Human Sperm Health Diagnosis with Principal Component Analysis and K-nearest Neighbor Algorithm

Abstract

Sperm morphology is an important diagnostic basis to identify if a sperm cell is healthy or not. This paper presents a method that using principal component analysis (PCA) to extract image features and k-nearest neighbor (KNN) algorithm to diagnose sperm health. We first accurately locate the position of sperm in the microscope images, and segment some small sperm division with a fixed size. Then some of divisions are selected as the training set to classify the remaining small sperm divisions. In this experiment, while the diagnosis accuracy depends on the training set, we have already selected a better training set and obtained a good performance with 87.53% compared with other feature extraction methods such as scale-invariant feature transform (SIFT) and other classifier such as back propagation neural network (BPNN).