REVIEW OF METHODS AND TOOLS FOR IMPROVING THE EFFICIENCY OF MEDICAL IMAGE CLASSIFICATION MODELS

Abstract

The article presents methods and tools for improving the efficiency of medical image classification models. It considers various approaches to enchancing classification accuracy and reliability, including the use of deep neural networks, fusion methods and the application of the cumulative effect of image smoothing and enhancement techniques on the datasets used for COVID-19 analysis. The datasets consist of four sets of categorised data, including COVID-19, pneumonia, viral pneumonia and bacterial pneumonia. Different combinations of CLAHE, gamma correction, HE, AMF, TVF, Gaussian and Median filtering methods are used to analyse the cumulative effect to determine the most effective combination of methods. 2-DGE images typically contain several anomalies that hinder spot detection and analysis. The article discusses methods for dealing with anomalies arising from the use of two-dimensional gel electrophoresis (2-DGE) using a pre-processing system consisting of three stages, normalization, noise reduction and background correction, are considered, which allowed to improve the image for a posteriori analysis. The fusion methods such as Uniform Voting, Distribution Summation, Dempster-Shafer, Entropy Weighting, Density-Based Weighting are analysed and compared with the newly developed BitClassification Fusion Model method on 15 benchmark data selected from the UEA and UCR time series classification repository. The proposed methods can be used for medical diagnostics, allowing to improve the accuracy and speed of medical image processing, which in turn contributes to improving the quality of medical services and reducing the time to diagnosis. Presented in the article information is an overview.