A Cognitive Model for Classifying Human Sperm Morphology using Convolutional Neural Network

Abstract

Evaluation of human sperm morphology is one of the important factors in the diagnosis of infertile male. Some Neural Network models was used to categorize the sperm morphology. In this proposed model, a deep neural network is drafted based on AlexNet, a pre-trained CNN. Human Sperm Head Morphology dataset (HuSHeM) was used in this model, which initially contains 216 RGB images. Training the model with 216 images was challenging and the accuracy obtained was oscillating between 65% to 69%. Data augmentation technique was used to overcome this challenge. The purpose to build the model is to automatically classify the sperm images based on the shape of the head and tail. Deep learning is put into the AlexNet in diverse ways such as, fine-tuning the weights of each layer and adding softmax layer, which has the cability to categorize the images accurately into the four different classes. Using this AlexNet, the training dataset obtained accuracy between 96% to 98% and testing dataset obtained 87% to 93%.