A Fuzzy Inference Algorithm for Personal Bio-Informatics Fusion in Home-care System

A Fuzzy Inference Algorithm for Personal Bio-Informatics Fusion in

Home-care System

Chi-Lu Yang, Chin-Yuan Hsu

Networks and Multimedia Institute, Institute for Information Industry

Tainan, Taiwan, R.O.C.

[email protected], [email protected]

Abstract

clearly understand what the other people mean when they communicate. This holds true as well whenever changes occur in their individual health conditions. The same situation is present in the judgment of multiple bio-signals because a similarly fuzzy situation exists. If personal bio-signals are collected, we can judge that a bio-signal overtakes its thresholds. In this research, we applied the fuzzy theory to fuse multiple bio-signals. In addition, we were able to identify the weak patients in a group. In this study, a fuzzy inference algorithm for fusing multiple bio-signals and referring to a person s health was proposed, as a result of which, the weaker persons in the same group were easily identified. Moreover, the inference results were outputted by semantic representations such as high, median, or low. The boundary values of degrees were also defined through special test cases. Furthermore, the degree range was divided into parts to differentiate between the healthy levels of patients.

Keywords: Fuzzy Inference, Bio-signals Fusion,

Health Promotion, Bio-signals Repository

1. Introduction

In the real world, human thought, language representation and feeling are all fuzzy phenomena. Generally, fuzziness is inaccurate, uncertain, and has multiple meanings. Language purpose is mostly judged from people's subjective feelings [1]. Everybody has a different interpretation of one particular sentence. People may not clearly understand what other people mean, more so with the changes that their own health’s undergo. Hence, a fuzzy situation also exists in judging multiple bio-signals. If personal bio-signals are collected, we are not only able to determine that any bio-signal overtakes its threshold limit value; moreover, we are able to fuse bio-signals to be fuzzily analyzed.

In this study, we applied the fuzzy theory to fuse bio-signals and to identify weaker persons in the same group. In Section 2, the fuzzy theory is briefly

discussed. In Section 3, a fuzzy inference system is first introduced. The main functionalities in the system are to collect, transmit, fuse, and analyze bio-signals. A fuzzy inference algorithm is bundled in the system. The algorithm is described in Section 4. Likewise, one practicable example is illustrated. In Section 5, the experimental results are evaluated and discussed. Finally, brief summaries are presented in Section 6.

2. Related Works

Fuzzy sets are sets whose elements have degrees of membership. Fuzzy sets have been introduced by L. A. Zadeh (1965) as an extension of the classical notion of set. Lee et al. proposed a genetic fuzzy agent for meeting scheduler system [2]. Delen and Pratt designed an intelligent decision support system for manufacturing [3]. Wang applied the concept of agent-based control theory in traffic and trans- portation management [4]. Yan et al. proposed a perception-based medical decision support system for the diagnosis of heart diseases [5]. Grossklags and Schmidt studied how software agents affect the consumer’s behaviors for human traders [6]. Gerla [7] proposed fuzzy set based on quantitative analysis of people’s thought. It is similar to the models of people’s thought. Each element in the fuzzy set can be expanded into two-value logic as well as into multi-value logic. The fuzzy degree that assembles the fuzzy set would lie between 0 and 1. The degree value is given by a function, which we named as the membership function. Owing to the difference among every individual’s subjective consciousness, the degree value is also definitely different for each individual.

Generally, the degree value is decided based on the knowledge of the domain experts. There are various formulas of membership function for a user’s defined function. Nonetheless, it is possible to deduce four kinds of membership functions: z-type, lambda-type, pi-type, and s-type [8]. If an inference is judged based on the fuzzy theory, they call it fuzzy inference. By comparing with the traditional logical inference,

fuzzy inference could match adaptive rules by the fuzzy degrees. Even if the inferable degrees do not exactly match the rules, the inference can also approximately apply the most suitable rules.

