Page 148 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

(for example, risk of hypokalemia in patients with arterial diseases, the statements about the method to eliminate the
hypertension) and predicting complications. 17,18 Almost all cause of the disease:
services based on machine learning provide versions of a <diagnosis , event , set-of {(observation , new-range
u
k
preliminary diagnosis without considering the dynamics of of observation values)}, delay >; jk iu
the patient’s observations. 19,20 For such tasks (preliminary ku
diagnostics, risk assessment, or forecasting), as a rule, an The statements about the impact on an organism for
intelligent service becomes inaccurate if it was “trained” on recovery start:
the data of one institution and it tries to operate in other <process variant , (period , interval ), treatment
in
kn
circumstances. event , set-of {observation , value range in (period (i+1)n ,
jkn
u
jkn
There are tasks for which no one has yet accumulated interval (i+1)n )>}}>;
adequate training material. In medicine, these are The statements about acting on a symptom to
corrections to disease treatment and differential diagnosis. alleviate it:
For this, intelligent services consultants trained on text <process variant , (period , interval ), observation ,
corpora (and GPT helpers) within the idea of hybrid values range, (event , delay ), values new range >. jk
in
kn
in
services to support the doctor’s work may be used. jkn u kj iu
The domain ontology represents all types of statements
3.1. The influence of the ontological model on the as a structural language (template) for introducing or
properties of system components describing knowledge. The knowledge base explicitly
For ClDSS to correctly formulate advice or results for contains sets of statements of the corresponding type
solving medical problems (risk, diagnostics, treatment, sufficient for this profession.
and prognosis), it needs to operate with concepts that The traditional architecture of KES is the knowledge
specialists use. For example, for knowledge of the task of base + fact base + intellectual problem solver + intellectual
monitoring the recovery process, one of the most common GUI. 15Finn2004 Knowledge bases are generated manually or
types of sentences (statements) is: inductively, including training samples from archives and
<process type , set-of {(period + interval ), set-of databases; this process involves inductive generalizations
k
ik
16
{characteristic , characteristic values range }}>. ki in machine learning. Bayesian classifiers, clustering
j ijk algorithms, and reinforcement learning 21,22 are sometimes
For the diagnosis of diseases, statements about the involved in this process.
relationships are needed:
The approach to creating systems with transparent
<diagnosis , process’ existence necessary condition , knowledge bases is based on an architecture expanded
k
k
set-of {factor }>;
km by a new component: ontology + ontological knowledge
<diagnosis, set-of {symptom complex | variant of base + ontological fact base + ontological interpretator
jk
j
jk
disease course}, [necessary condition]>; (problem solver) + intelligent graphical interface. It can
j include databases with reference information, operational
<symptom complex , set-of {feature, range of values information, and work with files.
k
kj
j
of feature}>;
Often in one domain, a set of interrelated tasks is
<symptom complex , set-of {sign , {period , duration solved; examples of related tasks are diagnosis, treatment,
ik
k
jk
of period , range of values of sign in period }}>;
i ijk j i and prognosis. To solve all tasks, one formal domain
<variant of disease course, set-of {(period + interval ), ontology can be created, but it is more convenient to map
in
n
in
set-of {observation , set-of {observation element , range of a separate ontological resource to separate tasks solved in
jk
k
values >}}>; the domain. A set of formal ontologies for related tasks
ijkn may be required when designing applied systems with
<necessary condition , set-of {factor }>. knowledge bases, which is followed by the creation of a set
km
k
All such concepts and relationships are explicitly written of ontological knowledge bases (for each task).
and “available” to algorithms when they are developed
based on ontologies and access to explicit knowledge 3.2. The influence of ontology on the integrability of
resources. The medical ontology was created by experts various components within a single architecture
and knowledge engineers. The factors without which a To combine various achievements of knowledge
disease does not begin can be the events or properties of engineering and AI in complex medical intelligent
an organism; they can also determine one of the options trustworthy systems, it is reasonable to choose the
for the development of the disease. For the treatment of document “patient’s medical record” as an integration
Volume 2 Issue 3 (2025) 142 doi: 10.36922/aih.5736