Artificial intelligence in the diagnosis of hereditary diseases

Abstract

Diagnostics of rare diseases presents significant challenges for both physicians and developers of artificial intelligence systems. This is due to the genotypic and phenotypic polymorphism of fuzzy clinical manifestations. Previously created computer systems did not include or insufficiently analyzed visual images of patients. The use of neural network technologies in phenotype analysis for disease recognition does not explain the proposed solution and their application is impossible on small samples of rare diseases. At the same time, a delay in nosological identification of diseases leads to the appearance of irreversible pathological changes that could be prevented with the timely appointment of pathogenetic therapy, which has recently appeared for a number of hereditary diseases. A version of the construction of a diagnostic system for rare diseases is presented, including visual image rows, accompanied by an explanation understandable to the physician. The peculiarities of constructing an expert system for differential diagnosis of hereditary diseases, implemented for lysosomal storage diseases, are indicated. A quantitative comprehensive assessment of the symptoms used in the diagnostic intelligent system is implemented based on expert confidence factors. The integrated approach includes measures of confidence in modality (diagnostic significance of a symptom), manifestation during a certain period of life, and severity. Options for constructing systems with user participation at the stage of evaluating the hypotheses put forward are considered. A prototype of a hybrid intelligent system implemented on the model of progressive Duchenne muscular dystrophy includes a module for analyzing precedents representing atypical variants of a rare disease. The development of systems for diagnostics of rare diseases can be implemented within the framework of a hybrid intelligent system built on the basis of a logical-linguistic-image paradigm in combination with decision-making based on precedents. The combination of symbolic systems on knowledge with subsymbolic ones based on neural network technologies will provide the possibility of joint decision-making using verbal and image components and providing users with meaningful explanations of the hypotheses put forward, taking into account the physician’s specialization.