A New Capsule Generative Adversarial Network for Imbalanced Classification of Human Sperm Images

Abstract

Male infertility as an effective factor can affect the lives of infertile couples. Sperm morphology is an important step in evaluating and examining semen in male infertility. The lack of samples of sperm head abnormalities compared to natural sperm samples can make the classification of sperm head images into an imbalanced classification problem. With the inability of common classification algorithms, capsule neural networks (CapsNet) provide a suitable platform for designing imbalanced classification models compared to other deep networks. Also, Generative Adversarial Networks (GANs) help improve the imbalanced classification of images by producing appropriate artificial samples. To this end, in this paper a new architecture is introduced based on CapsNet and GAN to evaluate the imbalanced classification of human sperm images. Reviewing and comparing the proposed model with other deep learning models in the balanced and imbalanced classification of human sperm images showed the superiority of the proposed model. Investigating the general methods of increasing data with the proposed model to increase data, it was concluded that the general methods have less resistance to reducing the number of data than the proposed model. Balanced classification of human sperm images was done by proposed model with 98.1 % accuracy. The proposed model also maintained a high sensitivity to the minority to the majority of 1:25, indicating its proper performance in the imbalanced classification of sperm images.

Keywords

• imbalanced classification
• human sperm
• infertility
• generative adversarial network (gan)
• capsule neural network (capsnet)
• deep learning