COPERNICUS - AN EXPERT SYSTEM SUPPORTING
DIFFERENTIAL DIAGNOSIS OF PATIENTS EXAMINED USING
THE MMPI TEST
An Index-rule Approach
Jerzy Gomuła
The Andropause Institute, Medan Foundation, Warsaw, Poland
Cardinal Stefan Wyszyski University in Warsaw, Warsaw, Poland
Krzysztof Pancerz, Jarosław Szkoła
Institute of Biomedical Informatics, University of Information Technology and Management in Rzesz
´
ow, Rzesz
´
ow, Poland
Keywords:
Diagnostic decision-support system Copernicus, Mental disorders, MMPI, Logical decision rules, Differential
diagnosis.
Abstract:
In the paper, we present the Copernicus system - a computer tool supporting differential diagnosis of patients
with mental disorders. As an example, we discuss results for a sample of 479 women examined by means of
the MMPI-WISKAD personality inventory. There have been used the inductive classifiers based on the C4.5
decision tree algorithm. We have accomplished an overview of indexes generated on the basis of examined
sample by several systems: Eichmann’s, Diamond’s, Goldberg’s, Taulbee-Sisson’s, and Panda-APAP (mixed
with respect to indexes).
1 INTRODUCTION
One of the main tasks of building expert systems is
to search for efficient methods of classification of
new cases. Our research concerns psychometric data
coming from the Minnesota Multiphasic Personal-
ity Inventory (MMPI) test (Lachar, 1974). MMPI
is used to count the personality-psychometric dimen-
sions which help in diagnosis of mental diseases.
In years 1998-1999 a team of researchers consist-
ing of W. Duch, T. Kucharski, J. Gomuła, R. Adam-
czak created two independent rule systems devised
for the nosological diagnosis of persons that may be
screened with the MMPI-WISKAD test (Duch et al.,
1999). Testing some algorithms for the rule genera-
tion from the MMPI data was described in (Gomuła
et al., 2010a), (Gomuła et al., 2010c). Our research is
focused on creating a new computer tool for multicri-
teria diagnosis of mental diseases. The first version of
this tool has been presented in (Gomuła et al., 2010b).
The Minnesota Multiphasic Personality Inventory
(MMPI) test (Lachar, 1974) delivers psychometric
data on patients with selected mental disorders. Orig-
inally, the MMPI test was developed and published
in 1943 by a psychologist S.R. McKinley and a neu-
ropsychiatrist J.Ch. Hathaway from the University of
Minnesota. Later the inventory was adapted in above
fifty countries. The MMPI-WISKAD personality in-
ventory is the Polish adaptation of the American in-
ventory. It has been used, among other modern tools,
for carrying out nosological differential diagnosis.
MMPI is also commonly used in scientific research.
The test is based upon the empirical approach and
originally was translated by M. Chojnowski (as WIO)
(Choynowski, 1964) and elaborated by Z. Płu
˙
zek (as
WISKAD) in 1950 (Płu
˙
zek, 1971). American norms
were accepted there. On the basis of the received
responses (”Yes”, ”Cannot Say”, ”No”) to selected
questions we may obtain the reference and clinical
scales as being directly related to specific questions
(items) and recalculate the outcome into T-scores re-
sults. The T-scores ([T]) scale, which is traditionally
attributed to the MMPI, represents the following pa-
rameters: offset ranging from 0 to 100 T-scores, av-
erage equal to 50 T-scores, standard deviation equal
to 10 T-scores. The profile that is built for such a
case always has a fixed and invariable order of its con-
stituents as distributed on the scales. The validity part
323
Gomuła J., Pancerz K. and Szkoła J..
COPERNICUS - AN EXPERT SYSTEM SUPPORTING DIFFERENTIAL DIAGNOSIS OF PATIENTS EXAMINED USING THE MMPI TEST - An
Index-rule Approach.
