Page 146 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

presence of an expert who can formulate an adequate set of that offer detailed explanations of their recommendations
knowledge – our research attempts to address the problem for knowledge bases. Various intelligent components of
across the compatibility and complementarity of these the system work with a single semantic description of the
paths, rather than interchangeability. source data, reading the elements they require.
Thus, the research scope is the manufacturing Each intelligent component (algorithm) usually
technology of continuously developing trusted systems to interprets its specific blocks of knowledge in clinical
support difficult clinical decisions. guidelines: diagnostic rules or treatment recommendations.
The engineering of such interpretable knowledge as
2. Materials and methods independent architectural components (knowledge about
2.1. The concepts of explanation in a decision diagnosis, risks, and treatment) is done based on their
3
6,7
support system ontologies. Existing methods and technologies make
creating, testing, and deploying knowledge bases and
Medical knowledge results from a person’s theoretical knowledge-based systems, or KES, possible.
and practical activities, reflecting the accumulation of
previous experience. Knowledge is the regularities of The most detailed part of the explanation is formed
the domain (relationships and rules) necessary to solve based on ontological knowledge (selected fragments in
problems based on new data (facts). Clinicians expect the analysis) and, in some cases, partially copies its cause-
recommendations or hypotheses from e-consultants with effect and structural connections. Forming the explanation
convincing explanations and systems with solid, reliable of the results in terms that are understandable to the doctor
knowledge. The production of detailed explanations is an and that correspond to his logic allows the application of a
important element of decision support systems in general full-fledged medical ontology.
and computer-interpreted clinical guidelines in particular. The explicit representation of all concepts and their
The generation of interpretable medical knowledge relations makes it possible to input, shape, store data (and
requires additional specialized mechanisms. The universal knowledge), and demonstrate results that can be generated
representation of knowledge in the form of rules processed automatically. Such ontological algorithms could be used
by a single “inference engine” has a limitation in medicine: in conjunction with machine learning approaches, either
since the number of rules is measured in thousands, it as a source of ground truth or as a thematic layer that could
becomes virtually impossible for experts to review, verify, be used to promote interaction or improve explainability.
refine, and correct them. When other AI modules are connected to generate advice,
recommendations, or solutions, they should work with a
Semantic models of the medical knowledge domain
are required to formulate clinical recommendations as single semantic description of the source data to ensure
compatibility in explaining all results.
components for ClDSS. These models must reflect the
logic that physicians use when appealing to their sentences Using medical ontology allows for creating a GUI in
and fragments. The description of a set of concepts, terms that are understandable to the doctor and also, as a
relationships, and constraints used by specialists when rule, forms a dialog script that corresponds to the logic of
solving problems and transferring knowledge between explaining the results. To support the doctor at all stages,
specialists is what we refer to as a knowledge ontology. it is important to combine AI modules as components of
medical knowledge, obtained in different ways, into a united
The hierarchies of classes of entities and their binary
relations are already a step toward the declarative ExClDSS, all based on a single terminological framework.
representation of a part of knowledge, but they cannot Existing explainable AI methodologies use large
cover the most important clinical connections (and we language models, requiring large capacities and rechecking
8,9
do not consider them to be a full ontological language). to mitigate potential hallucinations. The path used in this
Knowledge ontology is a part of the semantic description study is an improvement (based on ontology) of classical
of the medical domain, and the other part is the semantic explanation generation: during the decision process, the
representation of data and documents. In domain reasoner records its arguments in the knowledge base.
ontology, all the concepts of specialists, relations of
concepts, and restrictions on interpreting their meanings 2.2. Decision support system maintenance and
are defined. Together with them, the required types of viability concepts
statements are specified about factors such as necessary The developer of a clinical system has the task of designing
conditions, grouping, and cause-effect relationships. This and implementing the mechanisms that will provide
ontological approach makes it possible to develop systems further maintenance to meet the changing knowledge and

Volume 2 Issue 3 (2025) 140 doi: 10.36922/aih.5736