Comparative analysis of neural network algorithms for severity analysis of valvular regurgitation

Valvular regurgitation is one of the leading causes of cardiovascular disease along with death. This paper compares the classification performance of neural network algorithms with respect to the severity of mitral regurgitation (MR). Six different texture feature sets are used from the regurgitant region using the first-order statistics (FoS), Spatial Gray Level Difference Matrix, Neighbor Gray Tone Difference Matrix, Statistical Feature Matrix, Texture Energy Measurement of Law and Fractal Dimension Texture Analysis. The FoS features are initially used individually, however, depending on their limitations, other texture features are merged together and used in addition. In the study, Alyuda neurointelligence software was used. The current approach to taxonomy work uses 3 training algorithms, namely quick promotion, online backdoor promotion and batchback promotion through software. The classification accuracy is significantly improved by using these texture features in combination, which are provided as individual input to the classification. The classification Quick promotion, online backpropagation, batch backpropage from parasternal long-axis view have 90.50 %, 96.65 % and 74.11 % taxonomic accuracy respectively. The results of this paper suggest that the proposed CAD system may be useful in assisting radiologists in establishing (verifying) gentle, medium and hard MR stages.