A Belief Rule Based Clinical Decision Support System to Assess Suspicion of Heart Failure from Signs, Symptoms and Risk Factors

Transcription

1 A Belief Rule Based Clinical Decision Support System to Assess Suspicion of Heart Failure from Signs, Symptoms and Ris Factors Saifur Rahaman Computer Science and Engineering University of Chittagong Abstract A Clinical Decision Support System (CDSS) to assess suspicion of a disease would avoid unnecessary cost of medical diagnosis. Heart Failure (HF) is a complex clinical syndrome of cardiac disorder. In the present paper, a Belief rule based (BRB) CDSS has been proposed to assess suspicion of HF by using signs, symptoms and ris factors. The recently developed generic rulebased inference methodology using the evidential reasoning approach (RIMER) was considered as the methodology for developing this CDSS. etbean7.2 s GUI and MySQ server have been employed to develop the system. This belief rule based CDSS can deal with various types of uncertainties found in clinical sings, symptoms, ris factors and domain nowledge. The nowledge based for this system has been developed by taing account of real patient data, obtained in consultation with specialists. The CDSS has been tested by using the data taen from the patients with breathlessness. It has been observed that the results generated by the system is reliable and thus facilitates to tae decision to avoid unnecessary costly medical diagnosis. Keywords- Belief RuleBase; CDSS; uncertainity; RIMER; heart failure; Suspicion; signs, symptoms and ris factors I. ITRODUCTIO Cardiovascular disease mortality is more common among the population of South Asia, Europe and America[1]. Heart Failure (HF) is a syndrome of cardiovascular disease. ow-adays, many people of Bangladesh are suffering from heart failure. The rate of HF usually increases with the increase of age [2]. The diagnosis of HF is very expensive. In Bangladesh, it is generally observed that patients are commonly advised by the physicians to carry out costly laboratory tests without having preliminary suspicion of HF. This suspicion process of HF, consists of observation of a physician by using signs, symptoms and ris factors of a patient. However, this suspicion process contains error since the measurement of signs, symptoms and ris factors of a patient can t be achieved in a quantitative way. Consequently, the measurement of HF suspicion contains various types of uncertainty such as vagueness, imprecision, incompleteness, ambiguity, ignorance and randomness.. Therefore, by employing a methodology, which can handle the mentioned uncertainties, the development of a CDSS to assess suspicion of heart failure is necessary. Cost-effectiveness of medical diagnosis and treatment is of growing importance [3] [4], basically for the patients of Bangladesh. The proposed CDSS would facilitate patient to avoid costly lab investigations. In this paper Section II Mohammad Shahadat Hossain Computer Science and Engineering University of Chittagong Hossain_ms@hotmail.com presents an overview of HF, signs, symptoms and ris factors used to detect its suspicion along with a HF suspicion process. iterature review is elaborated in Section III.. Section IV introduces RIMER methodology employed to develop the CDSS. The system architecture, design and implementation of the CDSS are discussed in Section V. Section VI presents the results and discussions, while VII is the conclusion. II. HEART FAIURE Heart Failure (HF) occurs from any structural or functional cardiac disorder[2] that impairs the ability of the heart to function as a pump to support a physiological circulation. Patients having shortest of breath has been considered as the clinical target for the research presented in this paper.. HF is the leading cause of death in western world [2]. It frequently occurs due to coronary artery disease, tends to affect elderly people and often leads to prolonged disability. However, in early stage, HF detection is necessary. The evaluation of patients with suspected heart failure entails determining more than just whether the syndrome is present or not. Generally, signs, symptoms and ris factors of HF used in this suspicion process. The main signs, symptoms and ris factors are presented in next sub section. A. Main Signs, Symptoms and Ris Factor of Heart Failure Patients with heart failure may have a number of signs, symptoms and ris factors. One of the primary symptoms of heart failure is breathlessness. In this paper, we considered four symptoms, three signs and four ris factors. The most