A fuzzy expert system based on positive rules for depression diagnoisis

Nghiên cứu khoa học công nghệ Tạp chí Nghiên cứu KH&CN quân sự, Số Đặc san CNTT, 12 - 2020 33 A FUZZY EXPERT SYSTEM BASED ON POSITIVE RULES FOR DEPRESSION DIAGNOISIS Mai Thi Nu1*, Nguyen Hoang Phuong2 Abstract: Fuzzy set theory and fuzzy logic are highly suitable mathematical tools for developing intelligent systems in medicine. This paper presents a fuzzy expert system based on positive rules for diagnosing depression types. A knowledge base that includes more than 800 positive rules to determine diagnostic conclusions for 04 types of depression. The expert system has been tested on more than 200 medical records of depressed patients. Test results show the suitable accuracy of the system in diagnosis. Keywords: Fuzzy Expert Systems; Positive Rules; Diagnosis of Depression Types. 1. INTRODUCTION According to the World Health Organization, "depression" is a common mental disorder characterized by sadness, loss of interest, feelings of guilt or low self-esteem, sleep disorders, and eating drinking disorders, feeling tired and poor concentration. It affects approximately 264 million people worldwide [13]. Depressive disorder can manifest itself in several types, such as mild depressive disorder, moderately depressive disorder, major depressive disorder, and psychotic major depressive disorder. Depression disorder is the fourth leading cause of death worldwide and is predicted to be the second leading cause of death in 2030 [13]. In Vietnam, according to statistics of the Ministry of Health [12], in 2017, in Vietnam, about 15% of the population suffered from a mental disorder related to depression, with 3 million people suffering from serious mental disorders. The Institute of Mental Health reports that 30% of the Vietnamese population has mental disorders, of which the rate of depressive disorders accounts for 25%, with 50% of people suffering from suicide disorders every year. From the fuzzy set theory [10] given by Zadeh in 1965, there have been many studies on the fuzzy set applications in medical diagnosis. In particular, many scientists pay attention to apply fuzzy logic to develop fuzzy systems to help diagnose diseases [8-11]. The main reason is the ability to incorporate fuzzy reasoning in uncertain information processing. Adlassnig developed a famous system called CADIAG-2 [1-3]. On the theoretical basis of CADIAG2, the author develops a fuzzy system to support diagnose depressive disorders. It is a computer program that captures disease symptoms, and uses the inference engine of the system working with knowledge bases consulted by medical physicians to determine whether the patient has a depressive disorder or not. When developing this system, we faced with the following problems: the symptoms as a decrease in mood, loss of all interest, enjoyment, reduced concentration, attention, etc. are fuzzy in nature; therefore, we use fuzzy logic and approximate inference methods in representing fuzzy positive rules of the systems. 2. DEVELOPING A FUZZY EXPERT SYSTEM BASED ON POSITIVE RULES FOR DEPRESSION DIAGNOSIS An expert system is a software developed on the theoretical basis of CADIAG2. The expert system is developed on web base model, Visual C # .NET programming language, and Microsoft SQL Server 2012 database. The expert system has a friendly interface, easy to use. The system's main components are the knowledge base, the Công nghệ thông tin 34 M. T. Nu, N. H. Phuong, “A fuzzy expert system based on for depression diagnoisis.” inference mechanism, and a unit of explanation of diagnosis results described below. The knowledge base is a part of the expert system, and the knowledge base contains the knowledge of human experts. To have this knowledge, the knowledge engineer has collected wisdom from human experts and then encoded it into the