BRENNEN

Industries / Health / Healthcare / Fertility clinics

The embryo lab

The laboratory work between egg collection and embryo transfer: joining sperm and egg, growing the embryos for up to six days, choosing which to transfer, and keeping every dish matched to the right patient.

7

Milestones achieved, from 1978 to 2025.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1978

    First baby from IVF

    In vitro fertilization (IVF) joins egg and sperm in a dish outside the body. Patrick Steptoe, a gynecologist, collected eggs by laparoscopy (keyhole surgery), and Robert Edwards fertilized them in culture and grew them into early embryos. In 1977 Lesley and John Brown came to the team's clinic in Oldham, England, after nine years of trying for a child. An eight-cell embryo was returned to Lesley Brown, and Louise Brown was born on 25 July 1978; Steptoe and Edwards reported the birth in a short letter to The Lancet. Jean Purdy ran the laboratory, was the first to see that embryo divide, and co-wrote 26 papers with Edwards; her name joined theirs on a Royal Society of Biology plaque only in 2015. Edwards received the 2010 Nobel Prize in Physiology or Medicine.

    nobelprize.orgeuropepmc.orgblog.sciencemuseum.org.uk

    Credit

    Robert Edwards, Patrick Steptoe and Jean Purdy

  2. 1992

    One sperm into each egg

    Intracytoplasmic sperm injection (ICSI) pushes a single sperm into the egg with a fine glass needle. At the Centre for Reproductive Medicine of the Vrije Universiteit Brussel, a team injected 47 eggs from couples whose sperm was too poor for standard IVF; 31 were fertilized, and four babies were born from three pregnancies. It let men with very few working sperm have genetic children. Use then grew worldwide, mostly in couples without male infertility. A randomized trial of 1,064 such couples at two clinics in Ho Chi Minh City, Vietnam, published in 2021, found live births after the first transfer in 35% with ICSI and 31% with standard IVF, a difference within the range of chance (95% confidence interval −2.4 to 9.2 percentage points).

    LessonA method built for one problem spread to patients it was not shown to help.

    europepmc.orgeuropepmc.org

    Credit

    G. Palermo, H. Joris, P. Devroey and A. C. Van Steirteghem; the trial by V. Q. Dang, B. W. Mol and colleagues

  3. 1998

    Embryos grown to day five

    Embryos were usually transferred on day 3, at about the eight-cell stage, and clinics put back several to make up for low odds. A team grew embryos in two serum-free culture media used one after the other, each matched to a stage of development, to the blastocyst, the stage reached on day 5. In 23 patients at a private clinic, day-5 blastocysts implanted at twice the rate of day-3 embryos, so fewer embryos were needed for the same pregnancy rate, which lowers the risk of twins and triplets.

    europepmc.org

    Credit

    D. K. Gardner, P. Vella, M. Lane, L. Wagley, T. Schlenker and W. B. Schoolcraft

  4. 2004

    Two people check each step

    Before July 2002 four adverse incidents occurred in the reproductive medicine units of the Leeds Teaching Hospitals in England, among them a mix-up after which a white couple had mixed-race twins. Professor Brian Toft's inquiry, published on 22 June 2004, found "inadvertent human error and systems failure". The UK regulator, the Human Fertilisation and Embryology Authority (HFEA), had by then introduced double checking of patient identity at every stage of treatment, unannounced inspections and an alert system to share lessons between clinics. This check by a second person, called witnessing, is now standard, by hand or with barcode and RFID (radio) tags. In 2015 the HFEA counted 517 reported incidents in about 72,000 treatment cycles, and for the first time none was grade A, the class that includes an embryo transferred to the wrong patient.

    LessonMix-ups come from how a lab's steps fit together, so the fix was a check at every step.

    api.parliament.uknews-medical.netifqlive.blob.core.windows.neteuropepmc.org

    Credit

    Brian Toft; the HFEA

  5. 2024

    Time-lapse fails its trial

    Time-lapse incubators photograph each embryo repeatedly without taking it out of the incubator, and software can rank embryos by the timing of their cell divisions. In the TILT trial, 1,575 participants at seven IVF centers in the UK and Hong Kong were randomly assigned, between June 2018 and September 2022, to time-lapse culture with time-lapse selection, time-lapse culture alone, or standard care. Live birth rates were 33.7%, 36.6% and 33.0%, and neither time-lapse group differed significantly from standard care.