common symptoms are- breathlessness, fatigue, exercise intolerance, and fluid retention [2][6][21]. The common clinical signs arejugular venous pressure (JVP), tachycardia, and third heart sound[2][6][21]. The most common ris factors of heart failure are hypertension, smoing, diabetes and myocardial infarction (MI)[6]. B. Heart Failure Suspicion Process The diagnosis of heart failure consists of two stages, firstly, suspicion determination and secondly, lab investigation [2]. Figure 1 illustrates the algorithm of heart failure diagnosis. This paper deals with only suspicion stage of this algorithm, which uses signs, symptoms and ris factors. The result of this stage will in turn be used by the cardiologist to carry out lab investigation /13/$ IEEE

2 Figure 1. Full diagnosis algorithm of heart failure [ICE Guideline, 2010] III. REVIEW OF THE EXISTIG WORKS Clinical Decision Support Systems (CDSS) have become paramount in the medical environment. These systems enable the reduction of the intrinsic degree of uncertainty that any decision-maing process in the medical environment entails. In the literature, different methodologies have been used to develop CDSS based expert system with uncertainties. The first well-nown medical rule based expert system developed by Shortliffe [7][8] used to diagnosis bacterial infections. The rule based expert system, nown as PERFEX [9] used to support to solve clinicians problems. ITERIST-I [10] was developed to diagnosis complex problems in general in the internal medicine. However, none of the mentioned rule-based expert systems can handle various types of uncertainties as mentioned in Section I. There are some CDSS expert systems such as AAPhelp-eeds abdominal pain diagnosis system[11]. Environment and clinical wor based CDDSs for Emergency Department, Intensive Care Unit and laboratory [12][13][14]. Some non-nowledge-based CDSSs which learn domain nowledge through large historical data [24]. Since uncertainties in both clinical domain nowledge and clinical data are inevitable, these CDDSs have not addressed this issue. However, Bayesian theory with a raned list and posterior probability to each diagnosis was used in Iliad [15] and Fuzzy logic used in renal transplantation assignment [16]. The Bayesian theory and Fuzzy logic have their inherent limitations in representing all types of uncertainties. Therefore, the representation of Heart Failure (HF) suspicion with uncertain medical nowledge is crucial, requiring new methodologies and techniques [17]. Hence, this paper presents the employment of recently developed belief rule-base inference methodology using the evidential reasoning approach (RIMER)[18],which can handle various types of uncertainties in developing the CDSS to be used to assess HF suspicion. IV. OVERVIEW OF RIMER METHODOOGY The RIMER stands for Belief Rule-based Inference Methodology using the Evidential Reasoning approach [18]. This methodology consists of three parts. Firstly, Belief Rule Base (BRB) is used to represent domain nowledge with uncertain information. Secondly, the Evidential Reasoning (ER) [19] algorithm is used to deduce inference. BRB can capture complicated nonlinear causal relationships between antecedent attributes and consequents, which is not possible in traditional IF-THE rules [18]. This section introduces RIMER methodology. A. Domain Knowledge Re-orientation using BRB Belief Rules are the ey constituents of a BRB, which include belief degree and are the extended form of traditional IF-THE rules. In a belief rule, each antecedent attribute taes referential values and each possible consequent is associated with belief degrees [18]. The nowledge representation parameters are rule weights, antecedent attribute weights and belief degrees in consequents, which are not available in traditional IF-THE rules. A belief rule can be defined in the following way. R IF (P 1 is A 1 ) (P 2 is A 2 ) P T is A T THE C 1, β 1, C 2, β 2,, C, β R β j 0, β j 1 j =1 (1) with a rule weight θ, attribute weights δ 1, δ 2, δ 3,, δ T 1,, Where P 1, P 2, P 3 P T represent the antecedent attributes in the th rule. A i i = 1,, T, = 1,, represents one of the referential values of the ith antecedent attribute P i in the th rule. C j is one of the consequent reference values of the belief rule. β j j = 1,,, = 1,, is the degree of belief to which the consequent reference value C j is believed to be true. If j=1 β j = 1 the th rule said to be complete; otherwise, it is incomplete. T is the total number of antecedent attributes used in th rule. is the number of all belief rules