=Paper=
{{Paper
|id=Vol-2648/paper22
|storemode=property
|title=Model of Integral Evaluation of Expert Knowledge for the Diagnosis of Lysosomal Storage Diseases
|pdfUrl=https://ceur-ws.org/Vol-2648/paper22.pdf
|volume=Vol-2648
|authors=Nikolay A. Blagosklonov,Boris A. Kobrinskii
}}
==Model of Integral Evaluation of Expert Knowledge for the Diagnosis of Lysosomal Storage Diseases==
https://ceur-ws.org/Vol-2648/paper22.pdf
Model of integral evaluation of expert knowledge for
the diagnosis of lysosomal storage diseases
Nikolay A. Blagosklonova , Boris A. Kobrinskiia
a
Federal Research Center “Computer Science and Control” of RAS, Moscow, Russia
Abstract
This article proposes an approach to a comprehensive assessment of expert knowledge with using the
model. Implemented the ability to account for a fuzzy and incomplete clinical picture of diseases. Based
on the hypotheses, differential diagnostic series and comparison of reference models with personal
models of new cases are formed, that allows to rate the degree of similarity and identify the disease. A
comparative analysis of diagnostic hypotheses was carried out using special algorithms. The study was
carried out on the mucopolysaccharidoses as an example, which belong to the class of orphan inherited
lysosomal diseases.
1. Introduction
The diagnostics problem of monogenic lysosomal storage diseases that belong to the class of
orphan (rare) diseases is a global challenge [1]. The earliest possible detection of this pathology
is crucial to development severe changes leading to disability and death. Meanwhile, insuffi-
cient level of knowledge in the medical community in regards of the clinical signs of these
diseases is often the cause of late identification of diseases [2].
In the present study mucopolysaccharidoses, which are related to lysosomal diseases, and
include 15 clinical forms, are considered as an example.
An essential feature of lysosomal diseases is the fuzzy of the symptom’s degree of expres-
sion in the clinical manifestations. It is determined by the progressive accumulation of macro-
molecules due to a deficiency of specific enzymes to cause of gene defects. Signs of the disease
change progressively with age. Manifestations may be already present in the first year of life.
This explains the focus of computer diagnostic systems on early childhood. However, the weak-
ness of previously created and currently used systems is the problem of early detection of the
first signs of pathology, and underestimation of the dynamics of changes with children’s age
[3].
The comprehensive evaluation of the symptoms of diseases in terms of the onset, degree of
expression and frequency of occurrence of signs presents a great interest. However, the mani-
festations of the disease may be incomplete. With this in mind, the attention in the creation of
Russian Advances in Artificial Intelligence: selected contributions to the Russian Conference on Artificial intelligence
(RCAI 2020), October 10-16, 2020, Moscow, Russia
" nblagosklonov@gmail.com (N.A. Blagosklonov); kba_05@mail.ru (B.A. Kobrinskii)
© 2020 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org)
diagnostic decision support systems was focused on the assessments of the manifestation pe-
riod and degree of expression of clinical signs, made by the experts. It is necessary to consider
the character of the observed changes in the process of identifying diseases. Besides, the com-
plexity of differential diagnosis is determined by the similarity of the clinical manifestations of
different lysosomal diseases, in particular of the various types of mucopolysaccharidoses.
The difficulties in forming the knowledge base of the system are determined by the con-
tinuum of transitional states, variety of symptoms and a fuzzy clinical picture of lysosomal
diseases [4, 5].
2. Problem statement
In clinical practice, physicians in the diagnosis of lysosomal diseases, in particular mucopolysac-
charidoses, prior to conducting molecular genetic studies or enzyme analysis, focus on the
formation of a differential series of 3 to 5 nosological forms.
At creating an expert system, a number of stages have been identified.
At the first stage, knowledge about the clinical manifestations of diseases were sequentially
extracted from the world literature and from experts. The cognitologist synthesized the data
presented in the medical literature, and the experts specified the characteristic signs for each
from diseases and supplemented them with expert evaluations. Manifestation and degree of
expression of signs certainty factors [6] and the coefficient of modality were used to assess
the symptoms of diseases. Row scales were developed for: (a) age intervals, (b) periods of
manifestation, (c) fuzzy of signs, and (d) modality of signs.
At the second stage, “diseases – signs” matrix was formed, which included expert evaluations
for each form of the disease by age group.
At the third stage, a complex evaluation of each symptom and an integrated assessment of
the disease, which can be considered a model of diseases, were carried out.
At the fourth stage, six special algorithms for comparative differential diagnosis were devel-
oped, tested on a group of mucopolysaccharidoses. This made it possible to provide a choice
of the sequence of actions when recognizing a new case.
At the fifth stage, the “disease - signs” matrix was formed, an integrated presentation of the
clinical forms of diseases and algorithms for comparative diagnostics, which formed the basis
for the formation of the knowledge base.
