Page 156 - AIH-2-3

Artificial Intelligence in Health Explainable solutions from AI for HSSs

Inductive knowledge generation tools based on tabular ontological approach, we developed several diagnostic
data sets are used for two types of tasks. To test the consultants with a knowledge base based on the rules.
knowledge, tools are used to compare solutions in the The development of the first version of the diagnostic
reference case history sets with the generated results of service for a group of diseases (therapy, ophthalmology,
each AI component integrated into the medical IACPaaS and gastroenterology), including about 15 diseases, took
services. Two comparative analyses were carried out: an average of 20 months: for testing with debugging
(i) opportunities for generating advice: IACPaaS-ExClDSS (8 months), to expand with another disease (a month),
(including integrated ones) versus other available services, and subsequent testing with debugging (2 months).
and (ii) labor costs (as the number of employees) to With cloud implementation of KES (with separation of
maintain the relevance of the knowledge base. In the competencies) for a similar group of diseases, it takes on
cloud implementation of knowledge-enabled services, average 10 months (due to a more understandable form
there is a division of competencies versus traditional KESs of writing): Ontological Diagnostic Reasoner (5 months),
development. testing with debugging (3 months), to extend with another

The comparative analysis involved five medical internet disease (1 month), and new testing with debugging is a
services (https://symptoms.webmd.com, https://www. month.
everydayhealth.com/symptom-checker, https://www. Internet services provide savings by scaling their use.
mayoclinic.org/symptom-checker, https://kiberis.ru, and Theoretically, to offer many specialists and teams of the
https://symptomchecker.isabelhealthcare.com) and five of same profile, additional efforts are required only to integrate
the most recent IACPaaS “assemblies.” the presentations of patient data. However, in practice,
Five clinical cases were randomly selected from the services have a limited set of concepts that do not allow
accumulated set of clinical cases from practice, and five them to accept all the information about the patient. In the
clinical cases with the same diagnoses were taken from (new) cloud technology KES, with a separate declarative
publications on PubMed. As for internet services, for 10 knowledge base and a powerful glossary of terms, there are
reference cases, the true diagnosis was rarely in the top several savings due to a single center of knowledge update
three (Table 1): 2 – 4 times (out of 10 stars), and more and knowledge control (by accumulated case histories).
frequently in the top 10: 6 – 8 times. In contrast, with Similarly, the knowledge base about treatment is improving
IACPaaS, the true diagnosis appeared in the top three (but there is no way to compare because the team did not
7 – 8 times, and always within the top 10. have much experience in the past).
The results of treatment and prognosis are difficult to The situation to test the system’s viability was
compare, because only one internet service was ready to requested in early 2020 to expand the service for the
issue a cure, and two of our selected IACPaaS assemblies diagnosis, differentiation, and treatment of COVID-19. In
had a treatment module (prescription was issued for comparison with other service providers who presented
eight + six cases, 6 times it coincided with prescription updated versions a few months after the appearance
from the reference case). of diagnostic guidelines (Infermedica, klinica.com.ua,
and medicase.pro), in our technology, the addition of
The prognosis is always implemented separately from an existing knowledge base to describe several known
other tasks. It is aimed at a specific disease, while IACPaaS variants of manifestation, course, and diagnosing methods
services can predict disease course at the initial stages of of the new disease took several days.
diagnosis and predict recovery course (with the prescribed
treatment). For this extension of the KES, medical experts used two
knowledge base editors, adding several dozen statements
Before creating a cloud platform and developing an of diagnosis and treatment. The accuracy of the updated
knowledge base was evaluated using the first 15 case
histories available. A week later, a new cloud service was
Table 1. Comparative analysis of diagnostic hypotheses of
two types of Internet services launched to search for hypotheses about a patient’s possible
viral disease and differential diagnosis. This cloud service
Type of service The reference The reference diagnosis is an example of explanatory AI (Figure 6). It provides a
diagnosis was in the was in the top 10 rationale for the proposed solutions and recommendations
top three (unlike the services of klinica.com.ua or medicase.pro).
Internet 2 – 4 6 – 9 The service indicates which signs of the disease are/are not
Intelligent adaptive 6 – 8 10 included in the clinical picture of the disease and whether
clinical platform additional information is needed to confirm or refute it. At

Volume 2 Issue 3 (2025) 150 doi: 10.36922/aih.5736