Evaluation of the effectiveness of the decision support algorithm for physicians in retinal dystrophy using machine learning methods

Abstract

Electroretinography is a method of electrophysiological testing, which allows diagnosing diseases associated with disorders of the vascular structures of the retina. The classical analysis of the electroretinogram is based on assessing four parameters of the amplitude-time representation and often needs to be specified further using alternative diagnostic methods. This study proposes the use of an original physician decision support algorithm for diagnosing retinal dystrophy. The proposed algorithm is based on machine learning methods and uses parameters extracted from the wavelet scalogram of pediatric and adult electroretinogram signals. The study also uses a labeled database of pediatric and adult electroretinogram signals recorded using a computerized electrophysiological workstation EP-1000 (Tomey GmbH) at the IRTC Eye Microsurgery Ekaterinburg Center. The scientific novelty of this study consists in the development of special mathematical and algorithmic software for analyzing a procedure for extracting wavelet scalogram parameters of the electroretinogram signal using the cwt function of the PyWT. The basis function is a Gaussian wavelet of order 8. Also, the scientific novelty includes the development of an algorithm for analyzing electroretinogram signals that implements the classification of adult (pediatric) electroretinogram signals 19 (20) percent more accurately than classical analysis.