It is suggested that the best time when personal health should be initially promoted is when people are basically healthy. In medical science, this concept is called preventive medicine. People would like to find methods to maintain and promote their health. Nowadays, information and communication technology (ICT) have greatly advanced. Personal bio-signals can be easily measured and immediately delivered to even remote locations. A remote server could thus analyze bio-signals to quickly judge a person’s physiological status. Lee [9, 10] proposed the use of fuzzy inference architecture to filter the image noises. Their inference rules were optimized through an iterative learning. In this study, we adapted this architecture to fuse multiple bio-signals and to infer personal physiological status.

Fuzzy-inference System bio-signals

collection devices analysis systembio-signals

multiple bio-signals measurable devices Local homebox Local homebox pulse & blood pressure breath thermometer pulse & blood pressure breath thermometer

Figure 1. FIS system architecture

3. Fuzzy Inference System

The fuzzy inference system (FIS) is designed to manipulate multiple bio-signals. Its main functions in the FIS are to collect, transmit, fuse, and analyze bio-signals. Bio-signals include pulse, blood pressure, blood oxygen, blood sugar, breath, and body temperature among others. The applicable domains of FIS are chronic home-care, leisure treatment, and health promotion among others [11]. The various bio-signals represent their own specific features, which need to be artificially determined and interpreted by medical doctors.

However, numerous bio-signals need to be recorded over a significant period of time. Doctors and nurses cannot always stay near the patients to interpret the meaning of their bio-signals. In this situation, a real-time response could not be performed without the information system. We designed the FIS to cover these gaps between the providers and consumers. The FIS could automatically determine and interpret the incoming bio-signals. Thereafter, it could prompt adaptive actions to related persons through its service

processes. The FIS and its algorithms can immediately provide utilities protection and alert patients, hence reducing the manual operational cost.

In the scope of the FIS deployment, if the patient is in a public area, he could be first identified through an RFID tag or ZigBee module. This has been predefined in the database. Thereafter, the patient could measure his own bio-signals by using measurable devices in the public bio-station. The patient could even set up his own configuration after being identified. If the patient is in private area such as his room, he could then directly measure the bio-signals and check up on them by himself. He could naturally set up his own configuration in a private area. After collecting multiple bio-signals from an identified patient, the local homebox is able to converge and fuse the bio-signals. The homebox would then transmit personal data to the FIS for fuzzy-inferred analysis. The transmission is performed in a secure and private protocol. Likewise, the system architecture of collection and analysis bio-signals is shown in Figure 1. If any unusual signal is detected in the analytic results, the system would alert the related persons, such as the patient, family members, or local nurses. In the scope of FIS deployment, the local nurses could also actively monitor and adapt the patients’ activities through the collected and analyzed results.

For fusion and fuzzy-reference of multiple bio-signals, the system requires identification, static personal profile, and dynamic bio-signals. However, local nurses in most environments are not enough. It is quite difficult to manually manipulate these jobs. Therefore, the fuzzy inference system has been designed to solve these issues. In the system infrastructure, homogeneous bio-signals could be handled through international standard, such as Health Level 7. The messages are transmitted in a secure and private mechanism on a service-oriented architecture ([11], [12]). In this infrastructure, the bio-signals are ready to be analyzed. We thus have to focus on designing an analytic algorithm in the system. The details would be described in the next section.

4. Proposed Fuzzy Inference Algorithm

The fuzzy inference algorithm (FIA) in FIS was designed to identify the weaker persons in the same group. Through the identified information, local nurses or caregivers could pay more attention on the patients who were physiologically weaker. In the same algorithm, FIA could also identify persons who were healthier than others. The healthy levels of all patients could also be drawn out. The multiple bio-signals were collected by sensors, which were deployed in the environment. The patients’ bio-signals would first be fused into fuzzy set.

Thereafter, these data would be granularly computed in the fuzzy inference architecture. The inference results would thus be outputted by semantic representations, such as high, median, or low. These representations could be sent to pertinent medical professionals to alert them once it is needed. We define the FIS and FIA in the following section.