DOI: 10.5220/0003172503230328
In Proceedings of the International Conference on Health Informatics (HEALTHINF-2011), pages 323-328
ISBN: 978-989-8425-34-8
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
Figure 1: A graphical user interface of Copernicus.
consists of three scales: L - the scale of lying which is
sensitive to all false statements aiming at representing
ourselves in a better light, F - the scale which detects
atypical and deviational answers to all items in the
test, K - it examines self defensive mechanisms and it
detects subtler attempts of the subject being screened
at falsifying and aggravation. The clinical part con-
sists of ten scales: 1. Hypochondriasis (H p), 2. De-
pression (D), 3. Hysteria (Hy), 4. Psychopathic Devi-
ate (Ps), 5. Masculinity/Femininity (Mk), 6. Paranoia
(Pa), 7. Psychastenia (Pt), 8. Schizophrenia (Sc), 9.
Hypomania (Ma), 0. Social introversion (It).
2 THE COPERNICUS SYSTEM - A
GENERAL OVIERVIEW
The Copernicus system is a computer tool for mul-
ticriteria differential diagnosis of patients with men-
tal disorders. The tool was designed for the Java
platform. This makes the tool a modern platform-
independent, object-oriented, user-friendly applica-
tion. A graphical user interface is presented in Fig-
ure 1. The current version of this tool offers sev-
eral main functions. We can locate patients in a pro-
file space using a wide variety of measures and in-
dexes (e.g., general distance measures, specialized
measures, psychopathology indexes). We can match
patient profiles to patterns of disorders using dendro-
grams generated by different clustering methods with
a suitable visualization. In a tool, two approaches
to generating dendrograms (Gan et al., 2007) are ap-
plied: standard hierarchical clustering techniques as
well as an approach based on the so-called Wroclaw
taxonomy. We can visualize patient profiles on the
background of patterns of disorders as well as de-
cision rules generated by popular data mining sys-
tems. An important thing is a unique visualization
of decision rules (in the form of stripes put on pro-
files) supporting the nosological diagnosis. We can
characterize patients by a number of indexes of the
MMPI systems (Dahlstrom et al., 1986), (Kucharski
and Gomuła, 1998b), (Kucharski and Gomuła, 1998a)
like Diamond’s, Leary’s, Goldberg’s, Eichmann’s,
Panda-APAP, and some other important and useful in-
dexes. In the case of the Diamond’s and Leary’s sys-
tems, indexes are arranged in the form of appropriate
diagrams (a box diagram for the Diamond’s system
and an radar-angular diagram for the Leary’s system)
which enable the user to interpret a patient profile in a
graphical way. The design and implementation of the
presented tool take into consideration its modularity.
Therefore, the tool can easily be extended to other in-
telligent methods used in data mining and analysis, as
well as, to other kinds of data, for example, coming
from another inventories.
3 INPUT AND OUTPUT DATA
We have examined 479 clinical cases (women).
A sample was selected using the competent judge
method (the majority of two-thirds of votes of three
clinicians-diagnosticians with several years’ experi-
ence in clinical diagnosis using the MMPI-WISKAD
test). Each case is classified to one of ve psychi-
atric nosological types: neurosis (neur), psychopathy
(psych), schizophrenia (schiz), simulation (simu), dis-
simulation (dissimu) as well as to norm (norm). Data
vectors can be represented in a graphical form as the
so-called MMPI profiles. The profile has always a
fixed and invariable order of its constituents (scales).
In the Copernicus system, various procedures
have been implemented, i.e., procedures plotting
scale and index profiles, box diagrams, radar-angular
diagrams as well as the basis of classification (deci-
sion) rules and their striped visualization decidedly
improving differential diagnosis; the basis of noso-
logical patterns for different genders (separately for
men, separately for women) and classification (de-
cision) rules also with respect to the examined gen-
der; superimposing rules on patterns; visualization of
classification functions. The classification functions
(Cios et al., 2007) are definitely better and more pre-
cise than the rough Goldberg’s indexes and they con-
cern not four but twenty nosological classes (also with
respect to gender of the examined patient). Ranges
of the Goldberg’s indexes, determined by us, confirm
their high diagnostic relevance also for the MMPI-
WISKAD data (examined women). Many clinicians
feel an attachment to these indexes. Therefore, we
have used an index-rule approach to the Goldberg’s
indexes which enables us to determine values of their
ranges and classification accuracy of selected classes.