The sixth stage is focused on differential diagnosis and proposing a number of hypotheses
about the disease in the patient.
Schematic representation of the steps is shown in Figure 1.
3. Presentation of expert knowledge
A textological card was created to extract the signs characterizing the clinical picture of diseases
from the literature. Great attention was given to the modality of the signs, and the selection of
signs that did not require the use of complicated methods for primary diagnosis was carried out.
At expert evaluations took into account the fuzzy of the manifestations of those parameters.
Figure 1: Generalized scheme for creating an artificial intelligence system for diagnosing orphan dis-
eases
At first, all signs found in literary sources were included in the textological cards by a cog-
nitologist. Then diagnostically insignificant characteristics were excluded, and the modality of
relevant characteristics was evaluated, with the participation of two experts.
For the group of mucopolysaccharidoses, 22 parameters were selected. Each symptom was
accompanied by a coefficient of modality, and two certainty factors – for manifestation and for
degree of expression at each from the four age groups.
The scale correction for the expression of signs was adjusted at the results of testing the
artificial intelligence system for the differential diagnosis of mucopolysaccharidoses [7].
3.1. Age groups
The physiological characteristics of the development of children and the progressive course of
mucopolysaccharidoses division age groups. This allowed more efficient differential diagnosis.
In the course of the study, four previously formed age groups of children and adolescents were
modified [8]. Table 1 presents age periods and interval scales, allowing to take into account
the temporal fuzzy of the manifestation of signs.
In medicine, it is customary to formally consider a person (child) to have not reached the next
age before the date of birth (for example, at 2 years 11 months 29 days a child is still considered
a two-year-old). The “Intervals” column in Table 1 is presented as decimal numbers. This is
necessary in order to avoid erroneous estimates, for 2 years and 6 months is not synonymous
with 2.6 years.
Table 1
Age groups
Age groups Intervals Possible extension of the age interval
Period up to 1 year 0 – 0.99 0 – 1.25
Period 1-3 years 1 – 3.99 0.75 – 4.25
Period 4-6 years 4 – 6.99 3.75 – 7.25
7 years and older 7 – 17.99 6.75 – 17.99
The column “Possible extension of the age intervals” allows to take into account the fact that
a sign can manifest at the time other than indicated in the first column, for example not at 4
years, but at 3 years 10 months. This corresponds to the fuzzy of age-related manifestations
of signs. For this, an age “corridor” is used, during which signs of the nearest age group can
be considered. Based on the opinion of experts the reserve was determined to be 1/4 year (3
months), since the manifestations in most patients are covered by such an extension of the age
intervals.
3.2. Modality coefficients of signs
Modality involves evaluating a concept from a certain point of view [9]. In this study, this is an
assessment of the diagnostic significance of the symptom. In this case, we consider modality
as an appreciation of the diagnostic significance of the sign. Modality is characterized by a
coefficient obtained from experts, taking into account data from various sources, including the
frequency of occurrence of the signs in age groups in publications and international databases
Genetic and Rare Diseases Information Center - GARD (https://rarediseases.info.nih.gov/) and
Human Phenotype Ontology - HPO (https://hpo.jax.org/app/).
Ultimately, modality (M) determines the level of relevance of clinical features on the scale,
which is characterized by three gradations. Modality evaluations were introduced in corre-
spondence to the diagnostic role of signs:
• for main symptoms – 5;
• for necessary symptoms – 4;
• for secondary symptoms – 2.
This coefficient allows to determine more accurately the contribution of each individual
symptom to the clinical picture of the disease. In cases where the attribute is absent due to
the age the concept of modality cannot be used. In these cases the coefficient is replaced by 0.
Modality coefficients M1 , M2 , M3 , M4 are formally labelled in accordance with the presence of
the four age groups mentioned above.
3.3. Certainty factors for the manifestation of signs
The period of development of the clinical signs of the disease is called the manifestation of the
disease. The significance of the manifestation in the diagnostic process is determined by the
fact that due to the individual characteristics of the organism, the character of the development
of clinical picture in patients is different. The manifestation and degree of expression of signs
in monogenic diseases depends on the level of specific ferment deficiency and gene penetrance.
Thus, the difference in the period of manifestation (m𝑖 ) is assessed by a confidence measure
(certainty factor), that reflects the level of confidence of the experts that the symptom appears
at a given age (age group). A scale in the interval [-1; 1], set as a symmetric function [10], was
used to assess the certainty factor for the manifestation of signs:
• The value “-1” characterizes the impossibility of the manifestation of this trait in the
specific age group for physiological reasons.
• The value “0” characterizes the norm, either the absence of a sign due to the features of
the course of the disease, or the option when the sign has already manifested (in previous
age groups).
• Value in the range [0.1; 1] characterizes a confidence measure of experts’ that a sign will
appear (manifest) in a given age period.
It should be noted that values other than “-1” in total cannot exceed “1” for four age groups.