knowledge base through the knowledge representation method. There are many methods to organize and represent knowledge in computers; in this case, it chooses the method of knowledge representation by the "IF-THEN" production rules with some fuzzy degree. This method of representing knowledge effectively shows the fuzziness (uncertainty) in the knowledge of doctors. This method also has many advantages in accordance with a decision support model for diagnostics that is simple, easy to verify, easy to change, easy to modify, easy to expand, and take advantage of the experience knowledge of medical doctors. The (positive) rules of production take the following form: IF (ASSUMPTION) THEN CONFIRM (CONCLUSION) WITH (FUZZY DEGREE) Where "ASSUMPTION" is a symptom or a combination of symptoms that are combined by AND without using the NOT operator. “CONCLUSION” is a depression type. "FUZZY DEGREE" – a rule weight with the value in [0,1]. It indicates a degree of belief of the rule when "ASSUMPTION" is satisfied, then “CONCLUSION” will be confirmed. To ensure accuracy, the rule set in the knowledge base must satisfy the principle: there is not exist any two laws with the same premise and conclusion. The knowledge base of the expert system contains 857 positive rules, which include: 124 rules for diagnosing light depressive disorder; 146 rules for diagnosing middle depressive disorder, 263 rules for diagnosing serious depressive disorder and 324 rules for diagnosing depressive disorder with mental disorder. The number of rules will continue to increase as the system will be acquired knowledge from medical professionals. Most of the rules have been consulted by health professionals. To form these rules, the authors have listed all the clinical symptoms in patients diagnosed with depressive disorders, there are 13 such symptoms, of which: there are 3 typical symptoms, 7 universal symptoms, complications and 3 symptoms aggravate. The authors haves sorted these symptoms according to their frequency of occurrence in patients with every types of diagnosis. The authors then formulate all possible combinations of most typical, common, and aggravating symptoms and determine the extent to which this combination of symptoms confirms the type of diagnosis depressive disorder. The Inference Mechanism to process and control knowledge is represented in the knowledge base in order to respond to questions, user requests, and apply knowledge to solving practical problems. It is basically an interpreter for the knowledge base. Call S is the set of symptoms, S = {S1, S2, ..., Si, , Sn}, Si is the ith symptom. In our case, N = 13, including: Decreasing complexion, Losing all interest and pleasant, Decreasing energy, Decreasing attention, Decreased self-respectful and self-confident, Having idea of suicide, Feeling guilt, no worthy, feeling gray, Self-destructive / suicidal behavior ideas, Sleeping disorder, Eating disorders, Suicide, Delusions, Hallucinations. Nghiên cứu khoa học công nghệ Tạp chí Nghiên cứu KH&CN quân sự, Số Đặc san CNTT, 12 - 2020 35 Symptom Si take values μs, in [0,1]. The values μs, indicate the degree to which the patient exhibits the symptom Si. Intermediate combinations (fuzzy logical combinations of symptoms and diseases) were introduced to the model of the pathophysiological states of patients; and Symptom combinations 𝑆𝐶𝑖 are combinations of symptoms, diseases, and intermediate combinations. Both entities take values 𝜇𝐼𝐶𝑘and 𝜇𝑆𝐶𝑖 (respectively) in [0,1]. A relationship 𝑅𝑃𝑆𝐶 is established, defined by μRPSC(Pq, SCi) = 𝜇𝑆𝐶𝑖 for patient q where SCi = {𝑆𝐶1, , 𝑆𝐶𝑚} formally describes the symptom combinations observed on the patient. μRPCS(P,S) = min {μRPS1 (P,S1), μRPS2(P,S2), , μRPSi (P,Si),   μRSnP(P,Sn)} Call D is the set of diseases, D = { D1, D2,. , Dm}, Di is the ith depressive disorder. In our case, m=4, including: