Obstetrics_gynecology · IVF embryo assessment (time-lapse)
CHLOE BLAST
Fairtility Ltd.
CHLOE BLAST brought AI embryo assessment inside the FDA's perimeter: the August 2025 clearance — Fairtility's first, tracing its predicate to the 2014 Eeva De Novo — validated blastocyst-formation prediction on 1,726 embryos at an AUC of 0.88 overall, falling to 0.81 for normally fertilised embryos and 0.74 for the Good/Fair-morphology embryos the device is actually indicated for. In the reader study the adjunct sharpened embryologists' discrimination (odds ratio 5.67 against 3.77) mainly by raising specificity, while embryo-level sensitivity slipped. Two honest limits hold it at one mark: the load-bearing evidence is manufacturer-run validation and reader studies rather than independent prospective outcome trials, and the wider claims of the CE-marked European product (implantation and ploidy prediction, embryo ranking) sit outside what the FDA reviewed. Cleared for one year in the US and restricted to EmbryoScope version D images; the durable question — whether AI-guided selection improves live-birth rates — remains open in the peer-reviewed literature.
Performance Metrics
Clinical Evidence
The FDA evidence comes in two layers, both manufacturer-run and both described in the K243851 510(k) summary. The first is a standalone algorithm validation on 1,726 embryos from one US and one Norwegian clinic, annotated by three embryologists blinded to one another and to the model. Overall blastocyst-formation prediction reached an AUC of 0.88 (95% CI 0.86–0.90), clearing the pre-specified acceptance criterion of a lower bound above 0.8. Performance fell in the narrower populations that matter clinically: 0.81 (0.78–0.83) for embryos with two pronuclei, and 0.74 (0.69–0.78) for embryos graded Good/Fair — the population the device is indicated to help select from. The obese-BMI subgroup failed the acceptance criterion for blastocyst prediction; for the underlying morphokinetic-event detection (accuracy 0.82 on 1,094 embryos) two age groups (under 35 and 41 and over) and two BMI groups (underweight and obese) also missed it. A separate reproducibility test on optically augmented videos returned AUCs above 0.89; that figure measures robustness to acquisition artefacts, not diagnostic performance, and should not be read as the validation result. The second layer is a masked multi-reader study (NCT06111365; a pilot was registered as NCT05455281) on 703 archived embryos from 59 patients at three US sites, treated in 2020–2021. Five panelist embryologists predicted blastocyst formation from day-3 morphology alone and, after a two-week washout, with the CHLOE score added. On the primary endpoint — Good/Fair embryos at embryo level — the odds ratio linking prediction to outcome rose from 3.77 (2.97–4.79) to 5.67 (4.60–6.99). The gain came from the specificity side: [[specificity]] rose from 0.246 to 0.444 while [[sensitivity]] fell from 0.893 to 0.846; PPV moved from 0.569 to 0.629 and NPV from 0.675 to 0.721. The frequently quoted subject-level sensitivity (80.36–83.93 percent unaided against 87.50–92.86 percent with CHLOE, 56 patients with at least one blastocyst) uses a per-patient definition in which a "hit" among a patient's cohort counts as a true positive, and the summary notes that subject-level specificity could not be calculated because too few patients had every embryo arrest. Odds ratios exceeded 1 in every age, BMI and site subgroup, though the confidence-interval lower bound crossed 1 for the 41-and-over and underweight groups, and the race and ethnicity analysis was too small to support conclusions. The peer-reviewed record is real but manufacturer-weighted. The core algorithm paper (Erlich et al., 2022, PMID 35169194) is authored by Fairtility; the blastocyst-morphometrics study associating CHLOE-derived parameters with implantation potential (Fruchter-Goldmeier et al., 2023, n=608, PMID 37669976) is manufacturer-affiliated. Independent groups have begun using CHLOE as an annotation instrument — the GeneraLife network's cytoplasmic-strings study (Marconetto et al., Human Reproduction 2024, PMID 39354750) used it to register morphokinetic timings — which speaks to adoption, though it is not an independent validation of the prediction claim. No randomised trial has tested whether CHLOE-guided selection improves clinical pregnancy or live-birth rates, which remains the field-wide gap for AI embryo selection generally.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| FDA algorithm validation — standalone blastocyst-prediction test (manufacturer; K243851 510(k) summary §8.1) | RetrospectiveRetrospective | 1,726 | not reported (AUC-based validation) | not reported (AUC-based validation) | 0.88 (95% CI 0.86–0.90) overall; 0.81 (0.78–0.83) 2PN embryos; 0.74 (0.69–0.78) Good/Fair embryos | FDA 510(k) summary K243851, 2025; independent test set from one US and one Norwegian clinic, three blinded annotators; obese-BMI subgroup missed the 0.8 lower-bound criterion |
