A DESIGN OF DIAGNOSIS SYSTEM FOR MENTAL DISORDER

Mariko Sasakura, Kenichi Iwata and Susumu Yamasaki

Department of Computer Science, Okayama University, 311, Tsushima-Naka, Kita-ku, Okayam, Japan

Keywords:

Knowledge engineering, Diagnosis.

Abstract:

We propose a web application system which helps experts to make a diagnosis of mental disorders. We

construct a knowledge base from DSM and SCID which are diagnosis manuals made by psychiatrists. We

have two stages in our diagnosis system. In the ﬁrst stage, our application picks up suspicious disorders. In

the second stage, the application checks the suspicious disorders according to a ﬂow chart generated by the

knowledge base. We discuss problems gained by our experiment of constructing the knowldge base.

1 INTRODUCTION

Diagnosis can be included an area of knowledge en-

gineering. This view is supported by the existance of

many medical expert systems (Phuong et al., 2000;

Starita et al., 1995).

In this paper, we present an experiment system to

construct a prototype system for assisting to make a

diagnosis of a mental disorders. Diagnosis of men-

tal disorders is difﬁcult because of a lack of objective

biological markers. Therefore, to make a diagnosis,

psychiatrists are required a great deal of knowledge

of disorders and symptoms.

To assist to make an accurate diagnosis, we con-

struct a system by rearranging knowledge from DSM

(Diagnostic and Statical Manual of Mental Disor-

der) and SCID (The Structured Clinical Interview for

DSM), which are manuals to make a diagnosis of

mental disorder. DSM and SCID are a kind of knowl-

edge base constructed by experts: psychiatrists. The

original DSM or SCID are designed for checking all

disorders’ criteria. The process is time-consuming. In

order to decrease the time, our system is designed to

make a diagnosis by two stages. In the ﬁrst stage, the

system lists up suspicious disorders from symptoms.

In the second stage, the system assists an expert in

making a diagnosis of the selected disorders from the

result of the ﬁrst stage. In this stage, the expert checks

the criteria of suspicious disorders according to the

description of SCID, which are displayed on the com-

puter screen. To implement the system, we have to

reconstruct the knowledge of DSM for the ﬁrst stage.

In Section 2, we introduce the details of DSM and

SCID. In Section 3, we describe the knowledge base

we constructed. In Section 4, we show the prototype

system we developed. In Section 5, we discuss prob-

lems we ﬁnd out during our experience, and in Sec-

tion 6, we give a conclusion.

2 DIAGNOSIS MANUAL OF

MENTAL DISORDER

In this section, we mention the details of DSM (Amer-

ican Psychiatric Association, 2000) and SCID (First

et al., 2002) which are basis of our knowledge base.

2.1 DSM

Diagnosis of mental disorders has difﬁculties because

most of symptoms are subject of patients and not ob-

jective. Psychiatrists must give a diagnosis which dis-

order a patient has by major complaints and an obser-

vation of the patient. Therefore, the different patients

used to be given different diagnosis to the same symp-

toms.

American Psychiatric Association organizes the

relation of symptoms and the names of disorder as

a manual called “Diagnostic and Statical Manual of

Mental Disorder” (DSM). In this manual, psychia-

trists give the same disorder to the same symptoms.

The latest version of DSM is DSM-IV-TR published

in 2000.

368

Sasakura M., Iwata K. and Yamasaki S. (2009).

A DESIGN OF DIAGNOSIS SYSTEM FOR MENTAL DISORDER.

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 368-371

DOI: 10.5220/0002271203680371

 SciTePress

Table 1: DSM example : a part of criteria of panic disorder

without Agoraphobia.

A. Both (1) and (2):

(1) recurrent unexpected Panic Attacks

(2) at least one of the attacks has been followed

by 1 month (or more) of one (or more) of the fol-

lowing:

(a) persistent concern about having additional

attacks

(b) worry about the implications of the attack

or its consequences (e.g., losing control, having a

heart attack, “going crazy”)

the attacks

B. Absence of Agoraphobia.

(The rest is omitted.)

In DSM, a diagnostic criteria is given to each dis-

order in a natural language.

Table 1 shows a part of criteria of panic disor-

der. The criteria are declarative description. To judge

whether symptoms of a patient satisfy the criteria, the

knowledge of experts is needed. Therefore, only ex-

pert psychiatrists should be permitted to make a diag-

nosis using DSM.

