreover, sensors attached to the body of patient will not be receiving any data and the patient is to be diagnosed only on the basis of output made by the implanted sensors to the hospital.

The entire working of the proposed scheme can be divided into a number of phases as follows:

1.
Registration phase: In the registration phase, every sensor has assigned a unique ID, which is stored in the hospital database along with the patient’s detail. Such given IDs will be used to authenticate sensors by the hospital.
2.
Encryption phase: This phase implements the Knapsack based cryptosystem in this scenario. The steps followed by a hospital server are as follows:

a. a.
An arbitrary super increasing Knapsack vector say $A=\{a_{1},a_{2},a_{3},\ldots,a_{n}\}$ , such that $a_{j}=\sum_{i=1}^{j-1}a_{i}$ is generated.
b.
A number $Z$ is chosen such that the summation of the elements of $A$ is less than $Z$ ; Moreover, a positive integer $T$ is also chosen such that: GCD $(T,Z)$ is 1.

Table 1
Logistic regression notation

Notation Description

$H_{T}(X)$ Predicted value of output variable

$T_{i}$ (where $i\in(0,1,2,\ldots,n)$ Coefficient, $T_{0}$ called bias and $T_{i}$ (where $i>$ 0) called weight assigned to input variables

$X_{j}$ (where $j\in(1,2,\ldots,n)$ Input variables (sensor information in our paper)

$g(H_{T}(X))$ Sigmoid function over $H_{T}(X)$

$P$ Probability

c.
A vector $B=\{b_{1},b_{2},\ldots,b_{n}\}$ is computed as per the formula

$\displaystyle b_{i}=Mod(Ta_{i},Z)$ (2)

for $i=1,2,\ldots,n$ .
d.
The hospital server keeps parameters – $Z$ , $A$ , $T$ private and $B$ will act as public key and made open to all the sensors.
e.
Sensors will perform the encryption on any message, to be transmitted to hospital server in the following manner:

iii. i.
Message ‘ $m$ ’ together with the sensor ID and timestamp is split into block of $n$ (bits) where is ‘ $n$ ’ the vector length of $B$ .

Therefore, $M=(m||S_{ID}||TS)$ , such that the size of $M$ is $n$ bits.

where, $m$ denotes original message, $S_{ID}$ denotes sensor ID and TS denotes Time stamp.
ii.
Elements in the above formed blocks with the corresponding bit in $B$ equal to 1 are added yielding a number $\beta$ , in the following manner:

$\displaystyle\beta=\sum\limits_{i=1}^{n}B_{i}$ (3)

where $M_{i}=1$ .

Here, $i$ denotes index in vector $B$ and index of bit in $M$ .
iii.
$\beta$ is then transmitted to the hospital server by the sensors.

3.
Decryption phase: At hospital server, the message is computed as follows:

a. a.
$\displaystyle w=Mod(T^{-1},Z)$ (4)
b.
$\displaystyle h=Mod(\betaw,Z)$ (5)
c.
The instance $(A,h)$ is solved by the server in accordance with the Knapsack algorithm to retrieve the message and authenticate the sensor ID along with its database entry. In addition, the timestamp is used to avoid replay of message.

4.
Probability of disease is computed using the linear regression model with sigmoid function as follows: (Notations used are described in the Table 1.)

a. a.
The following notations will be used in subsequent paper:
b.
Compute

$\displaystyle H_{T}(X)=T+T_{1}X_{1}+T_{2}X_{2}+\ldots+T_{n}X_{n}$ (6)
c.
The vector form for above mentioned equation is as follows:

$\displaystyle H_{T}(X)=\begin{bmatrix}T_{0}\\ T_{1}\\ T_{2}\\ \ldots\\ T_{n}\\ \end{bmatrix}\begin{bmatrix}1\\ X_{1}\\ X_{2}\\ \ldots\\ X_{n}\\ \end{bmatrix}$ (7) $\displaystyle H_{T}(X)=\begin{bmatrix}T_{0}&T_{1}&T_{2}&\ldots&T_{n}\\ \end{bmatrix}\begin{bmatrix}1\\ X_{1}\\ X_{2}\\ \ldots\\ X_{n}\\ \end{bmatrix}$

$H_{T}(X)=T^{\prime}X$ , where $T^{\prime}$ is transpose of $T$ .
d.
Now, the desired probability is computed as per the sigmoid function as follows:

$\displaystyle g(Z)=1/{(1+e^{-Z})},$ (8)

where $Z=H_{T}(X)$ .

The above computed value is the probability for any disease, which is used by the respective doctor for disease prediction.