light depressive disorder, middle depressive disorder, serious depressive disorder and serious depressive disorder with mental disorder. A binary fuzzy relationship RPS is established, defined by μRPS(Pq, Si) = 𝜇𝑆𝑖 for patient Pq, where Pq = {𝑃1, 𝑃𝑝} and Si ∈ {𝑆1, , 𝑆𝑚}. μRPS(Pq, Si) [0,1]. A symptom - diseases relationship RPD is established, defined by μRPD(Pq, Dj) = 𝜇𝐷𝑗 for patient 𝑃𝑝, where Dj = {𝐷1, 𝐷𝑛𝑝} A fuzzy relationship RSD is established, defined by μRSD(S,DJ) [0,1]. This value represents the degree of confidence in the likelihood of having or not having DJ disease when a symptom or set of symptom S is present. Express the symptom-disease relationship as the follow: IF S THEN CONFIRM D WITH ( FUZZY DEGREE) RSD is now a confirming relationship that the patient has DJ disease when there is a symptom or set of S symptom S. The value μRSD(S,DJ) is fuzzy degree or rule weight. - μRSD(S, DJ) = 1 means the elementary conjunction S of symptoms iS definitely confirms the conclusion of DJ; - μRSD(S, DJ) = 0 means the elementary conjunction S of symptoms iS definitely not confirms the conclusion of jD ; - 0 < μRSD(S, DJ)< 1 means the elementary conjunction.S of symptoms i S confirms the conclusion of jD with some fuzzy degree. A fuzzy relationship RPD is established, defined by μRPD(P,DJ). Determining this relationship also means making a diagnosis of the patient's likelihood. Based on these fuzzy relationships, the MaxMin inference are used to deduce the fuzzy value μRPD(Pq,DJ) which indicates the degree of confirmation of disease Dj suffered by patient Pq from the observed symptoms. This MaxMin composition is the follow: RPD = RPS o RSD Where: - RPS is symptom S relationship or combination S (S1, S1,...,Sn) appeared in patient P. Công nghệ thông tin 36 M. T. Nu, N. H. Phuong, “A fuzzy expert system based on for depression diagnoisis.” μRPD(P,DJ) = maxSimin [μRPS(P,S) μRSD (S,DJ) ] (*) μRPD(P,DJ,rulet) = min [μRPS(P,S)  μRSD(S,DJ,rulet)] = min ({μRPS(Si,P)}, μRSD(S,DJ,rulet)) Where μRSD(S,DJ,rulet) is the degree of confirming DJ disease when there is an S symptom or a set symptom S on the rulet (weight of rulet). { μRPD (P,DJ, rulet), , μRPD(P,DJ, rule1), , μRPD(P,DJ, rulen)} t = 1,..,n. Calculate μRPD(P,DJ) from { μRPD(P,DJ,rulet) } according to the formula (*) μRPD(P,DJ) = maxSimin [μRPS(P,S) μRSD(S,DJ) ] μRPD(P,DJ) = maxSi [μRPD(P,DJ,rule1), . . ., μRPD(P,DJ,rulen)] - μRpd(P, DJ) = 1 means absolutely confirm of conclusion of DJ; - μRpd(P,DJ) = 0 means absolutely excludes of conclusion of DJ; - 0 < μRpd(P,DJ) < 1 means confirms the conclusion of jD with some fuzzy degree. The expert continues to acquire expert knowledge to complete the rules as well as determine the accuracy of the found results, so the author will choose the above formula with a high degree of confidence (accuracy). After acquiring expert knowledge, and verifying through practice, the formula will be improved and changed. Explain unit allows the program to explain its inference process to the user. These explanations include the arguments that justify the system's conclusions (answer to the how the system can gets the conclusion), explain why the system needs that data (the answer to the why question), and so on. To better understand the deductive process and diagnosis, The expert system can explain how and why to reach a certain conclusion about the possibility of a patient with a depressive disorder. During the diagnosis, the inference engine approves the rule and marks all the matching rules. When the diagnosis is completed, the explanation is formed by collecting all the satisfaction rules and step by step inference using each satisfaction rule. In this explanation, the expert system shows the diagnostic results, all of the patient symptom combinations used in the inference and the rules for individual patient satisfaction. Thus, the user