In the Copernicus system, there have been im-
HEALTHINF 2011 - International Conference on Health Informatics
324
plemented numerous matching and distance measures
enabling the user to inspect a degree of matching and
consistency of a given profile with a nosological pat-
tern or norm (see (Gomuła et al., 2010b)). An exam-
ined profile consists of scales. There is possibility of
excluding any scales like in a clinical practice. The
profile can be extended by various indexes which are
calculated, added to the profile, and displayed in dif-
ferent ways (tabular collation, diagrams).
4 EXPERIMENTS
In our experiments, the Copernicus and WEKA sys-
tems have been used. WEKA is a collection of ma-
chine learning algorithms for data mining tasks (Wit-
ten and Frank, 2005). The main goal of experiments
is generation of efficient classification (decision) rules
via decision trees on the basis of profiles of patients
and selected indexes calculated for profiles. Indexes
added to profiles (scales) have been calculated using
the Copernicus system. Next, for decision tree gener-
ation, the well-known C4.5 algorithm (Quinlan, 1993)
(implemented in WEKA) has been used. To avoid ex-
cessive specificity (overfitting) of a decision tree and
to improve its prediction value the standard approach,
called pruning the decision tree, has been used. The
complete decision tree has been generated, and next
non-significant branches have been removed (the so-
called post-pruning). If a decision tree is overfitted,
then it captures erroneous classification information,
which will tend to make it perform badly when clas-
sifying unseen cases. In the subsections, we present
several experiments performed for profiles (scales)
and selected indexes.
4.1 Classification Rules only from
Profiles (Validity and Clinical
Scales)
There is some problem with the 5.(M f ) scale. A fea-
ture selection analysis using several methods imple-
mented in the WEKA system, for example, Signif-
icanceAttributeEval together with Ranker, Genetic-
Search together with CfsSubsetEval confirms this fact
(see Table 1). The SignificanceAttributeEval method
shows that scales: 8.(Sc), 6.(Pa), 4.(Ps), 3.(Hy),
0.(It), 7.(Pt), 1.(H p), F, and 2.(D) are very impor-
tant in the rank whereas scales K and L are the weak-
est ones and they need psychometric revision (which
happened in the MMPI-2 test). Scale 9 needs refin-
ing (also in view of specificity and difficulty of test
examination of maniacal states/episodes). For a deci-
Figure 2: Visualization of profiles and rules.
sion tree generated using the C4.5 algorithm, we ob-
tain that only ten rules are sufficient to make clini-
cal differential diagnosis (see Table 3(a)) in the scope
of ve most important nosological classes and norm
with accuracy 85% and greater. The average accuracy
is 88%. Arrows indicate lower and upper bounds of a
rule. Exemplary visualization of profiles and rules (in
the form of stripes put on profiles) is shown in Figure
2. If all conditions of a given rule cut across the pro-
file, then the rule is satisfied and diagnosis indicated
by a rule can be assigned to the profile.
Table 1: Attribute selection using (a) SignificanceAttribu-
teEval together with Ranker, (b) GeneticSearch together
with CfsSubsetEval.
