Objectives of the research - Research on development a rule-based expert system for diagnosis of depressive disorder; - Building a positive rule base and a negative rule base for the ex
Trang 1ACADEMY OF MILITARY SCIENCE AND TECHNOLOGY
MAI THI NU
DEVELOPMENT RESEARCH ON FUZZY EXPERT SYSTEMS FOR
THE DEPRESSIVE DISORDERS DIAGNOSIS
Specialization: Mathematical Foundation for Informatic Code: 9 46 01 10
SUMMARY OF PhD THESIS IN MATHEMATICAL
Ha Noi, 2021
Trang 2ACADEMY OF MILITARY SCIENCE AND TECHNOLOGY MINISTRY OF NATIONAL DEFENCE
Scientific Supervisor:
1 Assoc Prof Dr Nguyen Hoang Phuong
2 Dr Duong Tu Cuong
Reviewer 1: Assoc Prof Dr Nguyen Tan An
Reviewer 2: Assoc Prof Dr Nguyen Long Giang
Reviewer 3: Dr Nguyen Manh Linh
The thesis was defended at the Doctoral Evaluating Council at Academy level held at Academy of Military Science and Technology
at … date ……., 2021
The thesis can be found at:
- The Liblary of Academy of Military Science and Technology
- Viet Nam National Liblary
Trang 31 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,
2017, Paper ID: #164
(Indexed in Scopus)
2 Mai Thi Nu, Nguyen Hoang Phuong, Hoang Tien Dung
STRESSDIAG: A Fuzzy Expert System for Diagnosis of Stress Types including Positive and Negative Rules, IFSA World Congress and NAFIPS Annunal Conference, June 18-22, 2019, USA, pp 371-
381, In Book: Fuzzy Techniques: Theory and Applications” Springer, 2019, volumn 1000
(Indexed in Scopus)
3 Mai Thi Nu, Nguyen Hoang Phuong A Fuzzy Expert System based
on positive rules for Depression Diagnoisis, Tạp chí Nghiên cứu khoa học và công nghệ quân sự - Viện Khoa học và Công nghệ quân
sự, Số đặc san CNTT, 12/2020, pp 33-39
Trang 4INTRODUCTION
1 The necessary of the thesis
According to the World Health Organization, "Depression" is a common mental disorder characterized by sadness, loss of interest, guilt or low self-esteem, disturbed sleep, or appetite, feeling tired and poor concentration It affects about 264 million people worldwide Especially when prolonged and
in moderate or severe intensity, depression can become a serious health condition It can cause serious harm and ineffectiveness at work, school and home Most worryingly, depression can lead to suicide Depression is the fourth leading cause of death worldwide, the second leading cause of death among young people aged 15-29, projected to be the second leading cause of death by 2030
According to statistics of the Ministry of Health in 2017, in Vietnam, about 15% of the population suffer from common mental disorders related to depression, 50% of people suffer from serious depressive disorders Although they are many effective treatments for depression but the proportion of people in low-income and middle-income countries who do not receive treatment for depressive disorder remains high There are many barriers to effective health care such as lack of resources, lack of medical professionals and social stigma Another barrier to effective health care is inaccurate diagnosis because the symptoms of a depressive disorder are difficult to assess and subjective, leading to inaccurate prescriptions Therefore, the treatment effect is not high
Experienced team of doctors and equipment are two of the dilemmas in the battle with disease According to statistics of the Ministry of Health, the average rate is about 8-9 doctors / 10,000 people The percentage of doctors specializing in psychiatry is even less, psychiatrists are always in a serious shortage, while the majority of qualified doctors are concentrated at the central level
Thus, it is necessary to exploit and use the knowledge of good specialists
at the central level hospitals to train and share experience in examination and treatment of depression disorders for doctors who do not have much experience, especially, for young medical doctors There are many methods
to exploit and use knowledge, the expert system method is one potential approach to develop intelligent systems for disease disgnosis with high accuracy and high confidence to users
Having many disease diagnostic support systems, the expert system method is the most appropriate approach, because the expert system method has a way of expressing knowledge by the production rules (IF… THEN…) This method has many advantages such as its syntactic closeness to the natural
