Identification of Human Sperm based on Morphology Using the You Only Look Once Version 4 Algorithm

Abstract

Infertility is a crucial reproductive problem experienced by both men and women. Infertility is the inability to get pregnant within one year of sexual intercourse. This study focuses on infertility in men. Many causes that can cause infertility in men including sperm quality. Currently, identification of human sperm is still done manually by observing the sperm with the help of humans through a microscope, so it requires time and high costs. Therefore, high technology is needed to determine sperm quality in the form of deep learning technology based on video. Deep learning algorithms support this research in identifying human sperm cells. So deep learning can help detect sperm video automatically in the process of evaluating sperm cells to determine infertility. We use deep learning technology to identify sperm using the You Only Look Once version 4 (YOLOv4) algorithm. Purpose of this study was to analyze the level of accuracy of the YOLOv4 algorithm. The dataset used is sourced from a VISEM dataset of 85 videos. The results obtained are 90.31% AP (Average Precision) for sperm objects and 68.19% AP (Average Precision) for non-sperm objects, then for the results of the training obtained by the model 79.58% mAP (Mean Average Precision). Our research show result about identification of human sperm using YOLOv4. The results obtained by the YOLOv4 model can identify sperm and non-sperm objects. The output on the YOLOv4 model is able to identify objects in the test data in the form of video and image.