An artificial intelligence system developed at Columbia University can now locate individual sperm cells in men who have been told they produce none, opening a new path forward in the treatment of male infertility.
The system, known as Sperm Track and Recovery (STAR), uses machine learning to scan semen samples at 300 images per second, identifying viable sperm cells hidden within debris that trained laboratory technicians would routinely miss. Once detected, a robotic microfluidic device extracts the cell within milliseconds, preserving its viability for fertilisation.
STAR targets men with azoospermia, a condition in which no sperm is detectable in the ejaculate using standard methods. Azoospermia affects approximately 10 to 15 percent of infertile men and around one percent of all men worldwide. Infertility itself affects one in every six people of reproductive age globally, with male factors contributing to roughly half of all cases.
Zev Williams, director of the Columbia University Fertility Center and lead developer of STAR, described the core challenge: a semen sample can appear entirely normal to the naked eye, yet reveal nothing but cellular debris under a microscope. His team built the system to solve precisely that problem.
Results so far are striking. Based on the most recent 175 patients treated, STAR found sperm in just under 30 percent of individuals previously told that a biological child using their own sperm was not possible. The system also recovered 40 times more sperm than a trained human technician working manually, and has achieved a sensitivity rating of 100 percent, meaning it can detect a single sperm cell in a sample if one exists.
The first baby conceived using STAR was born after its parents had spent nearly two decades trying to have a child. The technology has since been used with hundreds of patients from around the world, with demand continuing to grow.
One couple in the United States, whose identities have been protected for privacy reasons, became pregnant in late 2025 after two and a half years of trying. The husband had been diagnosed with Klinefelter syndrome, a genetic condition caused by an extra X chromosome that typically results in little or no sperm production. He underwent nine months of hormone therapy before a surgical testicular extraction provided a tissue sample that was processed through the STAR system. Eight sperm cells were isolated. One produced a viable embryo. Their baby is due in July 2026.
Williams drew the original inspiration for STAR in 2020 from astrophysics. Modern telescopes generate astronomical data sets too vast for human analysts to process manually, so researchers use machine learning to detect new stars and planets within them. The parallels to searching for rare sperm in a field of cellular debris were immediate.
STAR was named one of TIME magazine’s 2025 Best Inventions.
Experts welcome the advance but urge measured expectations. Siobhan Quenby, professor of obstetrics at the University of Warwick, described the combination of advanced imaging, engineering, and artificial intelligence as genuinely exciting, while noting that further research involving larger patient groups is needed before the technology’s full value can be established. Concerns have also been raised about the risk of overpromising outcomes to couples who have endured long and emotionally costly fertility journeys.
Beyond sperm detection, artificial intelligence is reshaping other areas of reproductive medicine, including personalised hormone dosing during ovarian stimulation and improved embryo selection in fertility treatment, pointing to a broader and accelerating transformation in how reproductive care is delivered.
