The Analysis of Mobile Platform based CNN Networks in the Classification of Sperm Morphology

Abstract

The diagnosis of male factor based infertility is performed by the evaluation of semen specimens in laboratories. Semen samples are investigated in terms of sperm concentration, morphology and motility. These investigations are generally performed manually by experts using microscopes instead of using computer based systems due to their high costs. However, manual observation also known as Visual Assessment (VA), has demonstrated significant subjectivity, including intra-observer and inter-laboratory variations. In this study, two CNN models especially for the possible usage in mobile platforms have been tested in the sperm morphology classification problem to eliminate the human factor in the analysis. In the analysis, three well-known sperm morphology data sets namely, HuSHeM, SMIDS and SCIAN-Morpho have been employed. Due to the data imbalance and scarcity problem of the utilized data sets, data augmentation and epoch analysis are also presented.