Page 150 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

A connector module is used to connect an external 4.3. Testing the quality of the system with a
microservice to the medical service, which solves an knowledge base
additional task for the same data (documents about the When testing the quality of ExClDSS work, “control
patient). It consists of standard tasks such as reading sets of clinical cases” should be used. This is a
the list of names of the required data in the declaration carefully selected set of documented medical histories
(specification) of this external service, finding the values containing the correct solution and facts sufficient to
of this data in the input document, composing a “PUT” develop the correct solution. Based on the “control set
request with the specified uniform resource locator, and of clinical cases,” a “control set of test cases” (CSTS)
sending it. If the microservice is not interactive, it is should be prepared by clearing out information that is
necessary to wait for a response, select fragments (specified
in the declaration) for explanation, and add them to the not important for the target task. This, in particular,
provided substructure of the final explanation. depends on the task being tested (treatment, diagnostics,
prognosis, or prevention).
4.2. Ontological approach to support and develop We believe it is important to use metrics to assess the
KESs quality of the ExClDSS components (such as sensitivity,
The dependency of all KES components on a domain specificity, positive predictive value, and precision) and
ontology supports the viability properties, which include metrics to evaluate the quality of ExClDSS performance
the replaceability of components for their improved in supporting the solution of specific clinical problems.
versions, admissibility of the improvement of the decision These metrics are defined to the CSTS. For example, for
method, permissibility of changing or adding functions (for the task of forming hypotheses about a diagnosis, we use:
example, the formation of additional results), adaptability CorrectnessEstimation (number of tests-with-a-finding/
of the user interface due to changes in the input data, and card of CSTS) and AccuracyEstimation (number of tests-
permissibility of expanding the ontology (adding concepts with-a-hit/card of CSTS), where a test-with-a-finding is a
and relationships). test clinical case of a certain disease, for which the ExClDSS
As a result, the structure of the KES and its components generated many hypotheses during testing, among which
does not require changes due to current maintenance and was this disease. The test-with-a-hit is a test clinical case
sustainable development. As mentioned above, this class of for which the ExClDSS generated its disease as the only
26
software systems (ClDSSs) requires specialized maintenance hypothesis. The card of CSTS is the power of the set of all
tools because clinical guidelines (and other medical prepared test clinical cases.
knowledge) are constantly evolving. Due to the importance These quantitative metrics are associated with clinician
of evolving knowledge bases, only ClDSSs integrated with a satisfaction. For almost every clinical task (except for
knowledge base management system should be considered. differential diagnostics), the metrics for assessing the
A toolkit for building application systems with accuracy and correctness should be determined separately
declarative (interpretable) knowledge is based on a domain for each of the diseases, knowledge of which is “embedded”
and problem ontology. If a separate formal ontology is in the ExClDSS. Each of these tasks has its own requirement
created for each task (diagnosis, treatment, prognosis, for the set of patient data in the test clinical cases used, and
etc.), it is easier to develop tools to ensure the quality they are not the same for acute, slowly progressive, chronic,
of homogeneous, localized knowledge. The tools for and hereditary diseases.
developing and verifying knowledge bases are desirable
to be integrated into the architecture of the decision 5. Manufacturing environment for viable
support system (to be a part of the integrated architecture clinical decision support systems
of the decision support system). Thus, a maintenance As noted above, this class of clinical software systems
environment has to provide knowledge base editors, tools in the construction of <ontology + set of ontological
for assessing knowledge bases by archives of etalons (solved knowledge bases, set of ontological interpreters, set of user
problems), tools for checking and evaluating the quality of interface components, sets of facts> requires specialized
knowledge bases, and tools for the inductive formation of development and maintenance tools where coding tools
knowledge base fragments. for new software units or their new versions may turn out

If the pre-existing solver is in accordance with to be unnecessary. The authors are aware of ontological
the problem statement and “building up” additional portals but not of development environments focused
functionality is not supposed to be used, then coding tools on quality assurance and the development of ontological
for program units may be unnecessary. components for such systems.

Volume 2 Issue 3 (2025) 144 doi: 10.36922/aih.5736