    LessonA clearer view of embryos did not by itself produce more babies.

    europepmc.org

    Credit

    Priya Bhide and colleagues, Queen Mary University of London and The Chinese University of Hong Kong

  6. 2024

    AI picks no better

    iDAScore is a deep learning program, a kind of AI, that scores time-lapse videos of embryos; its maker, Vitrolife, holds a pending patent on it, and two Vitrolife employees were among the authors. In a randomized, double-blind trial at 14 IVF clinics in Australia and Europe, 1,066 women under 42 had their embryo chosen either by iDAScore or by embryologists' standard grading. Clinical pregnancy rates were 46.5% with the AI and 48.2% with embryologists, and the trial could not show that the AI was within its preset margin of 5 percentage points.

    LessonRanking embryos well on past records is not the same as producing more pregnancies in a trial.

    europepmc.org

    Credit

    Peter J. Illingworth of Virtus Health and colleagues

  7. 2025

    A swap found years later

    Monash IVF, an Australian clinic group listed on the Australian stock exchange, revealed in 2025 that a patient at its Brisbane clinic had been given another patient's embryo and had given birth to a baby with no genetic link to her, in 2023 according to the ABC. In June 2025 it disclosed a second error: a patient's own embryo was transferred instead of her partner's, contrary to the treatment plan. Chief executive Michael Knaap resigned on 12 June 2025.

    LessonWitnessing rules did not stop a swap two decades after Leeds, and this one came to light only after the birth.

    rnz.co.nz

Watch for

  1. Robots handle eggs and sperm

    Conceivable Life Sciences, a New York company, reports that its automated systems, which it calls Pearls, prepared sperm, found eggs and stripped the cells around them, and injected sperm by ICSI in 11 cases at Hope IVF México in Guadalajara. In that study 64.3% of injected eggs were fertilized and five healthy babies were born, but staff still had to step in: the egg-finding robot found 92 of 110 eggs (84%), took about 76 seconds per egg and lost two. A follow-up adding automated freezing reports 16 live births among 26 patients, while its injection robot fertilized 59.6% of eggs against 80% for embryologists working on sibling eggs; two extremely preterm births, which the authors link to pre-eclampsia and infection, ended in the babies' deaths. The studies were company-sponsored and most authors are Conceivable employees or shareholders. In 2023 Overture Life of Barcelona reported two babies from its own injection robot.

    Achieved whena clinic with no financial tie to the maker runs a randomized comparison on sibling eggs in a few hundred patients and publishes fertilization and live birth rates and how often staff had to intervene.

    pmc.ncbi.nlm.nih.goveuropepmc.orgeuropepmc.org

    Claimed by

    Conceivable Life Sciences, December 2025 and October 2026

  2. Babies by mitochondrial donation

    Mitochondria, the parts of a cell that supply its energy, carry a little DNA of their own, and women who carry faults in it can have children with severe, often fatal disease. In pronuclear transfer, the parents' main DNA is moved from the affected woman's fertilized egg into a donor's fertilized egg whose own main DNA has been removed. A team at Newcastle University and the Newcastle upon Tyne Hospitals in England reports that 22 women have begun or completed pronuclear transfer and 8 children have been born, all healthy at birth, with the faulty DNA in their blood ranging from undetectable to 16%. One child developed high blood fats and an irregular heartbeat that responded to treatment, and another had infant epilepsy that went away.

    Achieved whenan independent group publishes the children's health and mitochondrial DNA levels over several years, or another licensed center reports births by the same method.

    europepmc.orgeuropepmc.org

    Claimed by

    Robert McFarland, Louise A. Hyslop and colleagues, July 2025

Open

  1. 01

    A camera that checks labels

    A camera station over an embryologist's bench that reads the label on every dish, tube and freezing straw as it is picked up, compares it with the patient on the current work order, and flags any mismatch. Test it only on mock items: empty dishes, tubes and straws labeled for made-up patients and handled by volunteers at a mock bench, with no real patient names, samples or embryos.