in the rule base. is the number of all possible consequent in the rule base. An example of belief rule in the domain of heart failure suspicion can be written in the following way. R : IF HF Symptoms is Hig AD HF Signs is Medium AD HF Ris facfor is Hig THE Heart Failure is Hig, 0.60, Medium, 0.40, ow, 0.00 (2) Where {(High, 0.60), (Medium, 0.40), (ow, 0.00)} is a belief distribution associated with Heart Failure consequent of the belief rule as represented in equation (2). The belief distribution states that the degree of belief associated with High Heart Failure is 60%, 40% degree of belief associated with Medium Heart Failure while 0% degree of belief is associated with ow Heart Failure. Here, High, Medium and ow can be considered as the referential values of the consequent attribute Heart Failure of the belief rule. In this belief rule, the total degree of belief is ( )=1 and hence, the assessment is complete. B. BRB Inference System The ER approach [16][19][22] developed to handle multiple attribute decision analysis (MADA) problem having

3 both qualitative and quantitative attributes under uncertainty. Different from traditional MADA approaches, ER presents MADA problem by using a decision matrix, or a belief expression matrix, in which each attribute of an alternative described by a distribution assessment using a belief structure. The inference procedures in BRB inference system consists of various components such as input transformation, rule activation weight calculation, rule update mechanism, followed by the aggregation of the rules of a BRB by using ER. The input transformation of a value of an antecedent attribute P i consists of distributing the value into belief degrees of different referential values of that antecedent. This is equivalent to transforming an input into a distribution on referential values of an antecedent attribute by using their corresponding belief degrees [20]. The ith value of an antecedent attribute P i i = 1,, T at instant point in time can equivalently be transformed into a distribution over the referential values, defined for the attribute by using their belief degrees [18]. The input value of P i, which is the ith antecedent attribute of a rule, along with its belief degree ε i is shown below by equation (3). The belief degree ε i to the input value is assigned by the expert in this research. H P i, ε i = A ij, α ij, j = 1,, j i, i = 1,, T (3) Here H is used to show the assessment of the belief degree assigned to the input value of the antecedent attribute. In the above equation A ij (ith value) is the jth referential value of the input P i.. α ij is the belief degree to the referential value j i A ij with α ij 0. j =1 α ij 1 i = 1,, T, and j i is the number of the referential values In this paper, the input value of an antecedent attribute is collected from the patient or from the physician in terms of linguistic values such as Severe, Moderate and Mild. This linguistic value is then assigned degree of belief ε i by taing account of expert judgment. This assigned degree of belief is then distributed in terms of belief degree α ij of the different referential values A ij [High (H), Medium (M), ow, ()] of the antecedent attribute. The above procedure of input transformation is elaborated by equations given below. if H vlue input vlue(ε i ) M vlue Tan, Medium = H vlue input vlue H vlue M vlue, Hig = 1 Medium, ow = 0.00 (4) if M vlue > input vlue(ε i ) vlue Tan, ow = M vlue input vlue M vlue vlue, Medium = 1 ow, Hig = 0.00 (5) It is assumed that α i is the belief degree of the one referential values A i (which is the element of A ij ) of the ith input P i in the th rule. Here it is called individual matching degree and there is α i α ij, i = 1,, T, j = 1,, j i, α ij can be calculated by using (4) and (5). When the th rule is activated, the weight of activation of the th rule, w is calculated by using the flowing formula [18]. ω = T θ α = θ α i δ i i=1 j=1 θ j α j T j θ j α δ jl j=1 l=1 l and δ i = δ i max i=1,,t δ i (6) Where δ i is the relative weight of P i used in the th rule, which is calculated by dividing weight of P i with maximum weight of all the antecedent attributes of the th rule. By doing so, the value of δ i becomes normalize, meaning that the range of its value should be between 0 and 1. α = T α δ i i=1 i is the combined matching degree, which is calculated by using multiplicative aggregation function. When the th rule as given in.