Detecting tubules and spermatogonial stem cells in mouse testis micrographs by using artificial intelligence techniques

Spermatogonial stem cell that provide male fertility, are few in number and difficult to detect. The cells numbers detection is critical in infertility due to prepubertal testicular damage. Several cell types can be counted by segmentation and feature extraction methods by image processing and artificial intelligence technique. In this study, we hypothesized to develop a bioengineering system that allows rapid, safe and easy quantification of spermatogonial stem cells by using segmentation, feature extraction and deep learning methods (YOLOv4). To test this hypothesis, paraffin serial sections were obtained from C57BL / 6 type newborn male mice testes and stained with hematoxylin eosin and SALL4 that is a spermatogonial stem cells marker. Classification process was performed using machine learning methods, support vector machines, decision tree, random forest, naive bayesian, k nearest neighbor, and logistic regression and seminiferous tubule data sets were obtained with active contour model in 66 digital micrographs, the detection rate was calculated as 96,5%, 96,5%, 97,1%, 90,4%, 89,1%, and 96,5% for respectively support vector machines, decision tree, random forest, naive bayesian, k nearest neighbor, and logistic regression. A data set was created using 12734 and 9889 data by YOLOv4 for seminiferious tubules and spermatogonial stem cell respectively. The detection rate was 94% and 93% for seminiferous tubules and spermatogonial stem cells repectively. In conclusion, a new artificial intelligence technique has been created by using testiscular section micrographs and it has been successfully worked for detecting spermatogonial stem cells. This new bioengineering product has a potential to be translated into in vivo experimental animal models in order to calculate spermatogonial stem cell pool number and then to be used in prepubertal children testicular biopsies in future.