Industries / Health / Healthcare / Fertility clinics
Testing
Finding out why a person or couple has not conceived: checking the sperm, whether the fallopian tubes are open, and how many eggs the ovaries still hold.
6
Milestones achieved, from 1920 to 2022.
1
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1920
Checking tubes without surgery
The fallopian tubes carry the egg from the ovary to the uterus, and a blocked tube stops sperm and egg from meeting. In April 1920 Isidor C. Rubin published in the Journal of the American Medical Association a way to tell whether the tubes are open without an operation, titled "Nonoperative determination of patency of fallopian tubes in sterility".
Credit
Isidor C. Rubin
2010
Semen norms from fertile men
The World Health Organization (WHO) first published its laboratory manual for semen in 1980 so that labs would measure semen the same way, and revised it in 1987, 1992, 1999 and 2010. For the 2010 edition, T. G. Cooper and colleagues pooled samples from more than 4,500 men in 14 countries and set lower reference limits from men whose partners had conceived within 12 months: 15 million sperm per milliliter, 32% swimming forward and 4% of normal shape. The sixth edition, of July 2021, adds that these cut-offs "do not represent distinct limits between fertile and subfertile men".
LessonA range drawn from fertile men describes a population; it is not a pass-or-fail line for one man.
Credit
WHO; T. G. Cooper and colleagues
2016
A sperm test by phone
On 8 November 2016 the US Food and Drug Administration (FDA) cleared the YO Home Sperm Test for sale without a prescription. A clip holds a slide over the phone's camera, and an app reports the concentration of moving sperm as "low" (under 6 million per milliliter) or "moderate/normal", and advises seeing a doctor if two tests read low. It followed an earlier home test, SpermCheck Fertility, cleared in May 2010. The WHO's 2021 manual says phone-based devices "cannot achieve the quality and accuracy needed for a full semen analysis".
LessonA home test answers one narrow question, and a normal reading does not rule out a male cause.
Credit
Medical Electronic Systems; US FDA
2017
Egg count is not fertility
Anti-Müllerian hormone (AMH) is made by the small follicles in the ovary, so a blood test for it estimates how many eggs remain. A study in JAMA followed 750 women aged 30 to 44 in North Carolina who had just started trying to conceive. Women with low AMH (under 0.7 ng/mL) had a 65% chance of conceiving within six cycles, against 62% for women with normal levels, which was not a significant difference. The European Society of Human Reproduction and Embryology (ESHRE) now recommends AMH or an ultrasound count of small follicles to predict how many eggs stimulation will yield, and recommends against using either to predict pregnancy or birth.
LessonA test that predicts the response to fertility drugs does not predict whether a woman will conceive naturally.
Credit
A. Z. Steiner and colleagues; ESHRE
2017
Oil contrast raises pregnancies
Hysterosalpingography (HSG) is an X-ray taken while a contrast fluid is flushed through the uterus and tubes. In a trial at 27 hospitals in the Netherlands, 1,119 women were assigned at random to an oil-based or a water-based fluid. Within six months, 39.7% of the oil group had an ongoing pregnancy against 29.1% of the water group, and after five years 74.8% against 67.3% had given birth.
LessonThe test can be part of the treatment, and the choice of fluid changed the outcome.
Credit
K. Dreyer and colleagues
2022
Ultrasound foam versus X-ray
Hysterosalpingo-foam sonography (HyFoSy) checks the tubes by pushing a foam through them while watching on ultrasound, with no X-rays or iodine contrast. In a trial at 26 Dutch hospitals, 1,026 women had both tests in random order. Treatment based on HyFoSy was estimated to lead to a birth in 46% of women against 47% for HSG; the trial could not show that HyFoSy was no worse within its preset margin of 2 points. HyFoSy hurt less: 3.1 against 5.4 on a 10-point pain scale.
LessonA gentler test is not automatically an equal one; it has to be tested on births, not on how the pictures compare.
Credit
N. van Welie and colleagues
Watch for
AI finds hidden sperm
In azoospermia a man's semen contains no sperm visible under a microscope. Columbia University's STAR system scans a semen sample with a high-speed camera, uses AI to spot sperm, and isolates them on a microfluidic chip, a plastic chip with hair-thin channels, from which a robot lifts each sperm. In the first reported case, published as a research letter in The Lancet, STAR scanned 2.5 million images of a 3.5 mL sample in about two hours, found two sperm, and the two embryos made with them led to a pregnancy. The team says larger studies are under way.
Achieved whena multi-patient study, ideally at a center outside Columbia, reports how often STAR finds sperm compared with a manual search or surgery, and the births that follow.
Claimed by
Columbia University Fertility Center (Zev Williams, Hemant Suryawanshi and colleagues), October 2025
Open
01
Ovulation tests on known standards
A bench test of 20 brands of home ovulation test strips, which detect the surge of luteinizing hormone (LH) in urine that comes a day or two before ovulation. Dip strips in synthetic urine spiked with known amounts of LH, from none to 100 mIU/mL, and with the related hormones hCG and FSH to check for false positives. Two people blind to the level read each strip, and its photo is also read by an AI model. Publish the level at which each brand turns positive, beside the level printed on its box. No human urine is used.