Evaluations of m𝑖 from previous age periods m1 , m2 , m3 , m4 complement each other (except
for negative values), which allows to obtain a total manifestation certainty factor (m𝑘 ):
4
𝑚𝑘 = ∑ 𝑚𝑖 , 𝑓 𝑜𝑟 𝑚𝑖 ≥ 0 (1)
𝑖=1
Thus, the certainty factors for the manifestation allow to indicate the onset of the develop-
ment of the sign at a certain age, and the total score for age groups reflects the manifestation
at a certain age.
3.4. Certainty factors for the degree of expression of signs
With the progression of the lysosomal storage diseases, an important characteristic of the signs
that makes up the clinical picture is the degree of their expression. This is a trend to change at
various stages of the development of the pathological process.
Thus, the assessment of expression (s) characterizes the confidence of experts in the force of
manifestation of this sign in a specific age group. Accordingly, evaluations are given for each
age period: s1 , s2 , s3 , s4 .
Unlike the manifestation scale [-1; 1], the scale for assessing the degree of expression of the
sign was determined in the interval [0; 10], where “0” corresponds with the situation when
the symptom is absent in the patient, and “10” corresponds with the maximal manifestation
of the symptom. Changes in degree of expression by age indicate indirectly the rate of the
development of the symptoms.
Scale [0; 10] was proposed to obtain mathematically correct estimates when integrating the
results on three scales.
pos=[ht]
Table 2
Form of the matrix “diseases – signs”
Disease name
Sign name up to 1 year 1-3 years 4-6 years 7 years and older
M𝑖 m𝑖 s𝑖 M𝑖 m𝑖 s𝑖 M𝑖 m𝑖 s𝑖 M𝑖 m𝑖 s𝑖
Sign 1 M1 m1 s1 M2 m2 s2 M3 m3 s3 M4 m4 s4
Sign 2 M1 m1 s1 M2 m2 s2 M3 m3 s3 M4 m4 s4
3.5. Formation of the matrix “diseases – signs”
The form of the matrix “diseases – signs” is displayed in Table 2, in which in the columns
are the diseases for four age groups, with indication of coefficients of modality and certainty
factors of manifestation and degree of expression of signs for each age group. Signs of disease
are arranged line by line.
Appreciations for each clinical form of mucopolysaccharidosis were evaluated by experts
when comparing clinical forms differing in the degree of expression This is advisable in order
to achieve more reliable description in connection with the similarity and fuzzy of symptoms
in the clinical picture of related forms of the disease.
An example of the matrix is displayed in Table 3, filled down for the type VII mucopolysac-
charidosis (Sly syndrome).
Table 3
Matrix “diseases – signs” mucopolysaccharidosis type VII (fragment)
Mucopolysaccharidosis type VII (Sly syndrome)
Sign name up to 1 year 1-3 years 4-6 years 7 years and older
M𝑖 m𝑖 s𝑖 M𝑖 m𝑖 s𝑖 M𝑖 m𝑖 s𝑖 M𝑖 m𝑖 s𝑖
Growth inhibition 2 0.2 1 2 0.1 4 2 0.2 5 2 0.2 7
Corneal clouding
(determined without 0 -1 0 4 0.3 2 4 0.1 4 4 0.1 7
using a slit lamp)
4. System of signs evaluation
4.1. Complex evaluation of signs
An operation of multiplying expert evaluations of the manifestation, degree of expression and
modality of symptoms in a certain age group was used to obtain a complex evaluation of signs
(P𝑖 ). It resulted in the following formula:
𝑃 𝑖 = 𝑀𝑖 ⋅ 𝑚 𝑖 ⋅ 𝑠𝑖 , (2)
where:
P𝑖 – sign (symptom),
M𝑖 –modality of the sign, characterizing its frequency,
m𝑖 – certainty factor for the manifestation of the sign,
s𝑖 – certainty factor for the degree of expression of the sign.
An array of P𝑖 estimates was obtained by the calculation of all parameters of the disease
according to this formula.
This complex evaluations may require revision or clarification only in the course of obtaining
new data on the clinical forms of diseases.
4.2. Integrated evaluation of disease
The next stage was implemented to obtain a summary index characterizing the clinical picture
of the disease as a whole, considering the previously obtained complex evaluations of signs
(P𝑖 ). This aggregate of clinical manifestations was called an integrated evaluation (I ) of the
disease for each age group:
𝑛
𝐼 = ∑ 𝑃𝑖 , (3)
𝑖=1
where:
I – integrated evaluation of the signs of the disease,
P𝑖 – sign,
i – the number of signs,
n – the aggregate of signs of the disease (group of diseases).
This formula was used in two variations: one for a total score for all the signs that may
occur with a particular disease in an age group, and another for a score for signs that occur in
a particular case (in a particular patient).
