Page 147 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

service conditions in the created software product. The Clinical decision support systems should have
same mechanisms help solve problems in implementing one part (knowledge) that is constantly evolving and
“continuous delivery,” a process in which software is always another part that can read and understand it, i.e., be an
kept relevant. interpreter. Medical knowledge, such as clinical guidelines,
The application of typical architectural solutions, is an evolving part of ExClDSS. ExClDSS are expected
declarative representation of components, and separation to remain useful and effective in an environment of
of competencies between developers of components of changing knowledge. Under conditions of variability in
different types are all used to create maintainable decision clinical knowledge, the viability of the medical system is
support systems. Information technology managers manifested in its ability to adapt and update in response to
struggle to scale AI projects because they lack the tools to new information and evolving practices.
create and manage a “production-grade AI pipeline.” 10 In medical knowledge, the influence of factors and
With the advent of complex software systems, the events on the patient’s state, their change over time,
problem of their long-term maintenance has become individual characteristics, and some of their processes
more and more critical. Maintenance is the possibility of on others is important. The development (evolution) of
adaptation (to hardware and system software, to new types such complex knowledge bases is the main “challenge”
of human-machine interfaces and users) and extensibility of modern “conditions” with (Ex)ClDSSs. “The ability to
(at the request of users). Modification of software systems is adapt under a change in the set of facts and knowledge” is
due to changes in operating conditions, user requirements, one of the aspects of intelligence. 15(p.5) For medicine related
and subject area. In the operation of applied systems to to solving intellectual problems, this implies the evolution
support professional activity, there is a need to add new of knowledge. The ontological approach to knowledge and
user-defined functions (and adapt to new devices or user programming for working with them was sufficient.
interface changes). The average maintenance cost of the 3. Clinical decision support systems
software system life cycle is about 50%, but according to
11
some reports, it can reach 80 – 90%. 12 as software systems that apply
In addition to maintenance, viability has become a understandable knowledge
modern, useful property of software systems. It is defined To ensure that doctors trust ClDSSs, their developers
as sustainability in a changing environment (maintaining need to demonstrate correctness (sometimes accuracy)
usefulness and operability), and the ability to evolve, as the on subsets of precedents (cases from practice), implement
13
ability to adapt with the least possible cost to requirements’ the ability to explain the proposed solution or hypothesis
variability, maintaining architectural integrity. We will (the explanation must be understandable, consistent with
14
specify “the viability” as software system resilience to some formalized knowledge), have a mechanism for permanent
functioning environment changes (the maintenance of improvement of the knowledge base that does not worsen
working capacity) and the ability to develop over the “life” its correctness, apply procedures for regular evaluation of
(evolvability). stored clinical knowledge, and provide the opportunity for
specialists to read and evaluate the included knowledge.
In the case of applied decision support systems
in intelligent tasks such as diagnosis, planning, and Knowledge must be formed considering standardized
forecasting, the situation is different. Here, knowledge clinical guidelines and under domain experts’ control. One
variability and the emergence of new solutions, such as method is to use trained text recognizers. Knowledge can
creating new diagnostic methods and identifying new sometimes be created by experts themselves (possibly with
influencing factors, are expected, rather than just the knowledge engineers and cognitive scientists). In this case,
extension of user functions. Therefore, the approach to experts fully participate in the development and maintenance
maintaining ClDSS should not be similar to maintaining process with programmers and designers. This requires
application software systems. knowledge bases to be presented in a form understandable
to medical domain experts. When knowledge is isolated
Many well-known tasks in diagnosis, treatment, and
prognosis in general and medicine, in particular, are quite and framed in independent architectural components and
knowledge bases, the system using them becomes a KES.
stable. Algorithms for solving them are described and
can be qualitatively programmed once for long-term use. Several AI, mathematical modeling, and machine
However, this is not the case for medical knowledge and learning methods for solving practical problems provide
clinical guidelines. They cannot be “sewn” into programs medical services based on hidden knowledge. In
16
because they are regularly updated in this dynamic field. medicine, these are most often the tasks of risk assessment

Volume 2 Issue 3 (2025) 141 doi: 10.36922/aih.5736