(1) is activated, the incompleteness of the consequent of a rule can also result from its antecedents due to lac of data. An incomplete input for an attribute will lead to an incomplete output in each of the rules in which the attribute is used. The original belief degree β i in the ith consequent C i of the th rule is updated based on the actual input information as[18]. β i = β i T t=1 τ t, Jt j=1 α tj T t=1 τ t, Where, t, = 1, if P i is used in defining R (t = 1,, T ) 0, oterwise Here β i is the original belief degree and β i is the updated belief degree. ER approach is used to aggregate all the pacet antecedents of the rules to obtain the degree of belief of each referential values of the consequent attribute by taing account of given input values P i of antecedent attributes. This aggregation can be carried out either using recursive or analytical approach. In this research analytical approach has been considered since it is computationally efficient than recursive approach [16][19]. Using the analytical ER algorithm [22], the conclusion O(Y), consisting of referential values of the consequent attribute, is generated. Equation (8) as given below illustrates the above phenomenon. : O(Y) = S( P i ) = C j, β j, j = 1,, (8) Where β j denotes the belief degree associated with one of the consequent reference values such as C j. The β j is calculating by analytical format of the ER algorithm [22] as illustrated in equation (9). β j = μ =1 ω β j +1 ω β j With 1 μ μ = ω β j + 1 ω β j j=1 =1 j=1 j=1 =1 1 ω j=1 β j =1 1 ω n 1 1 ω β j =1 j=1 (7) (9) The final combined result or output generated by ER is represented by C 1, β 1, C 2, β 1, C 3, β 1,, C, β, Where β j is the final belief degree attached to the jth referential value C j of the consequent attribute, obtained after combining all activated rules in the BRB by using ER. C. Output of the BRB System The output of the BRB system is not crisp/numerical value. Hence, this output can be converted into crisp/numerical value by assigning utility score to each referential value of the consequent attribute [16][18]- H(A ) = j =1 u C j B j (10) Where H(A ) is the expected score expressed as numerical value and u C j is the utility score of each referential value 1

4 V. BRB CDSS FOR HEART FAIUR SUSPICIO Heart Failure (HF) is diagnosis in two stages namely determination of suspicion and lab investigation as mentioned previously. The design and implementation of CDSS for Heart Failure suspicion is presented in this section. A. System architecture, Design and Implementation Architectural design represents the structure of data and program components that are required to build a computerbased system. It also considers the pattern of the system organization, nown as architectural style. In this paper the CDSS for HF suspicion adopts the three-layer architecture[23], which consist of patient information, interface layer, BRB application possessing application layer (RIMER Methodology with its components), data management layer database, and maing decision as output. Figure 2. System architecture of the BRB CDSS for HF suspicion Figure 2 illustrates the system architecture of the CDSS for HF suspicion, which includes the components as discussed above It is important to note that in the CDSS as illustrated in Figure 2, MySQ, which is a relational database has been used in bac-end to store and manipulate initial BRB, which represents nowledge based of the system. MySQ as bacend relational database has been chosen since it is flexible and portable in any system environment. Java application (J2SE), which is available in etbean 7.2 has been used to develop the interface and the components of the application layer. The reason for selecting is that Java application is platform independent, flexible and robust B. System components - System Input with Clarification The input information that use to perform the heart failure suspicion are classified in three categories namely symptoms, signs and ris factors. In this paper, we used four symptoms, three signs and four ris factors as input parameter of our proposed system. The input clarification of input antecedent Fatigue (A1), Breathlessness(A2), Exercise intolerance(a3), Fluid retention(a4), jugular venous pressure(jvp)(a5), Tachycardia (A6), Third heart sound(a7), MI(A8), Hypertension(A9), Smoing(A10) and Diabetes(A11) are transformed to referential value by equation (5),(6) on behalf of expert cardiologist. The input clarifications of this BRB system transformed to referential is shown in table I. TABE I. Sl. o. THE IPUTS ARE TRASFORMED I REFERETIA VAUE Input Antecedent Input Referential Value H M 1 Fatigue Yes Fatigue o Breathlessness o limitation