The briefCredit · unclaimed02
An answer key for results
A public set of 500 made-up fertility test reports (semen analyses, AMH and FSH results, follicle counts, HSG and HyFoSy reports), each paired with an answer agreed by specialists: what the result shows, what it does not show, and what to ask a doctor. Score current AI models against it, counting harmful mistakes such as telling a woman with low AMH that she cannot conceive.
The briefCredit · unclaimed03
A tracker for test claims
A repeatable check of what websites selling home fertility tests (AMH, FSH, LH and sperm) say each test can do, coded against the evidence, rerun every three months in ten countries, with a public record of which claims changed and when.
The briefCredit · unclaimed
Open 01 · the brief
Ovulation tests on known standards
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A bench test of 20 brands of home ovulation test strips, which detect the surge of luteinizing hormone (LH) in urine that comes a day or two before ovulation. Dip strips in synthetic urine spiked with known amounts of LH, from none to 100 mIU/mL, and with the related hormones hCG and FSH to check for false positives. Two people blind to the level read each strip, and its photo is also read by an AI model. Publish the level at which each brand turns positive, beside the level printed on its box. No human urine is used.
Why it is still open
The closest study compared five kits with blood LH in 23 patients over 97 test days, with only 13 surges; it found sensitivity from 38.46% (Clinical Guard) to 76.92% (Pregmate) and called all five accurate. It did not test against known concentrations. The FDA's clearance records list home ovulation tests from many makers since the 1990s, but the sensitivity figures behind each one come from its own maker. No independent bench test of brands against LH standards turned up in this search.
Built with what exists
Commercial synthetic urine; recombinant human LH, hCG and FSH sold as laboratory reagents; pipettes and a phone camera; the free image tool ImageJ to measure line darkness; AI models that read images, such as Claude, Gemini and GPT. The advance is vision models that read a faint test line from a photo the same way every time.
First step
Buy five brands and one vial of recombinant LH, make four dilutions, and test five strips of each brand at each level.
Done looks like
800 strips (20 brands, 8 LH levels, 5 strips each), each read by two blinded people and one model. A public table gives each brand's measured threshold, the lowest level read positive on at least 4 of 5 strips, against its label claim, with every photo and reading released under an open license.
Open 02 · the brief
An answer key for results
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of 500 made-up fertility test reports (semen analyses, AMH and FSH results, follicle counts, HSG and HyFoSy reports), each paired with an answer agreed by specialists: what the result shows, what it does not show, and what to ask a doctor. Score current AI models against it, counting harmful mistakes such as telling a woman with low AMH that she cannot conceive.
Why it is still open
A study published in December 2025 gave two chatbots, Google's Bard and Microsoft's Bing, a set of made-up semen analyses: Bard read 50% of them correctly and Bing 29%, and the authors called for training sets built for the task. It used no patient data and covered semen only. A 2025 study in JMIR Formative Research gave four models (GPT-4, GPT-4o, Claude 3.5 Sonnet and Gemini 1.5 Pro) 10 questions from Japan's fertility specialist and embryologist exams and a data task; it did not ask them to read test reports. Mate Health launched MateIQ, a chatbot that explains home semen test results, in November 2025, and published no accuracy figures. This search found no test of how AI models read AMH results, and the models tested so far are a generation or more old.
Built with what exists
The WHO 2021 semen manual and the ESHRE 2025 stimulation guideline as the standards for the answers; a script that generates realistic reports; Claude, Gemini, GPT and open-weight models; three reproductive specialists to agree the key. The advance is AI models that read a whole lab report and explain it in plain words.
First step
Write 30 semen analysis reports that span the WHO categories, and have one specialist draft the answers.
Done looks like
500 reports with a key agreed by three specialists; at least six models scored on correct readings and on harmful statements per 100 reports; reports, key and scoring code public.
Open 03 · the brief
A tracker for test claims
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable check of what websites selling home fertility tests (AMH, FSH, LH and sperm) say each test can do, coded against the evidence, rerun every three months in ten countries, with a public record of which claims changed and when.
Why it is still open
A 2023 study in JAMA Network Open coded 27 websites in seven countries that sold AMH tests directly to consumers, using pages captured in March 2022. It found that 96% said the test showed ovarian reserve and 74% that it showed the likelihood of conceiving, claims the evidence does not support for an individual. It was one snapshot, of one test.
Built with what exists
The open-source web agent Browser Use 1.0 to find and save the pages (see the Note "Deleting 80% to start over"); a small AI model built for extraction, such as Meta's Muse Ember (see the Note "The jobs agents repeat"), to pull out each claim; the 2023 study's coding frame; two people to code a sample by hand.
First step
Rerun the 2023 study's four search strings and save the top 50 results for each.
Done looks like
At least 100 websites in ten countries captured twice, three months apart; the automated coding agrees with two human coders on at least 90% of claims in a 20% sample; every dated capture and code published.