The Effect of Nonlinear Wavelet Transform Based De-noising in Sperm Abnomality Classification

Abstract

Morphological sperm analysis is one of the crucial steps in the male-based infertility diagnosis. Currently, analyses are mostly performed by visual assessment technique because of its easy implementation, quick response and cheapness properties. However, the expertise level of the observer has great importance in the visual assessment technique. Results can be different and misleading according to the observer analysis capability. Therefore, human factor should be eliminated and the analysis should be performed by an objective computerized system. In this study, we used descriptor-based features in the classification of the normal, abnormal and non-sperm patches. Additionally, we investigated the effects of two de-noising techniques in the classification performance due to the presence of noises in the patches. Results indicate that the de-noising processes have great importance in the classification performance. Moreover, a wavelet based adaptive de-noising approach dramatically increased the performance to 86% with support vector machine polynomial kernel classifier.