a)
Avg merit Avg rank Attribute
0.959 ± 0.016 1 ± 0 8.(Sc)
0.9 ± 0.015 2.4 ± 0.92 6.(Pa)
0.878 ± 0.013 2.8 ± 0.4 4.(Ps)
0.853 ± 0.007 4.6 ± 0.92 3.(Hy)
0.854 ± 0.012 4.7 ± 1 0.(It)
0.841 ± 0.01 6.2 ± 0.87 7.(Pt)
0.84 ± 0.005 6.3 ± 0.64 1.(H p)
0.813 ± 0.008 8 ± 0 F
0.781 ± 0.012 9 ± 0 2.(D)
0.741 ± 0.028 10 ± 0 9.(Ma)
0.676 ± 0.008 11 ± 0 K
0.607 ± 0.023 12.1 ± 0.3 L
0.558 ± 0.013 12.9 ± 0.3 5.(Mk)
b)
Number Attribute
of
folds
(%)
100 L
100 F
100 K
100 1.(H p)
100 2.(D)
100 3.(Hy)
100 4.(Ps)
90 5.(Mk)
100 6.(Pa)
100 7.(Pt)
100 8.(Sc)
90 9.(Ma)
100 0.(It)
The 5.(Mk) scale was excepted. A set of rules ob-
tained from profiles after the transformation of a deci-
sion tree is shown in Table 3(a). Rules have the form
of logical implications: IF conjunction of conditions
(concerning values of scales) is satisfied, THEN deci-
sion (nosological class) should be taken with a given
accuracy (certainty), for example, IF L <= 58[T ]
AND 1.(H p) <= 57[T ], THEN R1 : norm (with ac-
curacy 89%). All classification (decision) rules from
our set (see all tables with rules) have a separable
(hence differential) character. Separability is assured
by the inequality test applied in nodes of a decision
COPERNICUS - AN EXPERT SYSTEM SUPPORTING DIFFERENTIAL DIAGNOSIS OF PATIENTS EXAMINED
USING THE MMPI TEST - An Index-rule Approach
325
tree: a <= v, where a is an attribute (scale, index),
v is a threshold for which the criterion of splitting is
maximized. While using rule-based differential diag-
nosis, narrative descriptions can be restricted to scales
and their ranges occurring in conditions of the satis-
fied rule. Such narrative based on satisfied and closed
decision rules can be called a narrative diagnosis of
the first level. As it is known, the appropriate clinical
descriptions correspond to individual scales. Rules
can be also closed by minimal or maximal value, re-
spectively, for a given scale in a given class (or for all
classes). An example is shown in Table 2. A rule is
closed by minimal and maximal values of scales for
the schiz class. There is some paradox. A process of
Table 2: An example of closing a rule.
1.(H p) 6.(Pa) 9.(Ma) Rule No.:class (accuracy %)
> 64 <= 77 > 56 R6:schiz (85%) - before closing
65 89 47 77 57 85 R6:schiz (85%) - after closing
closing conditions of rules can lead to smaller predic-
tion values of rules, but it can increase the narrow-
range accuracy of a narrative description.
4.2 Classification Rules from Profiles
and Eichmann’s Indexes
For each pair of different scales, the Eichmann’s in-
dexes are calculated as a sum, a difference, and an
arithmetic average of two scales, for example L F,
L + F,
LF
2
, ..., L 1.(H p), L + 1.(H p),
L1.(H p)
2
,
etc. Originally, the Eichman’s system included five
components (the so-called L components, K compo-
nents, F components, M f components, and NP com-
ponents). A component is a scale subtracted from
another scale. In our approach, we extend the Eich-
mann’s indexes to all twelve components (subtraction
of the same scales is omitted). All indexes were cal-
culated only in [T] for the whole sample of women.
Classification rules obtained from data consisting of
profiles and the Eichmann’s indexes are shown in Ta-
ble 3(b). The average accuracy is 88.9%. We can
assign clinical rational interpretation to the selected
Eichmann’s indexes, for example, D Ma is the ac-
tivity (Diamond’s) or mood disturbance index. The
Eichmann’s indexes indicated by a decision tree ques-
tion credibility of code type systems. Not only the
height of the highest scales (clinical, >= 70[T ] or
>= 65[T ]) in a profile constituting a given code type
is important, but also sums and differences between
selected scales. It concerns both validity (L, F, K)
and clinical scales. All obtained rules can be treated
as new code types with the determined accuracy.
Figure 3: A hierarchy of classes for Goldberg’s indexes.
Figure 4: An exemplary Leary’s diagram.
4.3 Classification Rules from Profiles
and Goldberg’s Indexes
The Copernicus system enables the user to calculate
three Goldberg’s indexes. The ranges of the Gold-
berg’s indexes have been determined by rules for each
of three levels (see Figure 3). Simulation (simu) and
dissimulation (dissimu) have been excluded from the
analysis. A macroclass of deviational profiles in-
cludes three classes: neurosis (neur), schizophrenia
(schiz), and psychopathy (psych). A macroclass of
psychiatric includes neurosis and schizophrenia. A
class of sociopathy is formed by its counterpart, i.e.,
psychopathy. The Goldberg’s indexes have an excel-
lent property differentiating at each level with accu-
racy at least 91%. Classification rules obtained from
data consisting of profiles and the Goldberg’s indexes
are shown in Table 4(a).