Breathlessness Mild Breathlessness Moderate Breathlessness Severe Exercise intolerance Yes Exercise intolerance o Fluid retention Yes Fluid retention o JVP Range JVP Un range Tachycardia Yes Tachycardia o third heart sound Yes third heart sound o MI Yes MI o Hypertension Yes Hypertension o Smoing Yes Smoing o Diabetes Yes Diabetes o C. System components -Knowledge base constructed using BRB The heart failure diagnosis algorithm was investigated, which have two stages including suspicion and final lab investigations. In present paper, we wored on heart failure suspicion. In order to construct BRB nowledge base of this system we designed a BRB framewor to heart failure suspicion according to ICE guideline and clinical domain expert cardiologist. The BRB framewor of heart failure suspicion as illustrated in Figure 3, From the framewor, it can be observed that input factors that determine suspicion include Fatigue(A1), Breathlessness(A2), Exercise intolerance(a3), Fluid retention(a4), JVP(A5), Tachycardia (A6), 3 rd heart sound(a7), MI(A8), Hypotension(A9), Smoing(A10) and Diabetes(A11). This BRB nowledge base has different traditional rule to suspicion, which need to convert belief rules. Figure 3. BRB Framewor to suspect the Heart Failure (Showing All SRB) In such situations, belief rules may provide an alternative solution to accommodate different types and degrees of uncertainty in representing both clinical data and clinical

5 domain nowledge. A BRB can be established in the following four ways[25]- (1) Extracting belief rules from expert nowledge (2) Extracting belief rules by examining historical data; (3) Using the previous rule bases if available, and (4) Random rules without any pre-nowledge. In this paper, we constructed initial BRB by the domain expert nowledge. This BRB consists of four sub-rule-bases namely HF symptoms (A12), HF signs (A13), HF ris factors (A14) and Heart Failure (A15). HF symptoms sub-rule-base has four antecedent attributes. Each antecedent attribute consists of three referential values. Hence, this sub-rule-base consists of 81 rules. The entire BRB (which consists of four sub-rule bases) consists of ( ) = 216 belief rules. It is assumed that all belief rules have equal rule weight; all antecedent equal weight, and the initial belief degree assigned to each possible consequent by two cardiologists from accumulated patient data. To better handle uncertainties, each belief rule considered the three referential values are High (H), Medium (M) and ow (). An example of belief rule of this BRB system as follow- R1: IF Fatigue is Yes AD Breathlessness is Severe AD Exercise intolerance is Yes AD Fluid retention is Yes THE HF symptoms is{h(1.00), M(0.00), (0.0)} In the above belief rule, the belief degrees attached to the three referential values. For example, if we have 100 patients whose clinical information fully satisfies the antecedents of the rule then, there are all patients to be judged at High to suspect Heart failure from HF symptoms, and then we can assign the initial belief degrees to the consequents in the rule as shown above. D. System components - Inference Engine Design using ER This BRB CDSS designed using the ER approach [18][22], which is described in section IV.B. It is similar to traditional forward chaining. The inference with a BRB using the ER approach also involves assigning values to attributes, evaluating conditions and checing to see if all of the conditions in a rule are satisfied. The BRB inference process using the ER approach described by the following steps are input transformation, calculation of the activation weight, calculating combined belief degrees to all consequents, belief degree update and aggregate multiple activated belief rules. The inputs of clinical data are of two types, numerical and subjective. Input transformation of this system and input clarification are deduced in previous section and table I by using (4)(5). After the value assignment for antecedent symptoms, signs, and ris factors calculating the combined matching degrees between the inputs and the rule s antecedents, the next step is to calculate activation weight for each pacet antecedent in the rule base using (6). The belief degrees in the possible consequent of the activated rules in the rule base are updated using (7). Then aggregating all activated rules using the ER approach to generate a combined belief degree in possible consequents using (8)(9). Then expected suspicion of heart failure was calculated from its different suspecting consequents of HF signs, HF symptoms, and HF