Sperm-cell DNA fragmentation prediction using label-free quantitative phase imaging and deep learning

In intracytoplasmic sperm injection (ICSI), a single sperm cell is selected and injected into an egg. The quality of the chosen sperm and specifically its DNA fragmentation have a significant effect on the fertilization success rate. However, there is no method today to measure the DNA fragmentation of live and unstained cells during ICSI. We present a new method to predict the DNA fragmentation of sperm cells using multi-layer stain-free imaging data, including quantitative phase imaging, and lightweight deep learning architectures. The DNA fragmentation ground truth is achieved by staining the cells with acridine orange and imaging them via fluorescence microscopy. Our prediction model is based on the MobileNet convolutional neural network architecture combined with confidence measurement determined by distances between vectors in the latent space. Our results show that the mean absolute error for cells with high prediction confidence is 0.05 and the 90th percentile mean absolute error is 0.1, where the range of DNA fragmentation score is [0,1]. In the future, this model may be applied to improve cell selection by embryologists during ICSI.

Automated summary

In this study, a new method using unstained imaging data and deep learning techniques was proposed to predict DNA fragmentation in sperm cells. By comparing with cells imaged using fluorescence microscopy, we obtained the accuracy of the prediction model and found a low mean absolute error. This method has the potential to improve cell selection during ICSI.