2.2 SCID

”The Structured Clinical Interview for DSM-IV”

(SCID) is a semi-structured interview for the forth

version of DSM.

SCID is a list of questions which are constructed

like a ﬂowchart. An expert can make a diagnosis fol-

lowing SCID. SCID has same contents of DSM but

the contents are described in a procedural way. It

shows a procedure of a diagnosis.

Table 2 shows a part of list of panic disorder cri-

teria. Some questions in the list need specialized

knowledge to answer. SCID also needs expert psy-

chiatrists to perform.

The two problems for SCID have been pointed

out. DSM also has the problems because SCID and

DSM are same essentially. The ﬁrst problem is that

it is time-consuming to deliver, because it is designed

to check all disorders’ criteria in order. The second

problem is that the quality of the diagnosis depends

on the expert who uses it, because some questions in

SCID needs expert knowledge to answer.

Table 2: SCID example: a part of list of panic disorder with-

out Agoraphobia.

No. Questions Next question

F1 recurrent unexpected

panic attacks

if negative goto F25,

else goto F2

F2 at least one of the

follwoing: (b) worry

about the implications

of the attack; (a) con-

cern about having ad-

ditional attack; (c) a

signiﬁcant change in

behavior

if negative goto F25,

else goto F3

F3 Four (or more) of the

following panic attack

symptoms developed

abruptly and reached

a peak within 10

minutes

if negative goto F25

F4 (1) palpitations

F5 (2) swaeting

(The rest is omitted.)

3 A DESIGN OF KNOWLEDGE

STRUCTURE

Both DSM and SCID aim to diagnose a speciﬁed dis-

order. Usually, a psychiatrist may see a patient, and

check a criteria of the disorder of DSM or SCID. The

problem of DSM or SCID is time-consuming to de-

liver, because they are constructed for cheking all dis-

orders’ criteria.

Therefore, we divide a process of diagnosis into

two stages. The ﬁrst stage based on declarative

knowledge, DSM, is used to ﬁnd suspicious disor-

ders. The second stage based on procedural knowl-

edge, SCID, is used to diagnose the disorders.

3.1 The First Stage

In the prototype system we developed, the ﬁrst stage

is implemented simply.

1. The system lists up all symptoms related disor-

ders.

2. An expert decides a threshold rate.

3. An expert checks all symptoms that a patient has.

4. If the number of checked symptoms related to a

disorder goes over the threshold rate, the system

picks up the disorder.

5. All disorders the system picked up are displayed

on the screen for the second stage.

A DESIGN OF DIAGNOSIS SYSTEM FOR MENTAL DISORDER

369

3.2 The Second Stage

The second stage will start at a disorder that is listed

up by the ﬁrst stage of the system. The expert looks

the list of the suspicious disorders at ﬁrst, then picks

up a disorder that seems to be the possibly true dis-

order of the patient. Then the veriﬁcation stage (the

second stage) starts on here.

The system asks a question that follows SCID,

and then the expert answers. This answer includes

a thought of the expert. The expert interacts with

the patient, and judges how to answer to the ques-

tion from the system. This means the expert interacts

with the system also, not only with the patient. The

system stored the answer to the database, then shows

a new question to the expert. The system shows also

the process of the ﬂow. This helps the expert to judge

the disorder of the patient. The expert is possible to

think about the answer of the question from the sys-

tem too.

The cycle is continued until the ﬂow reaches to the

end. At the end of the ﬂow, the system shows if the

suspicious disorder of the patient would truly match

to the symptoms, or not.

The questions in the second stage of SCID have

3-valued logic. The answer for the question is one of

?, -, or +. The answer is interpreted in : ? means inad-

equate information, - means absent or sub-threshold

(negative), and + means present (positive). If the an-

swer is +, the patient has the symptom asked by the

question.

The questions of SCID can be categorized into

two types following:

Type 1. The next question is decided by the answer

of the this question.

Type 2. The next question is decided by the answers

of the previous questions including this question.

In this type of question, sometimes more than 10

previous questions are related to decide the next

question.

The system supports these two types of nodes and

controls ﬂows of questions in the second stage.

4 SYSTEM DESIGN

The proposed system aims to assist psychiatrists in

making a diagnosis. It is designed as a Web-based

system. The architecture of the system is shown in

Figure 1.

The system consists of four parts: Back End DB,

Suspicious Disorders List Up Engine, SCID engine

Back End

Expert

Patient

Suspicious

Disorders

List Up

Engine

SCID Engine

The System

The First Stage

The Second Stage

Browser

Figure 1: The system architecture.