In the first variation, this formula was called the reference integrated disease evaluation
(I𝑒 ), in this case, for each clinical form of mucopolysaccharidosis. The basis for the reference
evaluation is a set of signs that was determined at the stage of problem statement for creating
an intelligent diagnostic system.
In the second variation, when patients may have a different incomplete set of signs, the
concept of a personal integrated evaluation (I𝑝 ) was introduced. Thus, the identical formula I
was used in calculating both I𝑒 and I𝑝 . However, the number of signs n differ: when calculating
I𝑒 , it is 22 for mucopolysaccharidoses, and for I𝑝 it can exist in the range from 1 to 22, depending
on the presence of manifestations in the patient.
Thus, in contrast to the semantic similarity metrics for measuring the phenotypic similarity
between a new patient and the base of hereditary diseases annotated using HPO [11], we in
the intelligent diagnostic system correlate a personal integrated assessment with a reference
disease assessment based on the developed formula.
5. Diagnostic hypotheses based on model comparisons
For 15 clinical forms of mucopolysaccharidoses, the use of the model allowed, taking into ac-
count four age groups, the formation of 60 integrated reference estimates of I𝑒 . The personal
I𝑝 of a particular case is calculated according to the available signs for each clinical form, and
a comparison is made with 15 I𝑒 of the same age period as the patient.
A differential diagnostic series of hypotheses, including several forms of mucopolysacchari-
doses from the set I𝑒 is formed at this step of the algorithm.
Let’s have a look at an example of using the integrated evaluation model for a 3 months
old patient (description taken from publication [12]). The patient had the following signs:
coarse facial features, kyphoscoliosis, stiffness of large joints, hepatomegaly, splenomegaly,
cardiopathy. The result of calculations of I𝑝 and the corresponding values of I𝑒 is presented in
Table 4.
Table 4
Calculation according to the clinical case
The clinical form of mucopolysaccharidosis I𝑝 I𝑒
Hurler syndrome 38.0 93.2
Hurler-Scheie syndrome 2.5 19.6
Scheie syndrome 0.0 0.4
Hunter syndrome (severe course) 1.8 14.1
Hunter syndrome (mild course) 0.0 0.6
Sanfilippo syndrome A 0.0 1.6
Sanfilippo syndrome B 0.0 1.6
Sanfilippo syndrome C 0.0 0.0
Sanfilippo syndrome D 0.0 0.0
Morquio syndrome A (rapidly progressing) 0.0 0.8
Morquio syndrome A (slowly progressing) 0.0 0.0
Morquio syndrome B 0.0 0.0
Maroteaux-Lami syndrome (rapidly progressing) 0.0 15.6
Maroteaux-Lami syndrome (slowly progressing) 0.0 0.4
Sly syndrome 3.0 9.5
Despite the patient having an atypical picture of the disease (as a rule, signs of stiffness of
large joints and kyphoscoliosis do not occur at this age), the greatest similarity among the
differentiated diagnoses was found with type IH mucopolysaccharidosis (Hurler’s syndrome).
This corresponds with a diagnosis verified by genetic testing.
However, not in all cases, such a direct method of comparison by relevant features is effective.
In the medical practice there are patients whose to advance a hypothesis about the disease can
only on the basis of signs that are secondary to one or a number of diseases but are of great
importance for other differentiable forms. In such clinical cases, false positive hypotheses are
possible, when the system may put forward an assumption that is not relevant to this case.
In this regard, differential diagnostic algorithms were proposed and tested, taking into ac-
count the diagnostic significance of the signs for each disease.
6. The approach for choice of differential diagnosis algorithm
It is possible to use different approaches to the practical implementation of the differential
diagnosis of orphan diseases. In the course of the study, variants of algorithms were considered,
providing for a comparison of the calculated case estimates only among themselves (absolute
values), and a comparison in similarity with reference evaluations (relative values), which could
either take into account or not take into account the modalities of the signs. As an experiment,
options for diagnostic solutions were considered with exclusion of expert certainty factors of
signs. This made it possible to obtain comparative diagnostic results under the conditions of
using the developed model and in a simplified version by taking into account the patient’s signs
confirming or excluding a possible diagnosis.
6.1. Differential diagnostic algorithms
In the study developed six different algorithms below, which were then tested to identify the
positive and negative aspects of their application.
6.1.1. Algorithm one
For each disease, the sum of complex estimates of the signs of P𝑖 , called I𝑝 , is calculated (inde-
pendence of their modality). Then the hypotheses were ranked by the sum of I𝑝 from greater
to lesser. Thus an ordered set of hypotheses is formed.
6.1.2. Algorithm two
For each disease, the sum of complex estimates of the signs of P𝑖 , that is, I𝑝 (independence of
their modality), is calculated. Further, in contrast to algorithm one, the percentage of coinci-
dence of I𝑝 of a particular case with a reference integrated assessment of I𝑒 of a disease in a
given age group is calculated. Then the hypotheses are ranked by the percentage of coincidence
from larger to smaller.
