Software Procedures for Training a Recognition System for The Differential Diagnosis of Patients Based on Diverse Symptom Complexes

• Published in: Conference proceedings (IEEE International Scientific Conference Electronics and Nanotechnology. Online) pp. 446 - 451
• Main Authors: Oleksandr, Shulyak, Druzhynin, Vladyslav
• Format: Conference Proceeding
• Language: English
• Published: IEEE 13.05.2024
• Subjects: clinical case symptom complexes; Differential diagnosis; heterogeneous; Histograms; Mathematical models; Nephrology; parameter value distribution; Probability density function; probability models; Software tools; Statistical analysis; Testing; Windows
• ISSN: 2693-3535
• DOI: 10.1109/ELNANO63394.2024.10756819

The paper discusses the construction of probability density function models for parameter values in heterogeneous symptom complexes characterizing clinical cases of patients in their differential diagnosis with three possible diagnoses in the field of nephrology. The models are considered as a way to present practical experience of successful patient diagnosis and as informational support for making preferred diagnostic decisions. The subject of the work is the development of mathematical and software tools for modeling and using the obtained models in clinical case diagnostics. To construct the models, a database with descriptions of clinical cases by symptom complexes with eight quantitative parameters, annotated with confirmed diagnoses, is used. All available statistics of the database are transferred to the proposed step models, which differ in their properties from ordinary histograms and, unlike them, can provide unambiguous decision-making in the full ranges of parameter values in working windows defined by the database statistics. The relevance of the work lies in the fact that the mentioned probability density function models capture some characteristic aspects of successfully diagnosed types of symptom complexes and may add additional potential to the existing toolkit for diagnosing similar clinical cases. The content of the modeling method and its software implementation, as well as the voting criterion and the corresponding software procedure for selecting diagnoses by patient symptom complexes using the developed probability density function models, are discussed. The models and the procedures for their formation underwent initial statistical testing with estimates of sensitivity, specificity, and overall validity of decisions.