Let the FIS system = {U, A}, where U is a set of patients’ bio-signals. Let set U = {P1, P2, P3 … Pn},

where Pi represents the specific features of one

patient’s various bio-signals. Set A presents various bio-signals. If A= {h, w, blood, pulse, spo2,

breathe}, set A thus represents the generic features of height, weight, blood, pulse, spo2, and breath. Let

the group of bio-signals V =_vm_{, where m=1, 2, 3… 6.}

The fuzzy set was defined to correspond with fuzzy semantic presentations, such as [low, median, high]. Let the features of a membership function f(P) = [begin_support, begin_core, end_core, end_support]. The pi-type of the f(P) is represented in Figure 2.

1

begin_support

begin_core end_core end_support degree

Feature

Figure 2. The pi-type of the f(P)

P1


Pn h

^






^ class1 class2 class3

Layer1 (Input linguistic layer) Layer2 (Input term layer) Layer3 (Rule layer) ^ ^



Layer4 (Subrule layer) Layer5 (Output layer)

w blood pulse spo2 breath

Figure 3. Granular computing in fuzzy inference algorithm

The granulation of FIA is designed into layers. Personal bio-signals would be computed by these granulations in each layer. There are five layers defined in the FIA architecture, namely, input

linguistic layer, input term layer, rule layer, sub-rule

layer, and output layer. The architecture of FIA is

shown in Figure 3. The definitions of layers are described as follows:

Layer 1 (Input Linguistic Layer): Input of layer 1

is {P1, P2, P3… Pn}; Output of layer 1 is

ij ij=P

µ , i=1… n, j=1… 6. (1)

Layer 2 (Input Term Layer): Input of layer 2 is

(( P …11 P )… (16 P …n1 P )); Output of layer 2 n6 is 2 ( ) ij jk ijk= f P µ , i=1… n, j=1… 6, k=1, 2, 3. (2)

Layer 3 (Rule Layer): Input of layer 3 is

(( f11(P11)… f13(P11))… (f61(Pn6)…f63(Pn6))); Output of layer 3 is 3 ik µ , where _r∈ℜ3,4 _and, )} ( ),..., ( ), ( min{ 6 3 i 6k i2 2k i1 1k ik= f P f P f P µ , i=1… n, k=1, 2, 3. (3)

Layer 4 (Sub-rule Layer) 

Input of layer 4 is 3 ik µ ; Output of layer 4 is 4 d µ 

d

=

R µ = the value of the kth semantic rule of the 3

ik

µ .

Layer 5 (Output Layer): The degree is computed

by employing all the values from layer 4. Input is

4 ik µ . Output is = = n i ik k n 1 4 5 1 ₍

µ

µ

In the following section, we present pulse and

breath as examples to evaluate whether one patient in

his group is comfortable or not. The breath is depicted in the y-axis in Figure 4, while the pulse is depicted in the x-axis. The ranges of the x-axis and y-axis are separated by the fuzzy semantic presentation ([low, median, high]). Moreover, the whole area is divided into nine sub-areas. One distribution example of pulse and breath is presented in Figure 4.

The rules _ℜ3,4 _{between layers 3 and 4 are}

defined in Table 1. On the other hand, the rules _ℜ4,5

between layers 4 and 5 are defined in Table 2. For example, three compositions of all the inferable rules are listed in Table 2. Their referred results correspond to class1, class2, and class3. For instance, the

inferable architectures of the classes are shown in Figure 5. Class1 covers the union of Rule1, Rule2, and Rule4. Class2 covers the union of Rule2, Rule3, Rule5,

and Rule7. Class3 covers the union of Rule3, Rule6, Rule8, and Rule9.

low

medium

high

lo

w

m

ed

iu

m

hi

gh

pulse

breath

Figure 4. One distribution of breath and pulse Table 1. Rules _ℜ3,4 _{between layers 3 and 4}

pulse breath low high Rule1← ∧ pulse breath median high Rule2← ∧ pulse breath high high Rule3← ∧ pulse breath low median Rule4← ∧ pulse breath median median Rule5← ∧ pulse breath high median Rule6← ∧ pulse breath low low Rule7← ∧ pulse breath median low Rule8← ∧ pulse breath high low Rule9← ∧

