Automated Sperm Head Morphology Classification with Deep Convolutional Neural Networks

• Marco Antônio Calijorne Soares FUMEC University
• Daniel Henrique Mourão Falci FUMEC University
• Marco Flávio Alves Farnezi FUMEC University
• Hana Carolina Moreira Farnezi FUMEC University
• Fernando Silva Parreiras FUMEC University
• João Victor Boechat Gomide FUMEC University

Resumo

Background and Objective: The morphological analysis of sperm cells is considered a tool in human fertility prognosis. However, this process is manual, time-consuming and dependent on professional expertise. From a computational perspective, this is a challenging problem due to the high inter-category similarity between the objects of interest and the amount of data available. In this paper, we propose a Convolutional Neural Network model to automate morphology analysis of human sperm heads. Methods: We performed K-Fold cross-validation experiments over two publicly available datasets and assessed the performance of the proposed approach using Accuracy, Precision, Recall and F1-Score. We also compared the proposed model with well-known Convolutional architectures and previous approaches on the same task.Results: Experimental evaluation showed that our approach achieved a macro-averaged F1-score of 0.95 while our best model attained an accuracy of 97.7%. The error analysis revealed a balanced classifier over different sperm head classes. Conclusions: We proved that the proposed approach outperformed the previous state-of-the-art results on this task.

Palavras-chave: Measurement, Head, Image recognition, Computational modeling, Morphology, Computer architecture, Manuals