Trang 5language descriptions of medical doctors, the expression is quite simple and intuitive, and can be deduced according to different strategies: progressive inference, backward inference, mixed inference (forward - backward), can check the inconsistency between rules, high modularity, that is, adding, removing, removing rules completely does not affect the other rules as well
as the inference mechanism
Usually, the diagnostic criteria according to the ICD-10 diagnostic criteria include clinical symptoms such as "hypochromia", "loss of interest"
To diagnose depressive disorders, the symptom gets the values "yes" or "no"
In the fact, the symptoms are fuzzy data, can be represented by fuzzy sets and get values in the range [0,1] Thus, in order to reduce the possibility of losing information and correct representation of disease symptoms, patient data should be represented by fuzzy sets
2 Objectives of the research
- Research on development a rule-based expert system for diagnosis of depressive disorder;
- Building a positive rule base and a negative rule base for the experts system for depressive disorders diagnosis;
- Helping medical doctors in the provinces and country sides, contributing to more accurate diagnosis, reducing waiting time, reducing overload for specialized hospitals, upper psychiatric departments
3 Object and scope of the research
- The research object of the thesis is depressive disorders, the expert system, the CADIAG-2 expert system
- The research scope of the thesis topic is the fuzzy expert system based
on the rules for depressive disorder diagnosis representation of positive knowledge, negative knowledge and inference mechanisms combining positive knowledge and negative knowledge for fuzzy expert system
4 Content of the research
- Research on the mental health care system in Vietnam, methods of depressive disorders diagnosis, symptoms and diagnostic procedures for depressive disorders;
- Overview of research of fuzzy set theory, the expert system, the fuzzy expert system, the fuzzy expert system based on the rules, the CADIAG-2 expert system
- Researching and developing the positive rules and negative rules for depressive disorders diagnosis;
- Researching and proposing a model of an expert system based on the rule using positive rules for depressive disorder diagnosis
- Researching and proposing a model of the expert system based on the
Trang 6rules combining positive rules and negative rules for depressive disorders diagnosis
- Experimental implementation of the expert system with 264 depressive disorder patients
- Evaluating the experimental results of the expert system, recommending the applications of the expert system in the diseases diagnosis
5 Research method
- Theoretical research method: theoretical research on methods of diagnosing depressive disorder, fuzzy set theory, expert system, CADIAG-2 expert system;
- Experimental research method: collecting data on patients with depressive disorders; developing symptom-base, disease-base, and rule-base for expert systems; performing experiments on expert system software with collected data sets; evaluation of experimental results
6 Structure of the thesis
Apart from the introduction, the main contents of the thesis consists of three chapters:
- Chapter 1 presents some notions of depressive disorders and fuzzy expert systems, specifically the mental healthcare network in Vietnam, methods of diagnosing depressive disorders; fuzzy set theory, expert system, rule-based fuzzy expert system, CADIAG-2 expert system; Researching situation of development of diagnosing experts systems of depressive disorders in Vietnam and in the world, evaluation of limitations of studies about diagnostic systems of depressive disorders , and then proposing research directions of the thesis Through this chapter, the thesis shows an overview of the research problem of the thesis;
- Chapter 2 researches to propose a model of fuzzy expert system based
on rule, using positive knowledge (positive rules for confirmation of conclusion) in diagnosing depressive disorders, this expert system is developed on CADIAG-2 expert system’s Max-Min approach; The contents presented include: knowledge base of expert system (symptom base, disease base, positive rules base); inference mechanism of the expert system; Experimental system with 264 cases of patients with depressive disorder; and analysis and evaluation of experimental results