    The brief
    Credit · unclaimed
  2. 02

    AI embryo tools, checked

    A public table of every AI embryo-assessment product on sale: its maker, what it scores, its approval status in the US, the EU, the UK and Australia, and its best evidence, graded as randomized trial, independent study or the maker's own study, with a link for every cell. It is updated every three months, and each maker is invited to correct its row.

    The brief
    Credit · unclaimed

Open 01 · the brief

A camera that checks labels

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A camera station over an embryologist's bench that reads the label on every dish, tube and freezing straw as it is picked up, compares it with the patient on the current work order, and flags any mismatch. Test it only on mock items: empty dishes, tubes and straws labeled for made-up patients and handled by volunteers at a mock bench, with no real patient names, samples or embryos.

Why it is still open

Labs check identity by a second person or by barcode and RFID tags, and a 2026 committee opinion from the American Society for Reproductive Medicine (ASRM) and the Society of Reproductive Biologists and Technologists still lists many points in an IVF cycle where samples can be misidentified. Electronic witnessing cut witnessing time 3.1- to 5.2-fold in one private laboratory network, but every item must carry a tag. A 2022 study by K. C. Hammer, C. L. Bormann, H. Shafiee and colleagues matched embryos to patients from their images alone, which tracks embryos, not the labels on dishes, tubes and straws. A search in November 2026 found no published test of a camera that reads ordinary printed and handwritten labels at the IVF bench and checks them against the work order.

europepmc.orgeuropepmc.orgeuropepmc.org

Built with what exists

A phone or two USB cameras on a stand; AI models that read images, such as Claude, Gemini and GPT, and open-weight vision models such as Qwen2.5-VL; Tesseract OCR and the ZXing barcode reader as plain baselines; a short script holding a made-up work order. The advance is AI models that read printed and handwritten text in a photo and check it against a list in one step.

  1. First step

    Print labels for ten made-up patients, label 50 empty culture dishes, 50 tubes and 50 freezing straws, film 200 handling events at a desk, and score one model.

  2. Done looks like

    2,000 filmed handling events by at least five volunteers, with a planted mismatch in 10% of them (wrong name, wrong ID number, swapped dish or straw), keyed by two people. The system catches at least 99% of planted mismatches with no more than one false alarm per 100 events, and the video, key and code are public under an open license.

Open 02 · the brief

AI embryo tools, checked

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of every AI embryo-assessment product on sale: its maker, what it scores, its approval status in the US, the EU, the UK and Australia, and its best evidence, graded as randomized trial, independent study or the maker's own study, with a link for every cell. It is updated every three months, and each maker is invited to correct its row.

Why it is still open

The large randomized trials of choosing embryos by image, TILT for time-lapse and the 2024 iDAScore trial for AI, found no gain. A 2026 head-to-head test of six commercial AI programs on 1,000 genetically tested blastocysts at one clinic anonymized the products, so a clinic or patient cannot tell which program scored how. The FDA's database already lists six decisions under its code for embryo image assessment systems (PBH): Auxogyn's Eeva twice in 2014, Vitrolife's KIDScore D3 in 2019, Fairtility's CHLOE BLAST in August 2025, Alife Health's Embryo Predict in May 2026 and Future Fertility's MAGENTA BLAST in September 2026; it does not say which trial backs each, and other tools, such as iDAScore, are sold in Europe without a US clearance. A Europe PMC search in November 2026 found systematic reviews and that anonymized comparison, but no public product-by-product table of approvals and trials.

europepmc.orgapi.fda.goveuropepmc.orgeuropepmc.org

Built with what exists

The US Food and Drug Administration's device databases through the openFDA API; the EU's EUDAMED database; Australia's ARTG register; the UK's MHRA device register; the ClinicalTrials.gov and Europe PMC APIs for trials and papers; an AI research agent to draft rows, with every cell checked by a person against its link. The advance is AI agents that read regulatory records and papers and fill a table with a citation for each cell.

  1. First step

    List the AI embryo products named in papers and makers' websites, and look up the first five in the FDA and EUDAMED databases.

  2. Done looks like

    At least 20 products, each with every field filled or marked "not found", every filled cell linked to a public record, two reviewers agreeing on every row, and two quarterly updates published under an open license.