Non-Destructive and Real-Time Virtual Staining of Spermatozoa via Dark-Field Microscopy

Sperm morphology serves as a crucial indicator of fertilization potential; however, the fixation and staining required for its assessment irreversibly result in irreversible damage to sperm. Here, an improved Generative Adversarial Network (GAN) virtual staining model based on dark-field microscopy enables real-time conversion of label-free semen smears into high-contrast Papanicolaou-equivalent stained images. The experimental results demonstrated that our network completed virtual staining in ~0.047 s for a 2048 × 2048 image. Furthermore, the assessment showed that the mean squared error and the structural similarity between virtual staining and true Papanicolaou staining are 0.0044 ± 0.0031 and 0.905 ± 0.015, respectively. Our network bypasses the typically labor-intensive and costly histological staining procedures, enabling real-time, non-destructive virtual staining of motile spermatozoa without the need for laboratory quality control, and paves a novel way for selection of sperm for intracytoplasmic sperm injection (ICSI).