can see intermediate diagnostic conclusions from the steps of the diagnostic process and how the rules affect the final conclusion. Algorithm include 5 steps: Input: list of symptom (confirmed symptoms appear in the patient after the examination). Step 1: Query the knowledge base, find all the rules for which the premise is a subset of the set "List of input symptoms"; Step 2: Browse this rule set, group of rules that have the same conclusion DJ; Step 3: Within each group of rules there is the same conclusion about DJ disease, with the rules of DJ disease. For the set of rules of DJ disease = DJ = {rule1, rule2, rulei ,, rulen), calculate μRPD(P,DJ) follows: Substep 3.1: For each rulet of DJ disease: Nghiên cứu khoa học công nghệ Tạp chí Nghiên cứu KH&CN quân sự, Số Đặc san CNTT, 12 - 2020 37 rulet : IF S THEN CONFIRM DJ With μRpd(P,DJ,rulet) S can be a symptom S or a set of symptom S = (S1, S1, ...Sn) of the disease DJ. Substep 3.2: calculate μRPD(P,DJ,rulet) - positive degree P patient confirm DJ disease calculate depend on rulei, using formular (*) μRPD(P,Dj) = maxSimin [μRPS(P,S) μRSD(S,Dj) ] μRPD(P,DJ,rulet) = min [μRPS(P,S)  μRSD(S,DJ,rulet)] = min ({μRPS(P,Si)} , μRSD(S,DJ,rulet)) Calculate μRPD(P,DJ) from set of { μRPD(P,DJ,rulet) } μRPD(P,DJ) = maxSimin [μRPS(P,S) μRSD(S,DJ) ] μRPD(P,DJ) = maxSi[μRPD(P,DJ,rule1), . . ., μRPD(P,DJ,rulen)] Step 4: Similar calculation with other diseases; Step 5: Make a final conclusion based on the results obtained after the calculation. 3. TEST In this section, the result of diagnosis of depressive disorder will be evaluated and compared with the doctor's diagnosis in the medical record. 3.1. Experimental Result These tests implementation was conducted on the data set in the medical records collected from National Hospital of mental diseases. Data set includes medical records of patients such as with light depressive disorder, middle depressive disorder, serious depressive disorder and serious depressive disorder with mental disorder. In these tests with data sets of patients diagnosed by doctors with depressive disorders. From there, it is possible to evaluate, the accuracy of the expert system compared to the doctor's diagnosis in the medical record. This data set included 244 patients who came to the National Psychiatric Hospital for inpatient examination and treatment. Each medical record contains information related to examination and treatment of depressive disorders. Out of 244 medical records, 48 were diagnosed with mild depressive disorder, 60 were diagnosed with moderately depressive disorder, 50 were diagnosed with a disorder. severe depression and 86 medical records diagnosed with severe depressive disorder with psychosis. The following information is extracted from the medical record, relevant to the doctor's diagnosis of depressive disorder (some information is protected, some administrative information of the patient is not required to report). The information includes: - Administrative information: Patient name, age, gender, phone number, address. - Disease information: Decreasing complexion, Losing all interest and pleasant, Decreasing energy, Decreasing attention, Decreased self-respectful and self-confident, Having idea of suicide, Feeling guilt, no worthy, feeling gray, Self-destructive / suicidal behavior ideas, Sleeping disorder, Eating disorders, Suicide, Delusions, Hallucinations. The above information constitutes 13 input attributes for the test. This information is "fuzzialize" into fuzzy values and is valid in the range [0,1]. 