- Chapter 3 presents the research and proposed model of fuzzy expert system based on the rules combining positive knowledge (positive rules) and negative knowledge (negative rules); expert system was developed, improved from fuzzy expert system based on the rules of using positive knowledge for diagnosing depressive disorder which was studied in chapter 2; presenting improvements of representing rules in knowledge base, improvement in the
Trang 7inference mechanism combining positive rules with negative rules; expert system testing with 264 cases of patients with depressive disorders; and analysis and evaluation of experimental results
The conclusion part shows the research results of the thesis, the contribution of the thesis, the next research direction Finally, there are published works of the thesis
Chapter 1 Overview of depressive disorders and fuzzy expert systems 1.1 Overview of the depressive disorders diagnosis
* Tests for depressive disorder
- PHQ-9 consists of 9 questions, PHQ-9 is designed to screen and help monitor the patient's response to treatment status Evaluating of depressive disorder by total score: 0 ≤ total score ≤ 4 is no depression; 5 ≤ total score ≤
9 isrisk; 10 ≤ total score ≤ 14 is light depressive disorder; 15 ≤ total score ≤
19 is middle depressive disorder; 20 ≤ total score ≤ 27 is serious depressive disorder
- BECK includes 21 different indexes, BECK has been recognized by the World Health Organization for treatment Evaluating depressive disorder by total score, the total score <14 is no sign of depression, 14 ≤ the total score ≤
19 is light depressive disorder, 20 ≤ the summary score ≤ 29 is middle depressive disorder, the summary score ≥ 30 is serious depressive disorder HAMILTON is a reference tool to check your own condition HAMILTON includes 17 items Evaluating depressive disorder by total score, total score <14 is no depression, total score ≥ 14: beginning to show signs of depressive disorder
- ZUNG to measure and diagnose the cause of depression in patients ZUNG has 20 descriptions Maximum total score is 80 Evaluating depressive disorder by total score, total score <50 points is no depressive disorder; iscore
≥ 50 is have depressive disorder
* Diagnostic standard according to DSM-5 diagnostic standard DSM-5 has 9 symptoms Mild depressive disorder consists of 5-6 symptoms; Moderate depressive disorder consists of 7-8 symptoms; Severe depressive disorder consists of all 9 symptoms; Severe depressive disorder with psychosis consists of 9 symptoms and psychotic symptoms (hallucinations, delusions) Among them, at least one of the symptoms should be: a depressed mood or loss of interest or pleasure
* Standard diagnosis according to ICD-10 standards
ICD-10 has 03 main symptoms, 07 common symptoms, 03 aggravating symptoms, psychosis Mild depression: There are 2 main symptoms and 2 common symptoms; Moderate depression: has 2 main symptoms and 3-4 common symptoms; Severe depression: has 3 main symptoms and at least 4
Trang 8or more common symptoms; Severe depression with psychosis: satisfying symptoms of major depression with at least one psychotic symptom
1.2 Fuzzy set theory and fuzzy logic
* For the universe set or reference set X - which is not empty set The fuzzy set A in the universe X is a function of: A : X [0,1] x X defines
a membership value of x in the fuzzy set A as A A(x) [0,1] The characteristic function XA of a set A in set theory: XA : X {0, 1} XA{0, 1} such that: XA(x) = 1 x A; XA(x) = 0 x A
* The math operations on the fuzzy sets
- Intersection of two fuzzy sets: A B (x) = T (A(x),B(x)) x M, in which: fuzzy set A, B on space M with the functions ofA , B ; T là is a t-norm standard
- The union of two fuzzy sets: A B (x) = S (A(x),B(x)) x M, in which: fuzzy set A, B on space M with the functions of A,B ; S is a t-conorm standard
- The drag is a function L: [0,1]2 [0,1] satisfies If x z then L(x,y)
L(z,y) y [0,1]; If y u then L(x,y) L(x,u) x [0,1] => L(0,x) = 1 x
[0,1]; L(x,1) = 1 x [0,1], L(1,0) = 0
- The negation of a fuzzy set: The negation function N: [0,1] [0,1], does not increase satisfying the conditions N(0) = 1, N(1) = 0 is called a coverage function concentration Some negations: Standard negation functions n(x) = 1-x; Negation function: n(x) = 1-x2; Standard induction coverage function: n(x) = 1 for x = 0, n(x) = 0 for x > 0
- The offset of the fuzzy set A has the basis M and the belonging function
A(x) is a fuzzy set AC defined on the same basis M with the belonging function: Ac(x) = n(A(x))
1.3 The expert systems
* The main component of an expert system is the Knowledge Base and Inference Mechanism
