Organoids as one layer of an endometriosis diagnostic

What the work claims

This is a prospective observational cohort record led by Region Stockholm with the Karolinska Institutet as collaborator. It plans to enroll an estimated 365 women aged 18 to 45 with suspected or confirmed endometriosis; Swedish language comprehension is an inclusion criterion, and not having a uterus is the headline exclusion. The registered primary outcome is the diagnostic accuracy of a combined model of AI-assisted transvaginal ultrasound and multi-omics biomarkers for detecting endometriosis. Secondary outcomes include the accuracy of AI-assisted ultrasound alone, the performance of plasma and endometrial biomarker panels, and then three organoid-flavored measures: implantation rate in in vitro endometrial models, molecular characteristics associated with embryo or blastoid attachment, and organoid drug-response variability across the cohort¹.

The record is not yet recruiting, with a planned start of 2026-04-20, primary completion 2028-04-20 and full completion 2029-04-20. The protocol summary argues the burden case: endometriosis affects roughly 10 percent of women of reproductive age by the record's account, with chronic pelvic pain in 70 percent of cases and infertility in up to 40 percent, while existing imaging and biochemical markers lack the specificity to replace surgical confirmation¹.

How the stack is supposed to work

The design is a diagnostic cascade with four layers. Layer one is imaging: AI reads transvaginal ultrasound scans for lesions. Layer two is molecular: proteomics, transcriptomics and immune profiling are performed on patient-derived endometrial organoids, plus plasma biomarkers, to find signatures that separate diseased from healthy endometrium. The logic of running the multi-omics on organoids rather than raw biopsy tissue is that organoids standardize the cell population: they purge blood, stroma and menstrual-cycle noise and expand the epithelial compartment that is actually being assayed, so a biomarker panel derived from them should be more reproducible than one derived from heterogeneous tissue. Layer three is mechanism: an in vitro embryo-endometrium interaction model, where blastoid attachment to endometrial organoids is read as a proxy for the receptivity defect that ties endometriosis to infertility. Layer four is therapeutics: patient-specific organoids used for drug screening and treatment-response evaluation, with drug-response variability registered as an outcome in its own right¹.

For a field dominated by cancer avatars, this is a different contract for the organoid: not a predictor of how a tumor will respond to a drug, but a biomarker factory and a standardized receptivity substrate inside a diagnostics program.

Where a skeptic should push

First, the power question. The only outcome this study is clearly designed to power is diagnostic accuracy of imaging plus biomarkers. The three organoid endpoints are secondary, and the record attaches no sample-size justification to them: with the cohort split across suspected and confirmed disease, plus however many proceed to surgical verification, the subgroup actually carrying both an organoid culture and a histologic reference standard may be far smaller than 365. Organoid drug-response variability, read across that reduced and selected population, risks producing exactly the kind of uninterpretable variance it is named after.

Second, the reference-standard problem. A diagnostic accuracy study stands or falls on its gold standard, and the eligibility criterion says "suspected or confirmed," which merges two populations. The record does not state how many participants will receive laparoscopic verification, which is the only true reference for endometriosis. If verification rates are low, the combined model's accuracy will be validated against an imperfect label, and any organoid biomarker trained on that label inherits its noise.

Third, the blastoid step. Using blastoid attachment as a receptivity readout conflates the embryo's attachment competence with the endometrium's receptivity, and it is a contested proxy at best; a molecular correlate of attachment is a long way from a qualified assay of implantation. Finally, the Swedish-language inclusion criterion and the single health region mean the cohort is demographically narrow for a disease whose presentation and access to care vary widely across populations. That is a generalization ceiling the eventual model may never be tested against.

Organoids inside a diagnostic-first design

The opportunity is easy to miss: this is one of the few prospective, non-cancer cohorts anywhere that is positioned to validate an endometrial organoid assay against a histologic gold standard, because a large fraction of participants will plausibly pass through laparoscopy. Cancer organoid programs rarely get that anchor; their avatars are validated against response, not against truth. If the team prespecifies the concordance analysis between organoid readouts and surgical findings, endometriosis could deliver the cleanest organoid-validation dataset in the field, in a disease with an enormous unmet need and no incumbent avatar competition.

The threat is structural, and it applies far beyond this trial. When organoid work rides as secondary outcomes on a diagnostics grant, it is unfunded to powered conclusions and free to drift into ambiguity; a three-year, 365-patient study can end with a validated ultrasound algorithm and three organoid datasets that no design was built to interpret. That pattern, replicated across grant portfolios, is how exploratory organoid endpoints accumulate without ever becoming assays. The discipline fix is cheap: name one organoid measure, define its reference standard and target concordance, and power it in the confirmed-disease subgroup. Nothing in the record commits the team to any of that, and the field should read the eventual organoid claims accordingly.

The bottom line

Established: endometriosis diagnosis remains invasive and delayed, and AI ultrasound plus molecular biomarkers are a reasonable bet for a non-invasive test. Hypothesis: that endometrial organoids can contribute standardized biomarkers, a receptivity model and patient-specific drug screens within the same cohort. What would confirm the organoid layer: a prespecified concordance endpoint against laparoscopic verification in the confirmed subgroup. What would break it: organoid results confined to exploratory secondary analyses with an unverifiable reference standard, which is what the current registration sets up by default.