6.1.3. Algorithm three
For each disease, integral estimates of I𝑝 are calculated taking into account the modality of the
signs. Only signs with modalities “main” and “necessary” are taken into account. Then there
is a ranking of hypotheses by the sum of I𝑝 from greater to lesser.
6.1.4. Algorithm four
For each disease, integral estimates of the case of I𝑝 are calculated taking into account the
modality of the signs. Only attributes with modalities “main” and “necessary” are taken into
account. In contrast to algorithm three, the percentage of coincidence of the patient’s attributes
I𝑝 with the reference integrated assessment I𝑒 is calculated for signs with the “main” and “nec-
essary” modalities in this age group. Hypotheses are ranked by the percentage of coincidence
from larger to smaller.
6.1.5. Algorithm five
For each disease, the number of signs of any modality “for” and “against” is calculated. If the
number of signs denying a hypothesis exceeds a predetermined threshold, then the hypothesis
is rejected. The remaining hypotheses are ranked by the number of signs “for”.
6.1.6. Algorithm six
For each disease, the number of signs of any modality “for” and “against” is calculated. The
percentage of each group is calculated relative to the total number of patient symptoms. If
signs denying a hypothesis exceed a certain threshold relative to the total number of patient
attributes, then such a hypothesis is not considered. The remaining hypotheses are ranked
taking into account the percentage of signs “for” from the total number of signs of the patient.
6.2. Algorithm’s testing
Selection of the optimal differential diagnostic algorithm based on the clinical picture of 20
patients from different countries published in journal articles [13, 14, 15, 12, 16, 17, 18, 19,
20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]. Signs of patients with orphan diseases (including
age and gender) were included in the database. Then, an analysis was made of the results
for each algorithm with the allocation of the first 5 hypotheses in each case. If the clinical
form of the disease was confirmed by genetic testing in the article and was present among
the selected hypotheses, then differential diagnosis using a specific algorithm was considered
successful. The experimental results are presented in Table 5. In this test of the fifth algorithm,
the threshold for rejecting hypotheses was set with the number of signs “against” 3 or more, and
for the sixth algorithm more than 33% of the total number of patient symptoms. Both thresholds
are quite soft, but the initial testing was decided to be carried out under such conditions. In
the case of insufficient effectiveness of these thresholds, there was the possibility of a further
tightening of the selection criteria.
As can be seen from Table 5, the best result was shown for differential diagnostics using the
third algorithm for ranking hypotheses by the integral assessment I𝑝 taking into account the
modalities of the signs.
Next, an experiment was conducted to increase the accuracy of diagnostics by a pairwise
combination of algorithms. According to its results, the most promising was the approach with
preliminary exclusion of the hypothesis with the greatest number of “against” signs according
to the fifth algorithm, and among the remaining hypotheses, ranking according to one of the
first four algorithms.
For algorithm five, at a given threshold for rejecting hypotheses, the correct diagnosis was
rejected in 4 cases out of 20. That is, in 20% of cases the correct diagnosis was not made. In this
regard, a number of additional tests were carried out in order to clarify the “against” threshold
for algorithm five using a dynamic approach.
Table 5
Algorithm’s testing results
A confirmed diagnosis among
the first 5 hypotheses based on
Patient number Confirmed diagnosis
the results of the algorithm
1 2 3 4 5 6
1 [13] Hunter syndrome Yes Yes Yes No Yes Yes
2 [14] Maroteaux-Lami syndrome Yes Yes Yes No Yes Yes
3 [15] Maroteaux-Lami syndrome Yes Yes Yes Yes Yes Yes
4 [12] Hurler syndrome Yes Yes Yes Yes No No
5 [16] Hurler syndrome Yes Yes Yes Yes Yes Yes
6 [17] Sanfilippo syndrome C No Yes No Yes Yes Yes
7 [18] Sanfilippo syndrome B Yes Yes Yes Yes No No
8 [19] Maroteaux-Lami syndrome Yes Yes Yes Yes No No
9 [20] Sly syndrome Yes Yes Yes Yes No No
10 [21] Sly syndrome Yes No Yes No Yes Yes
11 [22] Sly syndrome Yes Yes Yes Yes Yes Yes
12 [23] Maroteaux-Lami syndrome Yes No Yes No Yes Yes
13 [24] Hunter syndrome Yes No Yes No Yes Yes
14 [25] Hurler-Scheie syndrome Yes No Yes Yes Yes No
15 [26] Sly syndrome Yes Yes Yes Yes Yes Yes
16 [27] Morquio syndrome A Yes Yes Yes Yes Yes Yes
17 [28] Morquio syndrome A No Yes Yes Yes Yes Yes
18 [29] Morquio syndrome B No Yes No Yes Yes Yes
19 [30] Sanfilippo syndrome A No Yes No Yes Yes Yes
20 [31] Hurler syndrome Yes No Yes No Yes No
Total 16 15 17 14 16 14
This approach was as follows: according to algorithm five, the number of diagnoses in the
differential series was calculated, which had 0 signs “against”. If the number of hypotheses is
less than 5, then the threshold for signs “against” is increased by one to expand the differential
series. This process continued until the selection of 5 or more hypotheses.
