The Effect of Differently Optimized Histogram Stretching Techniques on the Classification Performance of Sperm Morphology

Abstract

Infertility diagnosis in men is done by spermiogram test, which is based on the result of the sample being examined under a microscope in a laboratory environment. This test evaluates parameters like the morphological structure and motility of sperm cells. Due to its high error rate, this test is usually repeated twice, making the process time-consuming and labor-intensive. In this study, we have proposed an artificial intelligence supported decision-making mechanisms to classify the sperms for infertility diagnosis and to minimize the error rate. Two well-known datasets as HuSHeM and SMIDS were used in the experiments. Beyond the original datasets, image enhancement techniques have been performed to generate different versions of the datasets. After these enhancements, models were trained using deep learning algorithms with the data obtained. These image enhancement techniques are used to increase the accuracy of the classifications of images and to reduce the error rate. The main purpose of this study is to investigate the impact of diverse image enhancement techniques on the classification performance achieved with various deep learning network models, including EfficientNetB3, DenseNet, and Resnet50.