4.4 Classification Rules from Leary’s
Indexes
The Copernicus system enables the user to calculate
the Leary’s indexes (Leary, 1957). Indexes can also
be visualized as it is shown in Figure 4. For de-
cision rule generation, all of 497 profiles have been
used. For each profile, eight Leary’s indexes (called
styles) are calculated: (1) Managerial - Autocratic
style M A, (2) Responsible - HyperNormal style
R H, (3) Cooperative-Over - Conventional style
C C, (4) Docile - Dependent style D D, (5) Self-
Effacing - Masochistic style E M, (6) Rebellious
HEALTHINF 2011 - International Conference on Health Informatics
326
Table 3: (a) Rules obtained after transformation of a decision tree generated for all scales excluding scale 5, (b) Rules obtained
for profiles and Eichmann’s indexes, (c) Rules obtained for profiles and all indexes.
a)
L 1.(H p) 3.(Hy) 6.(Pa) 8.(Sc) 9.(Ma) 0.(It) Rule No.:class (accuracy %)
<= 58 <= 57 R1:norm (89%)
58 64 <= 59 <= 77 <= 68 R2:norm (89%)
> 64 <= 77 <= 56 R3:neur (85%)
58 64 > 59 <= 77 <= 68 R4:neur (85%)
> 58 <= 57 > 58 R5:psych (92%)
> 64 <= 77 > 56 R6:schiz (85%)
58 64 <= 77 > 68 R7:schiz (85%)
> 58 <= 57 <= 58 > 59 R8:simu (94%)
> 57 > 77 R9:simu (94%)
> 58 <= 57 <= 58 <= 59 R10:dissimu(85%)
b)
Hy + It L + Ps F D Hy Pt Ma F + Ps H y Ma H p + D L + Mk D Ma K Pt Rule No.:class (accuracy %)
> 119 <= 157 <= 13 <= 123 > 117 R1:simu (94.4%)
> 119 <= 157 <= 13 <= 123 <= 117 R2:norm (88%)
<= 119 <= 125 <= 58 R3:dissimu (70%)
<= 119 <= 125 > 58 R4:norm (88%)
<= 119 > 125 <= 5 R5:dissimu (70%)
<= 119 > 125 > 5 <= 11 R6:psych (94%)
<= 119 > 125 > 5 <= 11 R7:schiz (85.1%)
> 119 > 125 > 20 > 157 R8:neur (89.1%)
> 119 <= 20 > 157 R9:simu (94.4%)
> 119 <= 157 > 13 R10:neur (89.1%)
> 119 <= 157 <= 13 > 123 > 8 > 27 R11:schiz (85.1%)
> 119 <= 157 <= 13 > 123 > 8 <= 27 R12:neur (89.1%)
c)
SocAgg/ avgDrInt 1.M A avgHost SocInt 0.(It) Rule No.:class (accuracy %)
<= 1 <= 67 <= 5 > 12 R1:norm (96.0%)
> 1 > 67 <= 73 <= 12 R2:neur (93.1%)
<= 1 > 67 <= 73 R3:schiz 93.1%)
<= 1 <= 67 <= 5 R4:psych (96.8%)
<= 1 > 67 <= 5 > 73 R5:simu 87.9%)
<= 1 <= 67 > 5 > 55 R6:simu 87.9%)
<= 1 <= 67 > 5 <= 55 R7:dissimu (95.0%)
- Distrustful style R D, (7) Aggressive - Sadistic
style AS, (8) Competitive - Narcissistic style C N.