Next, ranking was carried out according to the first to fourth algorithms. The results of this
experiment testing algorithms are presented in Table 6.
As can be seen from Table 6, the best result is obtained according to algorithm two, when
ranking is carried out according to the percentage of coincidence I𝑝 of a particular case with a
reference estimate of I𝑒 . Improving diagnostics with a combined approach for this algorithm
took place in 3 cases. This allowed us to get 18 correct diagnoses out of 20 and, as a result,
made it possible to achieve 90% accuracy.
In addition, it should be noted that the use of the primary rejection of hypotheses on the
grounds of “against” excluded the correct diagnosis not in four, but only in 1 out of 20 cases,
that is, only in 5% of cases there was no possibility of making an accurate diagnosis.
Thus, the process of differential diagnosis using the developed models and algorithms is as
Table 6
Algorithm’s testing results
A confirmed diagnosis
among the first 5
Patient number Confirmed diagnosis hypotheses based on the
results of the algorithm
1 2 3 4
1 [13] Hunter syndrome Yes Yes Yes Yes
2 [14] Maroteaux-Lami syndrome Yes Yes Yes No
3 [15] Maroteaux-Lami syndrome Yes Yes Yes Yes
4 [12] Hurler syndrome Yes Yes Yes Yes
5 [16] Hurler syndrome Yes Yes Yes Yes
6 [17] Sanfilippo syndrome C Yes Yes Yes Yes
7 [18] Sanfilippo syndrome B Yes Yes Yes Yes
8 [19] Maroteaux-Lami syndrome No No No No
9 [20] Sly syndrome Yes Yes Yes Yes
10 [21] Sly syndrome Yes Yes Yes Yes
11 [22] Sly syndrome Yes Yes Yes Yes
12 [23] Maroteaux-Lami syndrome Yes Yes Yes Yes
13 [24] Hunter syndrome Yes No Yes No
14 [25] Hurler-Scheie syndrome Yes Yes Yes Yes
15 [26] Sly syndrome Yes Yes Yes Yes
16 [27] Morquio syndrome A Yes Yes Yes Yes
17 [28] Morquio syndrome A Yes Yes Yes Yes
18 [29] Morquio syndrome B No Yes No Yes
19 [30] Sanfilippo syndrome A No Yes No Yes
20 [31] Hurler syndrome Yes Yes Yes Yes
Total 17 18 17 17
follows:
• At the first step, the patient’s clinical picture analyzes the number of “for” and “against”
in relation to each hypothesis.
• In the second step, how many hypotheses are counted have 0 signs “against”. If 5 or
more, then the differential diagnosis process continues.
• In the third step, the percentage of coincidence calculated for the patient I𝑝 with the
reference I𝑒 for the selected hypotheses is compared.
• In the fourth step, the hypotheses are ranked and the percentage of coincidence of the
integrated estimates of I𝑝 from maximum to minimum, the first 5 are selected.
• In the fourth step, the hypotheses are ranked by the percentage of coincidence of the
integrated estimates of I𝑝 and I𝑒 from maximum to minimum, the first 5 are selected.
7. Differential diagnosis using an ontological system
The prototype of the differential diagnostic system is created on the basis of the IACPaaS plat-
form (https://iacpaas.dvo.ru/). Knowledge of orphan diseases, including expert assessments, is
presented in the form of an ontology. The solver will produce a list of ranked hypotheses at the
output of the system. As an explanation, the physician will receive a list of signs that served
as the cause of this hypothesis or hypoteses. Symptoms are grouped into 4 blocks according to
modalities: main, necessary, secondary and antisymptoms (signs “against” for this disease). In
the future, it is planned to request additional signs of the patient, information about which is
necessary to increase the likelihood of the hypothesis put forward.
8. Conclusion
The model for the integrated evaluation of expert knowledge for the differential diagnosis of
hereditary orphan lysosomal storage diseases has been implemented. The model is based on
an integrated approach to three parameters: modality, manifestation and degree of expression
of signs. It is the basis for comparing a new object with previously formed reference variants
of known clinical forms.
Expert evaluations (modality coefficients and certainty factors) were used to reflect the fuzzy
of the pathological changes. Linguistic and interval scales were introduced to characterize the
fuzzy signs and time boundaries of the manifestation of symptoms at different age periods.
During the study, differential diagnostic algorithms were developed for subsequent inclusion
in the IACPaaS platform. As a result of the experiment, the most effective combination was
selected among six algorithms. The experiment of differential diagnosis of mucopolysacchari-
doses was carried out using the clinical data of patients from articles in Russian and English.
