Page 157 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

the same time, the service asks which values of which signs and the personalization to the patient, (iii) the transparency
need to be obtained additionally. of all applicable knowledge, (iv) the explainability of advice
The new cloud service and a declarative method for based on the essence of the knowledge and with a link
accessing it (based on the existing solver) demonstrate the to the source, and (v) the integrability of ExClDSS with
feasibility of the technique and approach for evolving KESs neural network services, capable of inputting data from
and the adequacy of the proposed infrastructure for the a structured document, such as the medical history. Our
development and ongoing evolution of KESs. participation in piloting the (Ex)ClDSS in some medical
clinical institutions aligns with the rhetoric of conferences
7. Conclusion emphasizing the importance of AI for healthcare.
The application of the proposed approach ensured the Some of the limitations of the study include the lack of
construction of scalable medical software services to pre-existing converters of formalized knowledge (e.g., in
support specialists of different profiles at different stages of the Protégé paradigm) into our development and support
work. It has been demonstrated that the proposed method environment (this would provide an opportunity for both
for producing medical software assistants brings them to the integration of high-quality knowledge and the quality
the level of explainable AI, which is the consequence of the control of the accumulated archives of precedents), a high
interpretability of clinical guidelines and knowledge about “entry threshold” to the IACPaaS platform for Python-
the course of diseases and their management. savvy programmers, and insufficient attention to colorful
The proposed methodology and production visualization tools and flexibility of data input.
environment for viable systems proved easy to learn and Today, we are helping to bridge the gap between AI
convenient for teamwork. For a medical diagnostic system, innovation and real-world applications. The experience of
each significant knowledge extension (more than 20 such moving to trial operations in 2024 has shown that doctors
acts were performed in total) required from 5 h to 2 working welcome such important general characteristics of these
days for an expert, 5 – 8 h for quality control, 20 min for systems. These include the ability to explain hypotheses
an architect, and without a programmer, which would be (results), a mechanism for adding specific knowledge (e.g.,
unattainable in another production environment. After new in the clinical information guidelines), and specific
each update, the product characteristics analysis showed properties of specific software systems (for risk assessment,
that the results were consistent with the case samples diagnosis, and prognosis). Doctors particularly appreciate
received from real practice. systems that facilitate a dialogue to increase the result’s
Further research should focus on integrating the accuracy. For treatment-related software systems, the
developed tools with textual facts, knowledge parsers, and ability to apply knowledge from modern, regularly
third-party diagnostic and predictive tools. A detailed updated clinical research is also crucial. Developers of
study is required to demonstrate whether the components ExClDSS, using our technology, emphasize the importance
working with structured information, verbal text, images, of features such as procedures for regular evaluation of the
and digital arrays can be combined into a single complex. knowledge base by subsets of precedents (from archived
This approach would save valuable time for users in critical sets and cases from the practice of specialists), as well as
areas of activity. reading and directly evaluating the knowledge contained
in a specific system.
Work is currently underway to expand the capabilities
of the approach further. Today, the bottleneck for us is an As technology developers, we consider it important to
adaptable user interface. The technology allows you to have a procedure to verify the accuracy or correctness of
generate three user interfaces based on the explanation hypotheses based on any subsets of precedents provided
ontology, but these features are insufficient. We are currently by clients or potential users. Therefore, we believe there is
working on creating tools for automatically generating an potential for this ontological technology to bridge the gap
interface based on the user model, considering the usability between AI innovations and their real-world applications.
requirements. Acknowledgments

The main contributions of the study are: (i) the
automation of multiple physician tasks by filling a single None.
structured “medical history” (integrated with full electronic Funding
medical record), (ii) the integration of formalized
knowledge from clinical guidelines and other reliable The work was supported by the Ministry of Education and
sources to satisfy both the relevance of the methods used Science of Russia (No.: FWFW-2021-0004).

Volume 2 Issue 3 (2025) 151 doi: 10.36922/aih.5736