Supervised machine learning classification of human sperm head based on morphological features

Abstract

We developed an automatic framework to classify sperm heads as normal or abnormal using image processing and machine learning techniques. The framework segments each sperm head using a color-space-based classification method. A novel set of morphological features is proposed to better describe the sperm head morphology. Finally, a supervised learning model is trained and tested to analyze the feature data for classification. To train and test the model, a publicly available dataset of human sperm heads was used. All sperm samples were manually labeled as normal or abnormal according to the strict criteria of the World Health Organization laboratory manual for the examination and processing of human semen. The segmentation method preserves shape without losing key morphological aspects. The classification model based on morphological descriptors produces better discrimination as compared with the traditional shape descriptors, achieving a 92% accuracy in the discrimination of normal or abnormal spermatozoa.