Pregnant after 18 years as AI helps find Hidden Sperm

A team of skilled embryologists spent two days searching one man's semen sample for a single sperm cell. They found nothing.

One hour later, an AI system called STAR, found 44.

That moment, at Columbia University Fertility Center, sits behind the first confirmed pregnancy from AI-guided sperm retrieval, a new approach to one of the toughest problems in male infertility: azoospermia, a diagnosis of zero measurable sperm.

The diagnosis that blindsides healthy men

A standard semen sample typically holds hundreds of millions of sperm, but for a man with azoospermia, that count is zero.

Most men with the condition feel completely normal and have no symptoms.

Male-factor issues play a role in roughly 40 percent of infertility cases, according to Dr. Zev Williams, director of the Columbia University Fertility Center. Azoospermia affects an estimated 10 to 15 percent of men diagnosed with infertility.

Until now, in such cases, doctors relied on surgical extraction from testicular tissue, which carries risk of scarring and can only be repeated a limited number of times.

Sometime, hormone medication also helped, but only when an underlying hormone imbalance is the cause.

“This often is a really heartbreaking and shocking and unexpected diagnosis,” Williams said. “Most men who have azoospermia feel completely healthy and normal. There’s no impairment of their sexual function, and the semen looks normal, too. The difference is that when you look at it under a microscope, instead of seeing literally hundreds of millions of sperm swimming, you just see cell debris and fragments but no sperm.”

One cell in a thousand haystacks

STAR pairs high-speed imaging with an AI model trained to recognize a single sperm cell by shape alone. Here is what happens once a sample goes in:

1. A high-powered camera scans the sample, capturing more than 8 million images in under an hour.
2. AI software reviews the images in real time, flagging anything that matches the profile of a sperm cell.
3. A microfluidic chip, engraved with channels roughly the width of a human hair, isolates the flagged cell from surrounding debris.
4. A robotic mechanism lifts the cell out within milliseconds, with no lasers, stains, or centrifuging that could damage it.
5. The cell can be used right away to fertilize an egg, or frozen for later.

“It’s like searching for a needle scattered across a thousand haystacks, completing the search in under an hour and doing it so gently, without any harmful lasers or stains, that the sperm can still be used to fertilize an egg,” Williams said.

The name is a nod to astronomy. “They’ll look at a sky with hundreds of millions of stars in it, and they want to find one different than all the others,” Williams told the Columbia Daily Spectator. “So, we had to look through a field of hundreds of millions of fragments to find the one sperm.”

44 sperm, two embryos, one pregnancy

The clearest test of STAR came from a couple who had tried to conceive for close to two decades. Fifteen IVF cycles. Two surgical extractions. Years of manual searches under a microscope. None of it had worked.

Then came the sample the technicians spent two days searching by hand and came up empty. STAR found sperm in it within an hour.

“A patient provided a sample, and highly skilled technicians looked for two days through that sample to try to find sperm. They didn’t find any. We brought it to the AI-based STAR System. In one hour, it found 44 sperm,” Williams said.

“So right then, we realized, ‘Wow, this is really a game-changer. This is going to make such a big difference for patients.’”

Three hidden sperm cells came out of that sample. A case report on the pregnancy, later published in The Lancet, said two were confirmed viable and became embryos, drawn from a 3.5 mL sample that STAR scanned in about two hours.

“When they got the positive pregnancy test, the entire lab was jumping up and down, cheering,” Williams told the Columbia Daily Spectator. “To help bring life and joy into the world, to see that kind of success, that’s literally why we do everything that we do.”

The couple asked to stay anonymous. “We kept our hopes to a minimum after so many disappointments,” the wife said in a statement. “It took me two days to believe I was actually pregnant. I still wake up in the morning and can’t believe if this is true or not. I still don’t believe I am pregnant until I see the scans.”

Is this actually a breakthrough?

Physicians who had no role in building STAR say the concept fits AI's strengths well: spotting a rare, specific pattern buried in a huge amount of visual noise.

“The reason AI is so well-suited for this is AI really relies on learning: showing it an image of what a sperm looks like, what the shape is, what characteristics it should have, and then being able to use that learning algorithm to help identify that specific image that you’re looking for,” said Dr. Sevann Helo, a urologist at Mayo Clinic who specializes in male infertility.

“It’s very exciting. AI, in general, at least in the medical community, I think is a whole new landscape and really will revolutionize the way we look at a lot of problems in medicine.”

Dr. Aimee Eyvazzadeh, a San Francisco-based reproductive endocrinologist, called it additive rather than transformative on its own. “AI isn’t creating sperm, it’s helping us find the rare, viable ones that are already there but nearly invisible,” she said. “It’s a breakthrough not because it replaces human expertise, but because it amplifies it, and that’s the future of fertility care.”

Not everyone is convinced it is ready for wide use. Dr. Gianpiero Palermo, a professor of embryology at Weill Cornell Medicine who pioneered the sperm injection technique used across IVF labs worldwide, said AI models in reproductive medicine still need more validation.

“AI is gaining a lot of traction nowadays to offer unbiased evaluation on embryos by looking at embryo morphology. However, current available models are still somewhat inconsistent and require additional validation,” Palermo said. Of the STAR method specifically, he added: “In my opinion, this approach is faulty because inevitably some men will have no spermatozoa, doesn’t matter how their specimens are screened whether by humans or a machine.”

What to do if you've been diagnosed with Azoospermia

STAR is currently available only at Columbia University Fertility Center, and the results so far come from a handful of documented cases, not a large clinical trial. Williams has said the team plans to publish its methods so other centers can eventually offer it.

1. Ask your clinic directly whether AI-assisted sperm retrieval is something they offer or plan to offer, since it is still limited to specific centers.
2. A past azoospermia diagnosis from a manual search does not automatically rule out finding viable sperm with more sensitive methods.
3. A single viable sperm cell can be enough to attempt fertilization through IVF, so a very low count is not the end of the process.
4. Independent experts want to see the approach validated in larger studies before it becomes standard practice.