3.2 Scenario 2: When the patient is outside the premises of hospital

Notation	Description
$H_{T}(X)$	Predicted value of output variable
$T_{i}$ (where $i\in(0,1,2,\ldots,n)$	Coefficient, $T_{0}$ called bias and $T_{i}$ (where $i>$ 0) called weight assigned to input variables
$X_{j}$ (where $j\in(1,2,\ldots,n)$	Input variables (sensor information in our paper)
$g(H_{T}(X))$	Sigmoid function over $H_{T}(X)$
$P$	Probability

Similar to any standard sensor network working strategy, in this scenario, sensors implanted on patient’s body will transmit the message to the base station which in turn forward the message to the hospital server via public network and vice versa. Since the message is to be transmitted through public network, we need a secure algorithm along with Knapsack cryptosystem which must be efficient enough while working with devices viz. base station performing a handful number of computation throughout the day. Moreover, message authentication is not provisioned in the Knapsack algorithm, hereby; we also need to incorporate an improved batch verification scheme. As already discussed in the previous section of literature review, the message is embedded within the digital signature has been initially recovered and then verified by receiver.

The proposed batch verification digital signature scheme is an improved version of that proposed by Kang and Tang [8]. The proposed scheme enables multiple messages to be verified using a single step without compromising the security of underlying algorithm. The scheme comprises a number of steps described below to be performed by the base station and the hospital server separately.

Initialization step: Let $P$ be a prime number (large enough) and ‘ $g$ ’ be a primitive element in $Z_{P}$ . An integer ‘ $X$ ’ is chosen from 0 to $(P-1)$ by the hospital server and base station as their respective private key such that GCD $(X,P-1)$ . Afterwards, the base station and hospital server compute their public key as per the following expression:

$\displaystyle Y=Mod((g)^{X},P)$ (9)

where ‘ $Y$ ’ and ‘ $P$ ’ are relatively prime to each other. When the sender, either the base station or the hospital server is required to sign a message $M\in$ GF $(P)$ , the following steps are required to be implemented

Digital signature generation step: The message ‘ $m$ ’ and the timestamp are assorted into the digital signature as three new parameters – ‘ $s$ ’, ‘ $r$ ’, and ‘ $t$ ’, which are computed as follows:

a. a.

$\displaystyle s\equiv Mod(Y^{M},P)$ (10)

$r\equiv[s+M*(g)^{-q}]\bmod P]$ where $q$ is a random number chosen from $[0,P-1]$ .

‘ $t$ ’ is calculated by solving the following expression,

$\displaystyle(s+t)\equiv Mod([(X)^{-1}*(q-r)],(P-1))$ (11)

The triple $(s,r,t)$ is treated as the digital signature for the message. These parameters when encrypted using the Knapsack cryptosystem with public parameter ‘ $B$ ’ of the receiver become the resultant number $\beta$ to be transferred to the receiver.

Message recovery and batch digital signature verification step: Similar to the way as it is in the case when patient is inside the hospital (scenario 1), the triplet $(s,r,t)$ is recovered from $\beta$ by the receiver. Afterwards, contrary to that explained in the Kang’s scheme where each message is recovered and verified exclusively by the receiver, the proposed scheme will recover the messages separately from respective digital signatures and verifies them using single equation as one batch. More illustratively, the messages from the digital signatures $(s_{i},r_{i},t_{i})$ where $0<i<(n+1)$ are recovered as follows:

a. a.

$\displaystyle M^{\prime}_{i}\equiv\left\{\left(r_{i}-s_{i}\right)*g^{r_{i}}*Y^% {\left(s_{i}+t_{i}\right)}\right\}modP$ (12)

Proof given in [8].

Once all the messages are recovered, they are verified using a single equation as follows:

$\displaystyle\prod\nolimits_{i=1}^{n}S_{i}$ (13) $\displaystyle\equiv{Mod(\left(Y\right)}^{\sum\nolimits_{i=1}^{n}{Mod(M}_{i},% \left(P-1\right))},P)$

The above-cited equation establishes the fact that received messages originated from a single source. Moreover, timestamp would be used to avoid replay attack.