* The knowledge representation depend on the rule of production in hệ chuyên gia: IF (Condition 1, Condition 2,…, Condition m) THEN Conclusion (1.1)
* MYCIN is an expert system diagnosis of sepsis MYCIN uses a knowledge base of about 500 rules, the rules are represented as production rules with a certain theoretical approach
* The fuzzy inference: fuzzy rule IF A THEN B A' has the same set of universe as A, fuzzy inference will show that the conclusion B' has the same set of universe as B Each fuzzy rule of the form IF A THEN B can be represented by a fuzzy relation matrix M of size n x m (where n, m are the
Trang 9cosmic force of the fuzzy set A and B) There are two ways to construct this matrix: (1) Max-Min: Mij = min{ A(ai) , B(bj)} and (2) Max-Product Mij =
A(ai) x B(bj), B’j = maxmin{A(ai),Mij)} (i=1 n); the max is equivalent to the t-conorm standard, the min is equivalent to the t-norm standard
* CADIAG-2 Medical Expert System
- Call Pq is patient qth, Pq∈ [P1, , Pp]; Si is symptom ith, Si∈ [Si, … , Sn],
Dj is disease jth, Dj∈ [D1, , Dm] The value of symptom Si of patient Pq is
μPS(Pq, Si), SC = {S1 S2 Sn}, fuzzy value of SC symptom combination
of patient Pq is μRps(Pq, SC) = minSi ∈ SC (μRps(Pq, Si)) (1.2) Then, the degree
of disease Dj of patient Pq is μRPD(Pq, Dj) ∈ [0,1]; The relationship between
RSD: RPD= RPS o RSD SC = {S1 S2 Sn} then: RPD= RPSC o RSCD; The relationship between patient Pq and disease Dj is represented by the value
μRPD(Pq, Dj) = max min { μRPS(Pq, Si); μRSD(Si, Dj)}; In case of symptom set:
1.5 Conclusion of chapter 1
In chapter 1 presented the overview of research on depressive disorders and fuzzy expert system, comparing the results of case studies on building expert systems for diagnosing depressive disorders
The following chapters will present proposals on developing a rule-based fuzzy expert system for diagnosing depressive disorders according to ICD-10 diagnostic standards, including: 1) Developing a rule base of positive and negative rules for the diagnosis of depressive disorders 2) Developing a model of expert system based on the rule using positive knowledge for diagnosis of depressive disorders 3) Proposing to improve the expert system model based on the rule combining positive knowledge and negative knowledge for diagnosing depressive disorders 4) The experimental implementation of the expert system models on collected data and analysis
of diagnostic results of depressive disorders
Trang 10Chapter 2 Modeling a Fuzzy expert system based on positive rules for
depression diagnosis
This chapter presents a model of fuzzy expert system based on the use of positive knowledge for depression diagnoisis called is PORUL.DEP and experiments with 264 cases of patients, including:
- Building the knowledge base of PORUL.DEP expert system, including symptoms data, diseases data, a positive rule base
- Developing on the inference mechanism of the PORUL.DEP expert system
- Collecting experimental data
- Experimenting and evaluating on experimental results
2.1 Theoretical basis for research results
max min { μRPS(Pq, Si); μRSD(Si, Dj)}; Value of symptom combination SC:
μR
PS(Pq, SC) = minSi∈SC{μR
PS(Pq, Si)}; the positive relationship between patient and disease: μRPD(Pq, Dj) = max min {μRPS(Pq, Si)μRSD (Si, Dj)}; Determine the degree of confirmation of disease Dj for the rulet là
of confirmation of the possibility of Dj of patient Pq based on the set of rules
= {rule1, rulet2, … ruleh ,…, rulep) : μRPD(Pq, Dj) = max {μR
2.2 Developing on a model of fuzzy specialist system based on the rule using positive knowledge for depressive disorder diagnosis
Recalling the preface made by Zadeh, in narrow sense, fuzzy logic, FLn
is a logical system which aims at a formalization of approximate reasoning
In the broad sense, every thing dealing with fuzziness may be called “fuzzy logic” “Fuzzy logic” in medicine in broad sense In depressive disorder, all
symptoms are emotional The word “Decreasing complexion”, “Decreasing
energy”, “Decreasing attention”, are all in the rules, are fuzzy concepts, are fuzzy logic in the broad sense However, it is difficult to formalize and measure it like some other symptoms, such as the a “high fever” symptom
* Building in knowledge base of PORUL.DEP
- The basis of symptom: Set of symptoms S = {S1, S2, , Si, … , Sn}; Si
is ith symptom, i = 1 n, n = 13, the value of Si symptom is μSi [0, 1], this value represents the degree of Si symptom appearing in the patient S1= Decreasing complexion, S2 = Losing all interest and pleasant, S3 = Decreasing energy, S4 = Decreasing attention, S5 = Decreased self-respectful and self-confident, S6 = Having idea of suicide, S7 = Feeling guilt, no worthy, feeling