The model that takes fuzzy transitional pathological conditions [5] may improve the effi-
ciency of disease recognition with the incomplete and fuzzy clinical picture.
References
[1] M. R. Bellettato, R. Tomanin, M. Scarpa, Pathophysiological aspects of lysosomal storage
disorders, in: R. Parini, G. Andria (Eds.), Lysosomal Storage Diseases: Early Diagnosis
and New Treatments, John Libbey Eurotext, Montrouge, 2010, pp. 31 – 42.
[2] S. Tomatsu, T. Fujii, M. Fukushi, T. Oguma, T. Shimada, M. Maeda, K. Kida, Y. Shibata,
H. Futatsumori, A. M. Montaño, R. W. Mason, S. Yamaguchi, Y. Suzuki, T. Orii, Newborn
screening and diagnosis of mucopolysaccharidoses, Molecular Genetics and Metabolism
110 (2013) 42 – 53. doi:https://doi.org/10.1016/j.ymgme.2013.06.007.
[3] K. Kawamoto, C. A. Houlihan, E. A. Balas, D. F. Lobach, Improving clinical practice using
clinical decision support systems: a systematic review of trials to identify features critical
to success, BMJ 330 (2005) 765 – 772.
[4] S. Bavisetty, W. W. Grody, S. Yazdani, Emergence of pediatric rare diseases, Rare Diseases
1 (2013) e23579.
[5] B. A. Kobrinskii, Fuzzy and reflection in the construction of a medical expert system, Jour-
nal of Software Engineering and Applications 13 (2020) 15 – 23. URL: http://www.scirp.
org/journal/PaperDownload.aspx?paperID=99099. doi:10.4236/jsea.2020.132002.
[6] B. A. Kobrinskii, Certainty factor triunity in medical diagnostics tasks, Scientific and
Technical Information Processing 46 (2019) 321 – 327.
[7] V. Gribova, A. Kleschev, P. Moskalenko, V. Timchenko, L. Fedorisdiev, E. Shalfeeva, The
IACPaaS cloud platform: features and perspectives, 2017 Second Russia and Pacific Con-
ference on Computer Technology and Applications (RPC) (2017) 80 – 84.
[8] B. A. Kobrinskii, N. S. Demikova, N. A. Blagosklonov, Knowledge engineering in con-
struction of expert systems on hereditary diseases, in: S. O. Kuznetsov, G. S. Osipov,
V. L. Stefanuk (Eds.), Artificial Intelligence, volume 934, Springer International Publish-
ing, Cham, 2018, pp. 35 – 45.
[9] A. A. Ivin, A. L. Nikiforov, Dictionary of logic, Humanitarian publishing center VLADOS,
Moscow, 1997. (In Russian).
[10] I. Z. Batyrshin, Towards a general theory of similarity and association measures: similar-
ity, dissimilarity and correlation functions, Journal of Intelligent and Fuzzy Systems 36
(2019) 2977 – 3004.
[11] S. Köhler, M. H. Schulz, P. Krawitz, S. Bauer, S. Dölken, C. E. Ott, C. Mundlos, D. Horn,
S. Mundlos, P. N. Robinson, Clinical diagnostics in human genetics with semantic simi-
larity searches in ontologies, The American Journal of Human Genetics 85 (2009) 457 –
464.
[12] L. S. Namazova-Baranova, N. D. Vashakmadze, M. A. Babaykina, E. N. Basargina, N. V.
Zhurkova, A. K. Gevorkyan, L. M. Kuzenkova, T. V. Podkletnova, K. V. Zherdev, O. B.
Chelpachenko, T. D. Degtyareva, Effectiveness of modern methods of treating type I mu-
copolysaccharidosis patients, Pediatric pharmacology (Pediatricheskaya farmakologiya)
11 (2014) 76 – 79. doi:10.15690/pf.v11i6.1220, (In Russian).
[13] N. D. Vashakmadze, L. S. Namazova-Baranova, A. K. Gevorkyan, L. M. Kuzenkova, A. D.
Khristochevskiy, L. M. Vysotskaya, A. S. Dadashev, Mucopolysaccharidosis type II, Pedi-
atric pharmacology (Pediatricheskaya farmakologiya) 8 (2015) 66 – 68.
[14] A. Jurecka, E. Zakharova, V. Malinova, E. Voskoboeva, A. Tylki-Szymańska, Atten-
uated osteoarticular phenotype of type VI mucopolysaccharidosis: a report of four
patients and a review of the literature, Clinical rheumatology 33 (2013) 725 – 731.
doi:10.1007/s10067-013-2423-z.
[15] O. V. Paramei, S. S. Zhilina, Eye manifestations of Maroteaux-Lami syndrome, The Rus-
sian Annals of Ophthalmology (Vestnik oftalmologii) 120 (2004) 41 – 42. (In Russian).