| FDA pivotal masked multi-reader study, five panelists (manufacturer; K243851 510(k) summary §9) | RetrospectiveRetrospective | 703 | 0.846 with CHLOE vs 0.893 morphology alone (embryo level, Good/Fair embryos); subject level 87.50–92.86% vs 80.36–83.93% (56 patients) | 0.444 with CHLOE vs 0.246 morphology alone (embryo level, Good/Fair embryos); subject level not calculable (too few all-arrested patients) | — | FDA 510(k) summary K243851, 2025; NCT06111365; 703 embryos from 59 patients, three US sites, 2020–2021 cycles; primary endpoint OR 5.67 (4.60–6.99) vs 3.77 (2.97–4.79). The summary's §9.1 calls this a 'prospective assessment study' while §9.5 calls it 'retrospective, de-identified' and ClinicalTrials.gov registers a retrospective time perspective — recorded here as retrospective |
| Fruchter-Goldmeier Y, Kantor B, Ben-Meir A, et al. (manufacturer-affiliated) | RetrospectiveRetrospective | 608 | not reported as single figure | not reported as single figure | — | Sci Rep 2023; CHLOE-automated blastocyst morphometrics vs implantation potential |
Inside the algorithm
Editorial featureHow CHLOE BLAST calls day 5 from day 3.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
Time-lapse incubator video streams in from the IVF lab.
CHLOE BLAST works on the continuous video record that a time-lapse incubator captures of each embryo in culture — thousands of frames from insemination to day 3, acquired without removing the embryo from its controlled environment. The cleared indication restricts the input to images captured by the EmbryoScope version D incubator system; the FDA validation exercised the algorithm on those acquisitions, including robustness checks against the mechanical and image-capture artefacts that system produces.
The input is the embryo's whole developmental history to date, not a single snapshot.
Input
Time-lapse embryo video
Inference
Per-embryo, at cleavage stage (day 3)
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
Safety Record
No adverse-event report, recall or enforcement action for CHLOE or Fairtility was found in openFDA's MAUDE, recall and enforcement endpoints when queried on 3 September 2026. The US clearance is only a year old, so the post-market record is short by construction. The FDA reader study was image-based and non-interventional — the device was not used in treatment — so no safety endpoints were collected. The harm surface is indirect: the software predicts a developmental outcome and does not act on the embryo; a wrong prediction could contribute to a suboptimal selection or discard decision, which is why the cleared role is adjunctive and the embryologist's grading remains primary. The validation's weaker performance on Good/Fair embryos and the obese-BMI subgroup miss are the relevant caveats for a laboratory weighing how much to lean on the score.
Intended Use & Indications
CHLOE BLAST runs against time-lapse incubator video of embryos in culture. The software automatically annotates the morphokinetic events of early development and produces, at the cleavage stage (day 3), a "CHLOE Score" — a prediction of whether each embryo will go on to form a blastocyst by day 5, the developmental milestone that ordinarily gates transfer or freezing. The output is adjunctive: the cleared indication states that it is to be used only after the embryologist has completed an independent morphological assessment, when several normally fertilised embryos are already deemed suitable for transfer or freezing, and the laboratory retains full responsibility for which embryos are selected. The FDA clearance (K243851) restricts the device to the analysis of images captured by the EmbryoScope version D incubator system; this is a condition of the labelling, not merely the validation set. The same underlying platform is marketed in Europe as CHLOE EQ with a broader CE-marked scope — embryo viability assessment supporting blastulation, implantation and ploidy prediction and embryo ranking — which the US clearance does not cover; the two scopes should not be conflated.