Figure 2: The ﬁrst step.

and Browser. Back End DB keeps knowledge for di-

agnoses and records of diagnoses. Suspicious Disor-

ders List Up Engine is an implementation of the ﬁrst

stage mentioned in subsection 3.1. SCID engine is

an implementation of the second stage described in

subsection 3.2. Browser is an interface between the

system and psychiatrists. Any popular web browser

can be used as the Browser.

Figures 2, 3 and 4 show snapshots of the system.

Figure 2 is a snapshot of the ﬁrst stage. A psychi-

atrist asks and observes symptoms of a patient and

checks in boxes of present symptoms. Figure 3 show

the result of Figure 2: a list of suspicious disorders.

When the psychiatrist click the “go” button in Figure

3, the second stage starts. Figure 4 is a snapshot of

the second stage. In the second stage, questions are

displayed on the screen in sequence. The psychiatrist

inputs answers for the questions by checking an ap-

propriate item, then the result ﬂows: the patient has

the suspicious disorder or not.

5 DISCUSSION

By constructing the system, we ﬁnd the following

problems in DSM and SCID.

1. Knowledge for diagnosis described in DSM or

SCID has ambiguity. In DSM or SCID, the am-

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

370

Figure 3: The suspicious disorders.

Figure 4: The second step.

biguity should be resolved by psychiatrists when

they make a diagnosis for a patient. That is a rea-

son why only psychiatrists should be permitted to

make a diagnosis using DSM or SCID.

2. Knowledge for diagnosis is still growing and

changing. DSM has been upgraded in three

times from the ﬁrst published year, 1952, the sec-

ond:1968, the third:1980 and the forth:1994. And

next upgrade is under contemplation.

3. Knowledge described in DSM and SCID is strict

and exhaustive. Therefore the most serious prob-

lem of DSM or SCID for making a diagnosis is

time-consuming to deliver.

For the problems, our system can provide a partial

answer.

• For 1, the system does not resolve the ambiguity,

because the ambiguity is involved the knowledge

we use.

• For 2, the system should be designed ﬂexibly for

the change of knowledge. We will try to solve this

problem.

• For 3, we design two stages diagnosis so that the

system is expected to be shorten time for diagno-

sis. Our next work will verify this point.

For the ﬁrst stage of the system, we extract symp-

toms from the description of DSM by hand. In future,

we would like to extract them by computer. There are

problems in DSM to extract symptoms by computer.

They are natural language processing problems.

• Too many symptoms are mentioned so that it is

hard to distinguish which symptoms are “main”

or “critical”.

• Sometimes the same or similar symptoms are de-

scribed different representations. Therefore, for

the ﬁrst stage, we have to reduce them to repre-

sentative symptoms.

At ﬁrst, we will check whether our manual procedures

work well in practical cases, then, consider how the

work is done automatically.

6 CONCLUSIONS

We propose a web-based system which assists psychi-

atrists in making a diagnosis. We design a knowledge

base by reconstructing knowledge organized in man-

ual, DSM and SCID. We divide a process of making

a diagnosis in two stages. In the ﬁrst stage, a psychia-

trist selects suspicious disorders from the list provided

by the system. In the second stage the psychiatrist

makes a diagnosis of the selected disorders. We also

discuss problems we ﬁnd out during our experience.

REFERENCES

American Psychiatric Association (2000). Diagnostic and

statistical manual of mental disorders fourth edition

text revision DSM-IV-TR. American Psychiatric Pub-

lishing, Inc.

First, M. B., Spitzer, R. L., Gibbon, M., and Williams, J.

B. W. (2002). User’s Guide for the structured clini-

cal interview for DSM-IV axis I disorder. American

Psychiatric Press Inc.

Phuong, N. H., Dang, J. H., and Prasad, N. R. (2000). De-

veloping a supporting expert system for lung-disease

diagnosis using fuzzy logic. Biomedical Soft Comput-

ing and Human Science, 5(2):37–44.

Starita, A., Majidi, D., Giordano, A., Battaglia, M., and

Cioni, R. (1995). A tutorial expert system for the di-

agnosis of neurogenic diseases of the lower limb. Ar-

tifcial Intelligence in Medicine, 7(1):25–36.

A DESIGN OF DIAGNOSIS SYSTEM FOR MENTAL DISORDER

371