[16] O. Gabrielli, L. A. Clarke, A. Ficcadenti, L. Santoro, L. Zampini, N. Volpi, G. V. Coppa,
12 year follow up of enzyme-replacement therapy in two siblings with attenuated mu-
copolysaccharidosis I: the important role of early treatment, BMC Medical Genetics 17
(2016).
[17] H. J. Huh, J. Y. Seo, S. Y. Cho, C.-S. Ki, S. Y. Lee, J. W. Kim, H. D. Park, D.-K. Jin, The
first korean case of mucopolysaccharidosis IIIC (Sanfilippo syndrome type C) confirmed
by biochemical and molecular investigation, Annals of Laboratory Medicine 33 (2013) 75
– 79.
[18] Y.-E. Kim, H.-D. Park, M.-A. Jang, C.-S. Ki, S.-Y.Lee, J.-W. Kim, S. Y. Cho, D.-K. Jin, A novel
mutation (c.200T>C) in the NAGLU gene of a Korean patient with mucopolysaccharidosis
IIIB, Annals of laboratory medicine 33 (2013) 221 – 224. doi:10.3343/alm.2013.33.3.
221.
[19] J. A. Guio, G. AdolfoGiraldo-Ospina, Impact of enzyme replacement therapy in a
patient younger than 2 years diagnosed with Maroteaux-Lamy syndrome (MPS VI),
Journal of Inborn Errors of Metabolism and Screening 5 (2017) 1 – 8. doi:10.1177/
2326409817718849.
[20] S. Nampoothiri, M. Kappanayil, H. Ravindran, V. Sunitha, Sly disease: mucopolysaccha-
ridosis type VII, Indian pediatrics 45 (2008) 859 – 861.
[21] Y. Yamada, K. Kato, K. Sukegawa, S. Tomatsu, S. Fukuda, S. Emura, S. Kojima, T. Mat-
suyama, W. S. Sly., N. Kondo, T. Orii, Treatment of MPS VII (Sly disease) by allogeneic
BMT in a female with homozygous A619V mutation, Bone Marrow Transplantation 21
(1998) 629 – 634.
[22] A. C. Sewell, J. Gehler, G. Mittermaier, E. Meyer, Mucopolysaccharidosis type VII (𝛽-
glucuronidase deficiency): a report of a new case and a survey of those in the literature,
Clinical genetics 21 (1982) 366 – 373.
[23] S. S. Ibatova, T. T. Kerimbayev, G. N. Kasenova, Case report of mucopolysaccharidosis type
VI with a brief literature review, Journal "Neurosurgery and Neurology of Kazakhstan"
45 (2016) 36 – 41. (In Russian).
[24] N. I. Averianova, T. I. Rudavina, N. A. Domnina, Thrombocytopenia syndrome in a child
with type II mucopolysacharidosis, Perm Medical Journal 31 (2014) 110 – 114. (In Russian).
[25] E. K. Ryskulova, E. G. Khusnutdinova, A. E. Babushkin, G. Z. Israfilova, R. M. Mukhamet-
shina, A clinical case of Hurler-Scheie syndrome, Point of view. East-West (2016) 67 – 69.
(In Russian).
[26] P. Dubot, F. Sabourdy, G. Plat, C. Jubert, C. Cancés, P. Broué, G. Touati, T. Levade, First
report of a patient with MPS type VII, due to novel mutations in GUSB, who underwent en-
zyme replacement and then hematopoietic stem cell transplantation, International Jour-
nal of Molecular Sciences 20 (2019) 5345.
[27] K. Ramphul, S. G. Mejias, Y. Ramphul-Sicharam, Morquio syndrome: a case report, Cureus
10 (2018) e2270.
[28] S. N. Biswas, S. Patra, P. P. Chakraborty, H. Barman, Mucopolysaccharidosis type IVA
(Morquio A): a close differential diagnosis of spondylo-epiphyseal dysplasia, BMJ Case
Reports 2017 (2017) bcr–2017. doi:10.1136/bcr-2017-221156.
[29] Y. B. Sohn, H. D. Park, S. W. Park, S. U. Kim, S.-Y. Cho, A. R. Ko, C.-S. Ki, S. Y., D.-K. Jin,
A Korean patient with Morquio B disease with a novel c.13_14insA mutation in the GLB1
gene, Annals of clinical and laboratory science 42 (2012) 89 – 93.
[30] E. M. Ribeiro, A. C. Brusius-Facchin, S. Leistner-Segal, C. A. da Silva, I. V. Schwartz, Car-
diac disease as the presenting feature of mucopolysaccharidosis type IIIA: a case report,
Molecular Genetics and Metabolism Reports 1 (2014) 422 – 424.
[31] T. Gurumurthy, S. Shailaja, S. Kishan, M. Stephen, Management of an anticipated diffi-
cult airway in Hurler’s syndrome, Journal of Anaesthesiology, Clinical Pharmacology 30
(2014) 558 – 561.