After joining a profile (validity and clinical part ex-
cluding the 5.(Mk) scale) with the Leary’s indexes
and next generating decision rules, we came with the
following conclusion: the best clinical differentiation
is for styles 1.M A, 4.D D, and 7.A S. Two
Leary’s indexes determining two basic personality di-
mensions are not covered by a classifier and they are
not used in classification. Clinical scale 4.(Ps) rein-
forces the accuracy of differential diagnosis of psy-
chopathy to 60% and the average accuracy for the
whole classifier to 90%. Especially, this scale dif-
ferentiates well between norm and psychopathy for
women. Classification rules obtained from data con-
sisting of profiles and the Leary’s indexes are shown
in Table 4(b).
4.5 Classification Rules from Profiles
and All Indexes
In the last experiment, rules have been induced for
all system indexes (Diamond’s, Leary’s, Goldberg’s,
and PANDA-APAP (Pancheri et al., 1992)) and addi-
tional indexes important diagnostically. The PANDA
system includes also the Taulbee-Sisson’s system as
a subsystem. The Taulbee-Sisson’s system differenti-
ates between neuroticism and psychoticism profiles.
Table 4: (a) Rules obtained for profiles and Goldberg’s in-
dexes, (b) Rules obtained for Leary’s indexes (without va-
lidity and clinical scales).
a)
G1 G2 G3 Rule No.:class (accuracy %)
<= 137 <= 63 R1:norm (94.0%)
> 120 > 63 R2:psych (100.0%)
> 137 <= 63 R3:neur (85.1%)
<= 120 > 63 R4:schiz (94.1%)
b)
1.M A 4.D D 7.A S Rule No.:class (accuracy %)
<= 5 <= 192 <= 184 R1:norm (88%)
<= 42 > 192 <= 214 R2:neur (85%)
<= 5 <= 192 > 184 R3:psych (60%)
> 42 > 192 R4:schiz (87%)
> 192 > 214 R5:simu (95%)
> 5 <= 192 R6:dissimu (95%)
Copernicus counts the conditional Taulbee-Sisson’s
indexes and next indicates neuroticism (if at least thir-
teen indexes are satisfied), psychoticism (if at most
six indexes are satisfied), or an undefined state (oth-
erwise). Rules have also been generated for profiles
and all listed indexes. Classification rules obtained
from data consisting of profiles and all listed indexes
are shown in Table 3(c). The meaning of indexes is
as follows: SocAgg - social aggression (Diamond’s
system), avgDrInt - average dreams intellectualiza-
tion (Diamond’s system), avgHost - average hostil-
ity (Diamond’s system), SocInt - social introversion
(PANDA system). In this experiment, the strongly ex-
tended Eichmann’s system has been omitted. Its rule
analysis was performed in Subsection 4.2.
COPERNICUS - AN EXPERT SYSTEM SUPPORTING DIFFERENTIAL DIAGNOSIS OF PATIENTS EXAMINED
USING THE MMPI TEST - An Index-rule Approach
327
4.6 Summarry
Rule classification trees enable the user to reduce in-
terdependent, superfluous and strongly outlying data.
Moreover, rule trees obtained by us have the ”white-
box” character, i.e., the knowledge representation is
overt, readable and useful for diagnosticians and clin-
icians. Rule knowledge-based systems are small and
do not exceed fifteen classification (decision) rules.
The greatest number (twelve) of rules was generated
for the Eichmann’s system. All our analyses demon-
strate that in an index-rule approach the selected in-
dexes, Eichman’s (calculated in [T]), extended Di-
amond’s, Leary’s (without two basic personality di-
mensions) are used not only for supporting diagnosis
in the so-called interpretation quiet (mostly the range
40 - 60 [T]) or only for delivering additional inter-
pretation (descriptive) hypotheses. Indexes selected
by us and additionally transformed into the index-rule
form give the high accuracy (above 80%). Therefore,
they can not be overlooked. This is their new clini-
cal and psychometric application. The systems men-
tioned earlier are implemented in Copernicus and they
support quantitative and clinical differential diagnosis
based on an index-rule approach.
5 CONCLUSIONS AND FURTHER
WORK
In the paper we have shown an index-rule-based ap-
proach to differential diagnosis of patients with men-
tal disorders. In our experiments we have used the
Copernicus system and the WEKA system. The
Copernicus system is a tool created by our team.