3.2. Test and evaluate Công nghệ thông tin 38 M. T. Nu, N. H. Phuong, “A fuzzy expert system based on for depression diagnoisis.” In the test with the patient data set, the author fully updated the disease information of 244 medical records in the expert system software. The diagnostic results of the expert system were compared with that in the medical records, giving details as shown in the table below. Table 1. Compare the results of diagnosis for each type of depressive disorder. Type of depressive disorder Total In medical records In the expert system software rate Light depressive disorder 48 48 46 95,8% Middle depressive disorder 60 60 Inappropriate Serious depressive disorder 50 50 Inappropriate Serious depressive disorder with mental disorder 86 86 Inappropriate The above experimental results show that the expert system gave good results for light depressive disorder, the remaining depressive disorders are not accurate because diagnostic standards of these 4 types of depression overlap some symptoms, for example, serious depressive disorder with mental disorder when the patient has serious depressive disorder with suicide or delusions or hallucinations symptoms. 4. CONCLUSION This paper presents a development and proposal a fuzzy expert system based on positive rules for diagnosing depressive disorders. The test results of 244 patients gave good results for light depressive disorder. However, achieving good diagnostic results for all types of depressive disorders requires time and experience by "trial and error" many times to determine the complete values and functions for each problem. This is also a limitation of building the knowledge base and inference mechanism in the medical diagnostic specialist system. In the coming time, the author will continue to improve the system by maintaining and updating new rules for the knowledge base and improving the inference mechanism for the expert system. REFERENCES [1]. Adlassnig, K-P, “A Fuzzy Logical Model of Computer - Assited Medical Diagnosis”, Methods of information in Medicine 19/3, 141-148, 1980. [2]. Adlassnig, K-P, “Fuzzy Sets Theory in Medical Diagnosis”, IEEE Transaction on Systems, Man and Cybernetics, Vol-SMC-16, No.2, 260-265. [3]. Adlassnig, K-P., “CADIAG-2: Computer – Assisted Medical Diagnosis Using Fuzzy Subsets”. In Gupta, M.M. & Sanchez, E. (Eds.) Approximate Reasoning in Decision Analysis, North-Holland Publishing Company, Amsterdam, 219 - 247. 1982. [4]. ICD - 10, Medical Publiser, Vietnam, 2010 (in Vietnamese and in English). [5]. J.-S.R.Jang, C.-T.Sun, E.Mizutani – “Neuro-Fuzzy and Soft Computing, Acomputational Approach to Learning and Machine Intelligence” – Prentice Hall Upper Saddle River, NJ 07458. [6]. Mai Thi Nu, Nguyen Hoang Phuong, Hoang Tien Dung, “STRESSDIAG: A Fuzzy Expert System for Diagnosis of Stress Types including Positive and Negative Rules”, 2019 IFSA World Congress and NAFIPS Annunal Conference, June 18-22, 2019, USA, pp. 371-381, (In Book: Fuzzy Techniques: Theory and Applications”. Springer, Cham, 2019) Nghiên cứu khoa học công nghệ Tạp chí Nghiên cứu KH&CN quân sự, Số Đặc san CNTT, 12 - 2020 39 [7]. Mai Thi Nu, Nguyen Hoang Phuong, K. Hirota, “Modeling a Fuzzy Rule Based Expert System combining Positive and Negative Knowledge for Medical Consultations using the importance of Symptoms”. In Proc. of IFSA-SCIS’2017, Otsu, Japan, June 27-30, Paper ID: #164, 2017. [8]. Nguyen Hoang Phuong, “Towards intelligent Systems for integrated Western and Eastern Medicine”. TheGioi Publishers, Hanoi, 1997. [9]. Nguyen Hoang Phuong, V. Kreinovich, “Fuzzy Logic and its Applications in Medicine”, International Journal of Medical Informatics 62, 165-173, 2001. [10]. Zadeh, L.A. Fuzzy Sets, Information and Control, Vol. 8, 338-353, 1965. [11]. Zadeh, L.A., “The role of fuzzy logic in the management of uncertainty in expert systems”, Fuzzy Sets Syst. , 11, 199, 1983. [12]. [13]. https://www.who.int/mental_health/management/depression/en/. TÓM TẮT HỆ CHUYÊN GIA MỜ DỰA TRÊN LUẬT KHẲNG ĐỊNH CHO CHẨN ĐOÁN RỐI LOẠN TRẦM CẢM Lý thuyết tập hợp mờ và logic mờ là một công cụ toán học rất thích hợp để phát triển các hệ thống thông minh trong y học. Bài báo này trình b