Page 149 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

point. The document “medical history of any patient” the user interface of the knowledge editing tool must meet
contains a structured set of facts observed or objectively the requirements and expectations of domain experts.
measured in the considered situation (medical case) The main challenge in medical systems manufacturing
regarding which the problem is solved. is ensuring that the knowledge reflects current knowledge
All results and decisions are recorded in the same (e.g., clinical guidelines) and continuous improvement.
document (medical history), regardless of the method Continuous improvement of the knowledge base allows it to
in which they were collected. The place in the document become a reliable source (repository) of expert knowledge,
structure must be strictly connected with the essence hoping to create a “reference” knowledge base. Its quality
of the result (diagnosis in one place and prognosis in will determine the success of the use of this knowledge.
another). Such a document structure is part of the domain The relevance of a knowledge base is achieved through
ontology. Thus, it can be asserted that ontology ensures the three main ways: 23-25 interactive change of the knowledge
integrability of various components. base, usage of machine learning methods (tools of

Often, it is necessary to add a pre-existing software inductively generating knowledge from selected precedents
service with hidden knowledge (trained model) to the and tools of knowledge discovery from “big data”), or a
system to solve a specific problem. Typically, this task falls combination thereof. The “success” of adaptability depends
into one of three categories: risk assessment, prediction, on several conditions and principles.
or recognition of a class of pathologies. A structured
description of the service is sought, which includes the 4.1. Architectural properties of clinical systems
following elements: (i) name and author of the method, enabled with declarative knowledge
(ii) essence of the result, (iii) vector of initial data, (iv) This intelligent software system class, which explains
conditions of applicability (entering values in limited decisions, requires specialized development and
ranges), (v) manner of launching the service, and (vi) if maintenance tools. The key principle is the special role
the expected response of the service is numeric, then the of the knowledge ontology (as a model of professional
description should also include the interpretation of the concept relations). Its formalized representation, separated
result. from the professional knowledge itself, allows for the
For the mutual exchange of data and results with independent development of each ontological component,
software services (with hidden knowledge), a single relying on its integrability. The medical ontology makes it
semantic template is used: <name of method, author, possible to create a GUI that is understandable to the doctor
essence of the result, vector of initial data, [conditions and, as a rule, to create a dialogue script corresponding to
of applicability], description of result interpretation, the logic of explaining the results.
launch method or full address of microservice>. For The interpretation of knowledge consists of choosing
example, for a software service for assessing the risk of each hypothesis and transitioning from it (along the
developing a disease, the description of the interpretation chain of connections) to the expected values areas of
of the result = a set of pairs <threat level value, range of observations for subsequent comparison with facts, as
calculated values>. Adding such a semantic template (with well as constructing an explanation with the collected
a description of the interpretation of the result) to the arguments. The “structural” complexity of the ontological
medicine ontology ensures the explainability of connected, interpretation algorithm is determined by the number and
intelligent services. The vector of initial data (from the the length of the chains of cause-and-effect relationships in
semantic template) should be formed only with the help the statements, the degree of fuzziness prescribed in cause-
of the terms of the “medicine” thesaurus. The “medicine” and-effect relationships and statements, and the structure
thesaurus (a dictionary of terms for observing a patient of observation, description, or conditions for a decision.
and studying the patient’s body) is traditionally considered To develop interpreters (task solvers), coding tools
part of the ontology of this domain area. Hence, the for new software units or their new versions, tools for
ontology is a structural basis of both tools for experts and cataloging units for reuse, and tools for integrating reusable
users (editing tools) and for software components of KESs. units and new ones into new solvers or their new versions
4. The viability model of clinical are needed. Solvers built according to a given ontology (for
knowledge-enabled decision support systems problems of diagnosis, prognosis, etc.) must be reusable
reasoning engines of medical services. A version of clinical
An ontology, as a structural basis for viewing and editing knowledge is their input parameter. Therefore, regular
knowledge bases, provides a basis with a declarative updating of the knowledge base does not require changes
property. However, for the knowledge base to be adaptable, in other components of the KES.

Volume 2 Issue 3 (2025) 143 doi: 10.36922/aih.5736