Trang 11S10 =Eating disorders, S11= Suicide, S12 = Delusions, S13 = Hallucinations
- The basis of diseases: Set of diseases D = { D1, D2, … , Dm}, Dj is Dj is the jth depressive disorder, j=1 m, in the PORUL.DEP system, m = 4 The degree of Dj disease is μDj[0, 1] D1 = light depressive disorder, D2 = middle depressive disorder, D3 = serious depressive disorder and D4 = serious
depressive disorder with mental disorder
- The basis of positive rules: IF (hypothesis) THEN (conclusion), FD; FD
is the degree of conclusion FD (Fuzzy Degree) is the weight of the rule Table 2.1 Example of the positive rules for light depressive disorder
if Losing all interest and pleasant Then light depressive disorder 0, 35
if Decreasing energy Then light depressive disorder 0, 35 Table 2.2 Example of the positve rules for middle depressive disorder
if
Decreasing complexion; Decreasing
attention; Losing all interest and pleasant;
Feeling guilt; Having idea of suicide
Then middle depressive disorder
1
if
Decreasing complexion; Decreasing
attention; Losing all interest and pleasant;
Having idea of suicide; Self-destructive /
suicidal behavior ideas
Then middle depressive
Table 2.3 Example of the positive rules for serious depressive disorder
if Self-destructive / suicidal behavior ideas Then serious depressive disorder 0,05
if Sleeping disorder Then serious depressive disorder 0,05
Table 2.4 Example of the positive rules for serious depressive disorder with
mental disorder
if
Decreasing complexion; Losing all interest
and pleasant; Decreasing energy; Having
idea of suicide; Feeling guilt, no worthy,
feeling gray; Self-destructive / suicidal
behavior ideas; Delusions
Then
serious depressive
disorder with
mental disorder
0,73
if
Decreasing complexion; Losing all interest
and pleasant; Decreasing energy;
Decreasing attention; Self-destructive /
suicidal behavior ideas; Sleeping disorder;
Eating disorders; Suicide; Delusions
Then
serious depressive
Trang 12Set of symptoms S = {S1, S2, , Si, … , Sn}; n=13
Set of diseases D = { D1, D2, … , Dm}, m = 4
μRPS(Pq,Si) [0, 1] SC is the symptom combination, Si is ith symptom, i, i = 1,…,m, S = S1 S2, , Sm, then
μR
PS(Pq, SC) = minSi∈SC{μR
PS(Pq, Si)} (2.2)
- Symptoms and Disease Relationship denoted by RSD is present μRSD(Si,
Dj) μRSD(Si, Dj) represents the degree of disease Dj when appears symptom
Si μRSD(Si, Dj)is the weight of the positive rule μRSD (Si, Dj)= 1 means confirming the possibility of disease Dj when appears symptom Si μRSD(Si,
Dj)= 0 means that the possibility of disease Dj is not confirmed when appears symptom Si (does not mean negative) 0 < μRSD(Si, Dj)< 1 means confirming
μRSD(Si, Dj) [0, 1]
- Patient and Patient Relations are denoted by RPD, RPD is present μRPD(Pq,
Dj) [0, 1] μRPD(Pq, Dj) = 1 means completely confirmed that patient Pq has disease Dj.μRPD(Pq, Dj) = 0 means that patient Pq is not confirmed with disease
Dj (does not mean negative) 0 < μRPD(Pq, Dj) < 1 means that patient Pq has disease Dj with a fuzzy value is μRPD(Pq, Dj) [0, 1] RPD is determined to be
a combination of two components RPS và RSD RPD = RPSo RSD Calculate
μRPD(Pq, Dj) by Max-Min inference of CADIAG-2:
μRPD(Pq, Dj) = maxSimin {μRPS(Pq, Si)μRSD (Si, Dj)}
{rule1,…,rulet,…,rulen) with disease Dj, rulet is the tth rule, t=1 n Then
μR
according to the rulet
Calculate μRPD(Pq, Dj) from { μR
μRPD(Pq, Dj)=max {μRPD rule1(Pq, Dj), , μRPD rulen(Pq, Dj)} (2.4)
* The algorithm of PORUL.DEP is described in Figure 2.3
The description of PORUL.DEP algorithm is as follows:
Input value symptoms (symptoms defined appear in patients after the medical examination)
Step 1: Query knowledge base, find all rules where the premise is a subset of the set "Input symptom value" Step 2: Review this set of rules, group of rules with
Trang 13Figure 2.3 The algorithm diagram of PORUL.DEP
Step 3: In each pattern of rules, there is the same conclusion that disease D j { rule 1 , rule 2 , rule t , , rule n ), calculate μRPD(P q , D j ) as follows:
Step 3.1: For each rule, there is a conclusion of disease: Rule D j : IF S THEN D j , FD, S can be a symptom or a symptom combination {S 1 , S 1 , S n }
Step 3.2: Calculate μRPDluậtt(P q , D j ) is the confirmed value
of the patient with disease D j according to the rule t , using formula 2.3 in section 2.2, obtained the set of the degree