ris factors. Finally, presenting the system inference results of Heart Failure (A15) consequent which is not crisp/numerical value, then it is converted into crisp/numerical value for recommendation using(10). VI. RESUT AD DISCUSSIO In this present paper, the patients having shortness of breath or previous MI history was chosen as a target for providing this suspicion BRB CDSS of Heart failure. If a patient has shortness of breath or previous MI history then a clinician will enquire about the patient s signs, symptoms and ris factors of heart failure. This patient information was used in this system to determine whether the patient is suspect or not suspect of heart Failure. The clinically simulated data set variables used in this paper to determine heart failure suspicion are one patient s clinical signs, symptoms, ris factors and outcome include Fatigue, Breathlessness, Exercise intolerance, Fluid retention, JVP, Tachycardia, third heart sound, MI, Hypertension (HT), smoing, Diabetes and Outcome. Where the outcome is dependent variable, it is used to present outcome as having heart failure suspected or not, if Heart failure suspicion result is greater than 50% (High ris), then suspicion outcome is considered one otherwise zero. The real data set of 100 patients were collected and simulated having suspicion of Heart failure with different clinical level. The five patient s simulated data set with suspicion outcome is presented as example in table II. This table represents overall heart failure outcome from patient s information. The suspicion result of this system is measured in percentage for recommendation. The output of this system was generated based on output utility equation (10). In this paper, the utility score of 100% assigned to High, 50% assigned to Medium, and 0% assigned to ow. For example, we can estimate overall system output heart failure suspicion as 99.69% (crisp/numerical value), if the suspicion Fuzzy result of the system is {(High, 0.94), (Medium, 0.04), (ow, 0)}. In the case study, the heart failure suspicion of 50 patients using this system (blue curve), cardiologist s manual system (red curve) and clinical history result (green curve) is shown in figure 4. The clinical historical results were considered as benchmar. From figure 4 it can be observed that CDSS generated result has less deviation than from benchmar result. Hence, it can be argued that CDSS output is more reliable than manual system. Therefore, it can be concluded TABE II. EXAMPE PATIET SIMUATED DATASET FROM SIGS AD SYMPTOMS AD RISK FACTORS (FIE-TUIG SUSPICIO OUTCOME BY DOMAI EXPERT) ID Fatigue Breathlessness Exercise Fluid JVP Tachycardia 3 rd Heart Failure Heart Sound MI HT Smoing Diabetes intolerance retention Suspicion P1 Yes Moderate Yes Yes Un range Yes Yes o Yes Yes o 1 P2 o o limitation Yes o Range o o o o Yes Yes 0 P3 Yes Mild Yes Yes Un range Yes Yes o Yes o o 1 P4 Yes Severe Yes Yes Range Yes Yes Yes Yes Yes Yes 1 P5 o o limitation o o Range o o Yes o Yes o 0

6 that if the assessment of suspicion of heart failure is carried out by using the CDSS, eventually this will play an important role in taing decision to avoid unnecessary costly lab investigation. Figure 4. Heart failure suspicion graph of BRB system output, Expert manual system and Benchmar of Historical data. VII. COCUSIO The development and application of a belief rule based CDSS to assess suspicion of heart failure by using signs, symptoms, and ris factors of patients have been presented. The prototype CDSS is embedded with a novel methodology nown as RIMER, allows the handling of various types of uncertainty and hence, be considered as a robust tool can be utilized in assessing suspicion of heart failure. Consequently, the prototype CDSS can handle various types of uncertainties found in clinical domain nowledge as well as in signs, symptoms and ris factors of a patient. Most importantly, the system will play an important role in reducing the cost of lab investigations. The system will facilitate patients in taing precautionary measures well in advance. The system can provide a percentage of suspicion recommendation, which is more reliable and informative than from the traditional cardiologist s opinion. The prototype CDSS can only be used to assess the suspicion of heart failure by using signs, symptoms and ris factors of a patient but cannot be used to diagnose the heart disease. In future, an attempt will be taen to enhance the system with the capability to support the diagnosis of heart disease ACKOWEDGMET The authors are grateful to Dr. Muhammad