Table 2. Rules _{ℜ between layers 4 and 5} 4,5 4 2 1

1 Rule Rule Rule

class ← ∨ ∨

7 5 3

2 Rule Rule Rule

class ← ∨ ∨

9 8 6

3 Rule Rule Rule

class ← ∨ ∨

The feature degrees of a patient’s breath and pulse are shown in the feature table of Figure 6. Subsequently, we could induce rule table based on the feature table and the rules in Table 1. The values in the degree table are computed through the min column in the rule table. Likewise, we could obtain that the degrees of classes (class1, class2, and class3.)

are 0.3, 0.7, and 0.0, respectively. Thus, we can infer that the patient is conformable in his group.

class₁ class₂ class₃

low medium high

lo w m ed iu m hi gh pulse breath

low medium high

lo w m ed iu m hi gh pulse breath

low medium high

lo w m ed iu m hi gh pulse breath Lbreath Mbreath HbreathLpulse Mpulse Hpulse (a) (b) (c) Lbreath Mbreath HbreathLpulse Mpulse Hpulse Lbreath Mbreath HbreathLpulse Mpulse Hpulse

Figure 5. Architectures of the referred classes

0.0 0.0 0.0 0.0 0.3 0.0 0.0 0.7 0.0 0.7 0.7 0.8 0.3 0.3 0.8 0.2 0.7 0.2 0.2 0.3 0.2 0.0 0.0 0.8 0.0 0.0 0.2 min pulse breath [rule table] max max max pulse breath feature 0.0 0.2  high 0.7 0.3 0.8 0.0 P1  median  low data [feature table] 0.0 0.7 0.3 degree green red red classes [degree table] [Decision] P1is conformable .

Figure 6. Referred result example Table 3. Basic values in fuzzy set Input

Terms Low Median High

SBP 80,80,110 95,130,160 150,200,200 DBP 40,40,65 60,75,90 85,100,100 Pulse 50,50,70 60,80,100 90,110,110

5. Experimental Evaluations

In the experimental design, the membership function f(x) could be displayed as a triangle shape (lambda-type), given that we defined begin_core =

end_core. The f(x) was formally defined as follows:

≥ < < − − = = < < − − ≤ = port End_ x port End_ x End_core End_core port End_ x port End_ End_core x Begin_core x Begin_core x port Begin_ port Begin_ Begin_core port Begin_ x port Begin_ x x f sup , 0 sup ), sup /( ) sup ( or , 1 sup ), sup /( ) sup ( sup , 0 ) (

A manometer with three input terms was chosen as a bio-signal device. It could measure systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse. The initial values in the fuzzy set are presented in Table 3. In addition, they corresponded with the semantic presentation.

In this experiment, one semantic presentation went along with one fuzzy linguistic term hedge (FLTH) of the fuzzy set. The const parameter was defined to modify the range of the fuzzy set. Their relationships are listed in Table 6. We set three fuzzy hedges by changing the value of . Furthermore, three hedges with their semantic presentations are listed in Table 4.

Table 4. setting of fuzzy hedges fuzzy

hedges Low( ) Median( ) High( )

Hedge 1 1 1 1

Hedge 2 2 0.5 2

Hedge 3 0.5 2 0.5

In Figure 7, we were to deduce that fuzzy hedges were affected by the value of . The inferred results were depended on the hedge settings. By comparing the hedges in Figure 7, hedge 2 had more significant differences than others. Specifically, the inferred results of hedge 2 would be more precise than the others.      data number de gr ee Hedge1 Hedge2 Hedge3

Figure 7. Hedge comparisons Table 5. Special test case

Input

[SBP, DBP, Pulse] Output degree [80,40,30] 0.67 [200,100,110] 0.0

[130,75,65] 4.5

The boundary values of degrees were also defined by special test cases. This is shown in Table 5. The worst degree is 0, while the best degree is 4.5. If the degree range is equally divided into three parts, the range of the healthy persons is from 3.0 to 4.5. In contrast, the range of the weak persons is from 1.5 to

2.9, while the range of the critical persons is from 0 to 1.4. However, a large amount of data samples is required to accurately define the boundaries before the algorithm is used in specific domains.