Proof According to Fermat’s theorem, Let $P$ be a prime number and a is a number such that GCD $(a,P)\equiv 1$ then:

$\displaystyle a^{p-1}\equiv Mod(1,p)$ (14)

Therefore,

$a^{n*(p-1)}\equiv Mod(1,p)$

Thus,

$a^{(r+n(p-1))}\equiv Mod(a^{r},p)$

For $n$ messages:

$\displaystyle\prod_{i=1}^{n}{Mod(S_{i}},P)$ $\displaystyle\equiv Mod(\{Mod\left(Y^{M_{1}},P\right)*{Mod(Y}^{M_{2}},P)$ $\displaystyle*\ldots*{Mod(Y}^{M_{n}},P)\},P)$ $\displaystyle\equiv Mod(\left\{Y^{M_{1}+M_{2}+\ldots M_{n}}\right\},P) % * By using Eq. () % If the above defined batch verification fails, then to identify maliciously modified message, we are proposing to follow an efficient tree based approach. In this approach the batch of messages, for which batch verification failed, will be divided into two equal halves as shown in the Fig. % . The two halves will be verified independently as a batch. The process will further divide the sub-batches to find the malicious messages. This approach will perform better as comparison to validate individual messages if batch verification failed. % % In average case, the number of sub-batch digital signature verification will be limited to Log% {}_{2}% n as comparison to verifying each message independently which will incur message verification complexity to ‘% n’, where n represents number of messages. Therefore, the proposed scheme will still be efficient in case of few malicious messages. After the successful authentication of messages, logistic regression will be used to compute the probability of disease based upon received information from sensors implanted on patient body. The steps for logistic regression are same as depicted in Scenario-1 step 4.$

References

Al-Janabi

Al-Shourbaji

Shojafar

Shamshirband

. Survey of main challenges (security and privacy) in wireless body area networks for healthcare applications. Egypt. Inf. J. 2017; 18(2): 113-122.

Sangari

Manickam

JML

. Public key cryptosystem based security in wireless body area network. in: International Conference on Circuits, Power and Computing Technologie. 2014; 1609-1612.

Shieh

Lin

Yang

Sun

. Digital multisignature schemes for authenticating delegates in mobile code systems. IEEE Trans. Veh. Technol. 2000; 49: 1464-1473.

Chang

Pon

Lee

. Applying Harn-Kiesler multisignature scheme to electronic document systems. in: Proceedings of National Information Security Conferenc. 1995; 35-38.

Zhang

. Cryptanalysis of Chang et al’s. signature scheme with message recovery. IEEE Commun. Lett. 2005; 9(4): 358-359.

Chien

. Forgery attacks on digital signature schemes without sing one-way hash and message redundancy. IEEE Commun. Lett. 2006; 10(5): 324-325.

Chang

. Signing a digital signature without using one-way hash functions and message redundancy schemes. IEEE Commun. Lett. 2004; 8(8): 485-487.

Kang

Tang

. Digital signature scheme without hash functions and message redundancy. J. Commun. 2006; 27(5): 18-20. (in Chinese)

Wong

Lam

. Digital signatures for flows and multicasts. in: Proceedings of the Sixth International Conference on Network Protocols. (ICN 1998); 198-209.

10.

Zhu

. A comment on “MABS: multicast authentication based on batch signature”. IEEE Trans. Mob. Comput. 2012; 11(11): 1775-1776.

11.

Park

Chong

EKP

Siegel

. Efficient multicast packet authentication using signature amortization. in: Proceedings of IEEE Symposium on Security and Privacy. 2002; 227-240.

12.

Bellare

Garay

Rabin

. Fast batch verification for modular exponentiation and digital signatures. in: EUROCRYPT Springer. Nyberg

. ed. 1998; 1403: 236-250.

13.

Zeadally

Kumar

Lee

. Anonymous authentication for wireless body area networks with provable security. IEEE Syst. J. 2017; 11(4): 2590-2601.

14.

Ibrahim

Kumari

Sangaiah

Gupta

Choo

KKR

. Anonymous mutual authentication and key agreement scheme for wearable sensors in wireless body area networks. Comput. Netw. 2017; 129: 429-443.

15.

Goodman

RMF

McAuley

. New trapdoor-knapsack public-key cryptosystem. IEE Proc. E – Comput. Digit. Tech. 1985; 132(6): 289-292.

16.

Kanchan

Kishor

. Study of machine learning algorithms for special disease prediction using principal of component analysis. International Conference on Global Trends in Signal Processing, Information Computing and Communication, Jalgao. 2016; 5-10.

17.

Chen

Hao

Hwang

Wang

. Disease prediction by machine learning over big data from healthcare communities. IEEE Access. 2017; 5: 8869-8879.

18.

Kumar

Tomar

Chaurasiya

. Enhanced secure transmission of data in wireless body area network for health care applications. 3rd International Conference on Next Generation Computing Technologies (NGCT 2017), Springer. 2017; 828: 138-145.

19.

Tomar

Singh

Sharma

Arya

. Traffic management using logistic regression with fuzzy logic. International Conference on Computational Intelligence and Data Scienc. 2018; 132: 451-460.