Abdullah Al Anis, MBBS (CMC), FCPS, woring as the medical officer of ational Hospital (Pvt.) td, Chittagong, Bangladesh and Dr. Saiful Islam Tipu Chow. MBBS, FCPS (Medicine part-1), Studying MD in Cardiology in BSMR Medical University, Dhaa, for providing domain nowledge. REFERECES [1] arendra Singh, MD; Milan Gupta, MD, clinical characteristics ofsouthasian patients hospitalized withheart failure, Ethnicity & Disease, Volume 15, Autumn, 2005, p , [2] Steven Barnes, Professor Martin Cowie, Jonathan Mant, Jennifer Roberts, et al., CHROIC HEART FAIURE ational clinical guideline for diagnosis and management in primary and secondary car, The ational Collaborating Centre for Chronic Conditions, ROYA COEGE OF PHYSICIAS, SB X, 2003 [ICE Gd.] [3] Mar DB, Hlaty MA, Califf RM, et al. Cost effectiveness of thrombolytic therapy with tissue plasminogen activator as compared with streptoinase for acute myocardial infarction. Engl J Med; 1995, 332: ,. [4] Weintraub WS, Mauldin PD, Becer E, Kosinsi AS, King SB, III. A comparison of the costs of and quality of life after coronary angioplasty or coronary surgery for multivessel coronary artery disease. Results from the Emory Angioplasty Versus Surgery Trial (EAST). Circulation; 1995, 92: , [5] D.. Xu, J. iu, J.B. Yang, G.P. iu, J. Wang, I. Jeninson, J. Ren, Inference and learning methodology of belief-rule-based expert system for pipeline lea detection, Expert Systems with Applications 32 (2007) [6] ici R. Colledge, Brian R. Waler, Stuart H. Ralstonl, Davidson s Principles and Practice of Medicine 21st Edition, ISB-13: , 2010 [7] Shortliffe, E. H. Computer-Based Medical Consultations: MYCI, ew Yor, Elsevier, 1976 [8] Buchanan and Shortliffe, Stanford University, MYCI expert system, 1984 [9] Ezquerra et al., Emory University Hospital, 1992 [10] Kumar, K. A., Singh, Y. & Sanyal, S. Hybrid approach using case-based reasoning and rule-based reasoning for domain independent clinical decision support in ICU. Expert Systems with Applications, 2009, 36,65-71 [11] De Dombal, F. T., eaper, D. J., Staniland, J.R., Mccann, A. P. & Horrocs, J. C. Computer-aided diagnosis of acute abdominal pain. British Medical Journal, 1972, 2(5804) 5-9, [12] Graham, T., Bullard, M., Kushniru, A., Holroyd, B. & Rowe, B. Assessing the sensibility of two clinical decision support systems. Journal of Medical Systems, 2008, 32, , [13] Mac, E. H., Wheeler, D. S. & Embi, P. J. Clinical decision support systems in the pediatric intensive care unit. Pediatric Critical Care Medicine, 2009, 10,23-28, [14] Grams, R. R. Clinical laboratory test reference (CTR). Journal of Medical Systems, 1993, 17,59-67, [15] Warner, H.R.J.,. Iliad: moving medical decision-maing into new frontiers. Methods Inform. Med. 1989, 28, [16] Yuan, Y., Feldhamer, S., Gafni, A., Fyfe, F., udwin, D.,. The development and evaluation of a fuzzy logic expert system for renal transplantation assignment: is this a useful tool? Euro. J. Oper. Res, , [17] Musen, M., Shahar, Y. Shortliffe, E.,Clinical Decision-Support Systems. In: Shortliffe, E.H. Cimino, J.J. (Eds.) Biomedical Informatics, 2006 [18] Yang, J. B., iu, J., Wang, J., Sii, H. S. & Wang, H. W. () Belief rulebase inference methodology using the evidential reasoning approach - RIMER. IEEE Transactions on Systems Man and Cybernetics Part A- Systems and Humans, 2006, 36, [19] Yang, J. B. & Sen, P. (1994) A general multi-level evaluation process for hybrid MADM with uncertainty. Systems, Man and Cybernetics, IEEE Transactions on,24, [20] J. B. Yang, Rule and utility based evidential reasoning approach for multi-attribute decision analysis under uncertainties, Eur. J. Oper. Res., vol. 131, no. 1, pp , [21] Cowie MR, Wood DA, Coats AJ, Thompson SG, Suresh V, Poole- Wilson PAet al. Survival of patients with a new diagnosis of heart failure: a population based study. Heart,; 2000, 83: , [22] Y.M. Wang, J.B. Yang, D.. Xu, Environmental impact assessment using the evidential reasoning approach, European Journal of Operational Research 174 (2006) [23] Roger S. Pressman, Software engineering: a practitioner s approach. 5th ed. p. cm. (McGraw-Hill series in computer science) Includes index. ISB , p [24] Berner, E. S. & a ande, T. J., Overview of Clinical Decision Support Systems. I BERER, E. S. (Ed.) Clinical Decision Support Systems: Theory and Practice. 2nd ed. ew Yor, Springer-Verlag, 2007 [25] D.. Xu, J. iu, J.B. Yang, G.P. iu, J. Wang, I. Jeninson, J. Ren, Inference and learning methodology of belief-rule-based expert system for pipeline lea detection, Expert Systems with Applications 32 (2007)