Among others, it enables the user to calculate a wide
variety of indexes used in diagnosis of psychopatholo-
gies. Some of them can also be visualized by means
of appropriate diagrams. WEKA is a collection of
machine learning algorithms for data mining tasks.
Our experiments deliver important information for
diagnosticians and clinicians about individual scales
(their validities and utilities) used in the MMPI test-
ing of patients. In the future work, among others, we
are going to incorporate rule generation algorithms
and visualization modules into the Copernicus sys-
tem. Our main goal is to deliver to diagnosticians and
clinicians an integrated tool supporting the compre-
hensive diagnosis of patients. The Copernicus system
is flexible and it is dedicated for supporting differen-
tial diagnosis of profiles of patients examined using
multiphasic personality inventories.
REFERENCES
Choynowski, M. (1964). Multiphasic Personality Inventory
(in Polish). Psychometry Laboratory, Polish Academy
of Sciences, Warsaw.
Cios, K., Pedrycz, W., Swiniarski, R., and Kurgan, L.
(2007). Data Mining. A Knowledge Discovery Ap-
proach. Springer.
Dahlstrom, W., Welsh, G., and Dahlstrom, L. (1986). An
MMPI Handbook, vol. 1-2. University of Minnesota
Press, Minneapolis.
Duch, W., Kucharski, T., Gomuła, J., and Adamczak, R.
(1999). Machine learning methods in analysis of psy-
chometric data. Application to Multiphasic Personal-
ity Inventory MMPI-WISKAD (in Polish). Toru
´
n.
Gan, G., Ma, C., and Wu, J. (2007). Data Clustering. The-
ory, Algorithms, and Applications. SIAM, Philadel-
phia, ASA Alexandria, VA.
Gomuła, J., Paja, W., Pancerz, K., and Szkoła (2010a).
A preliminary attempt to rules generation for mental
disorders. In Pardela, T. and Wilamowski, B., edi-
tors, Proc. of the HSI’2010, pages 763–765, Rzesz
´
ow,
Poland.
Gomuła, J., Pancerz, K., and Szkoła, J. (2010b). Analysis
of MMPI profiles of patients with mental disorders -
the first unveil af a new computer tool. In Grzech,
A.,
´
Swia¸tek, P., and Brzostowski, K., editors, Appli-
cations of Systems Science, pages 297–306. Academic
Publishing House EXIT, Warsaw, Poland.
Gomuła, J., Pancerz, K., and Szkoła, J. (2010c). Classifi-
cation of MMPI profiles of patients with mental disor-
ders - experiments with attribute reduction and exten-
sion. In Yu, J. et al., editors, Rough Sets and Knowl-
edge Technology, volume 6401 of LNAI, pages 411–
418. Springer-Verlag, Berlin Heidelberg.
Kucharski, T. and Gomuła, J. (1998a). Introduction to the
MMPI-2 questionnaire (in Polish). Toru
´
n.
Kucharski, T. and Gomuła, J. (1998b). Introduction to the
MMPI-WISKAD questionnaire (in Polish). Toru
´
n.
Lachar, D. (1974). The MMPI: Clinical assessment and au-
tomated interpretations. Western Psychological Ser-
vices, Fate Angeles.
Leary, T. (1957). Interpersonal diagnosis of personality:
a functional theory and methodology for personality
evaluation. The Ronald Press Company, New York.
Pancheri, P., Biondi, M., Portuesi, G., and P.L., M. (1992).
Psicodiagnostica ed evoluzione nosografica: la ver-
sione dsm iii del sistema mmpi-apap. Rivista di
Psichiatria, 1:7–22.
Płu
˙
zek, Z. (1971). Value of the WISKAD-MMPI test for
nosological differential diagnosis (in Polish). The
Catholic University of Lublin.
Quinlan, J. (1993). C4.5. Programs for machine learning.
Morgan Kaufmann Publishers.
Witten, I. H. and Frank, E. (2005). Data Mining: Practi-
cal Machine Learning Tools and Techniques. Morgan
Kaufmann.
HEALTHINF 2011 - International Conference on Health Informatics
328