6. Conclusions and Future Works

In the real world, people may not always clearly understand each other’s thought. Likewise, they also find it difficult to understand the changes in each other’s health conditions. We could judge one bio-signal whenever overshoots its thresholds, that is, generic or personal. In this study, a fuzzy inference algorithm for fusing multiple bio-signals and determining a person’s health was proposed. We applied the fuzzy theory to fuse multiple bio-signals and to identify weaker persons in a group. In the experimental results, the weaker persons could be successfully identified by semantic representations. Likewise, the boundary values of degrees were defined. The degree range could be divided into parts to differentiate the patients’ health levels and could be used in specific domains.

In the future, we will apply FIA in specific domains, such as computer officers and leisure health promoters. The initial values of fuzzy set have to be defined by the specific domain experts. The best value of fuzzy hedge will be adapted if FIA is used in different users’ domains.

Acknowledgments

This research was supported by the Applied Information Services Development and Integration project of the Institute for Information Industry (III) and sponsored by MOEA, Taiwan R.O.C.

References

[1] Novak, V. “Are fuzzy sets a reasonable tool for modeling vague phenomena?” Fuzzy Sets and Systems, vol. 156, no. 3, pp. 341-348, Dec. 2005.

[2] C. S. Lee, C. C. Jiang, and T. C. Hsieh, “A Genetic Fuzzy Agent Using Ontology Model for Meeting Scheduling System,” Information Sciences, Vol. 176, No. 9, pp. 1131-1155, 2005.

[3] D. Delen and B. Pratt, “An Integrated and Intelligent DSS for Manufacturing System,” Expert Systems with Applications, Vol. 30, No. 2, pp. 325-336, 2006. [4] F. Y. Wang, “Agent-based Control for Networked

Traffic Management Systems,” IEEE Intelligent System, Vol. 20, No. 5, pp. 92-96, 2005.

[5] H Yan, Y. Jiang, J. Zheng, C. Peng, and Q. Li, ”A multiplayer Perceptron-based Medical Decision Support System for Heart Disease Diagnosis,” Expert Systems with Applications, Vol. 30, No. 3, pp. 272-281, 2006. [6] J. Grossklags and C. Schmidt, “Software Agents and

IEEE Transactions on Systems, Man, and Cybernetics-PartC: Applications and Reviews, Vol. 36, No. 1, pp. 1-13, 2006.

[7] G. Gerla, “Effectiveness and Multivalued Logics,” Journal of Symbolic Logic, vol. 71, pp. 137–162, 2006. [8] V. Novak, "On fuzzy type theory," Fuzzy Sets and

Systems, vol. 149, no. 2, pp. 235-273, Jan, 2005. [9] C. S. Lee, S. M. Guo, and C. Y. Hsu, “Genetic- Based

Fuzzy Image Filter and Its Application to Image Processing,” IEEE Trans. on Systems, Man and Cybernetics Part B, vol. 35, no. 4, pp. 694-711, Aug. 2005.

[10] S. M. Guo, C. S. Lee, and C. Y. Hsu, “An Intelligent Image Agent Based on Soft-Computing Techniques for

Color Image Processing,” Expert Systems with Applications, vol. 28, no. 3, pp. 483-494, 2005. [11] Chi-Lu Yang, Yeim-Kuan Chang and Chih-Ping Chu, ”Modeling Services to Construct Service-Oriented Healthcare Architecture for Digital Home-Care Business,” The 20th International Conference on Software Engineering and Knowledge Engineering (SEKE'08), San Francisco, USA, July, 2008.

[12] Chi-Lu Yang, Yeim-Kuan Chang and Chih-Ping Chu, ”A Gateway Design for Message Passing on SOA Healthcare Platform,” The Fourth IEEE International Symposium on Service-Oriented System Engineering (SOSE 2008), Jhongli, Taiwan, Dec. 2008.

Table 6. Relationship between FLTH and

FLTH( =2) : Very FLTH( =1) : Normal FLTH( =0.5) : More or Less

SBP degree 1 95 160 2 , ] [

_µ

_δ

_µ

_δ

_µ

_δ