AI for Endometriosis Diagnosis: How PelviQuest Works, What It Can and Cannot Do

The problem the model is built to solve

Endometriosis has no blood test, is often invisible on imaging, and was for decades confirmed only by surgery. The result is an average diagnostic delay of 7 to 11 years. In 2019 a group of endometriosis specialists published a "call to action" arguing that the disease should be diagnosed clinically, from symptoms and examination, the way most chronic diseases are. Current NICE and ESHRE guidelines agree.

The obstacle to clinical diagnosis is consistency. Which symptoms matter most, how they combine, and how much weight a given examination finding carries are learned over years of specialist practice and are not available to most primary care clinicians or general gynecologists. PelviQuest exists to make that pattern recognition reproducible.

What goes in

• A standardized patient questionnaire covering menstrual pain, non-cyclical pelvic pain, pain with sex, bowel and bladder symptoms, fatigue, back and leg pain, menstrual history, fertility and family history, with each question phrased and scored the same way for every patient.
• A structured physician examination recorded in the clinician app: findings such as tenderness, nodularity behind the uterus, uterine mobility and pelvic floor tone, captured as discrete fields rather than free text.
• Daily symptom tracking over time, which adds the dimension a single appointment cannot capture: how symptoms move with the cycle.

Where the training data comes from

The model was developed from the surgical practice of PelviQuest’s founding clinicians, endometriosis and pelvic nerve surgeons. Patients completed the standardized questionnaire and were examined using the structured protocol before surgery. Their intraoperative findings, whether endometriosis was present, where, and how extensive, were then matched to their pre-operative answers. This pairing of symptoms and examination with surgically confirmed outcomes is what allows the model to learn which patterns predict disease.

The dataset reflects one surgical practice in Hackensack, New Jersey. That is both its strength, because every case was confirmed by an experienced surgeon using a consistent protocol, and its principal limitation, discussed below.

What comes out

The model produces a 0 to 100 score that shows how closely the answers match symptom patterns commonly reported with endometriosis. In the patient app this is presented as a result that explains which symptom domains contributed and what to do next, together with a report you can view as a PDF and bring to any appointment. In the clinician platform the same report appears alongside the questionnaire and examination findings as one input among several. The score is a screening aid, not a diagnosis, and it has not been validated as a diagnostic test. The score does not change a clinician’s responsibility to evaluate the patient; it changes how early the right questions get asked.

What PelviQuest does not do

• It does not diagnose endometriosis. Diagnosis is a clinical decision made by a licensed clinician, and surgical confirmation remains the reference standard.
• It does not detect endometriomas, deep nodules or ureteral involvement. Those require ultrasound or MRI by trained operators.
• It does not rule out other causes of pelvic pain. Adenomyosis, pelvic floor dysfunction, bladder pain syndrome and bowel disease frequently coexist and need their own evaluation.
• It does not replace the relationship between a patient and her doctor. It is designed to make that conversation start sooner and go better.

Limitations we think patients and clinicians should know

• Single-practice data: the model has been trained on patients from one specialist practice. Patients who reach a tertiary endometriosis surgeon differ from the general population, and performance in other settings has not yet been established.
• External validation is ongoing: PelviQuest is actively seeking partnerships with endometriosis surgeons worldwide to enrich the dataset and test the model on independent populations. Until then, the score should be read as decision support from one expert’s experience, systematized.
• Superficial versus deep disease: the score reflects how closely symptoms match a pattern, not the stage or location of disease.
• A low score is not an all-clear. Women with persistent symptoms should be evaluated regardless of any score.

How this compares with other approaches

Researchers have pursued several routes to non-invasive diagnosis. Biomarker studies in blood, urine and saliva have so far not produced a test accurate enough for clinical use, and guidelines recommend against relying on them. Imaging research is improving detection of deep disease but cannot see superficial lesions. Symptom-based prediction, the approach PelviQuest takes, has the advantage of using information every patient already has. Agarwal and colleagues (2019, cited below) argue that a structured symptom history and examination can support a clinical diagnosis of endometriosis without waiting for surgery. Academic symptom-tracking projects have shown that patient-reported data at scale is informative; PelviQuest differs in combining patient data with physician examination and surgical confirmation in a clinical workflow.

Privacy

Health information entered into PelviQuest is encrypted in transit and at rest, access is role-based and audited, and the platform is designed to meet HIPAA requirements. Research use is limited to de-identified or aggregated data. The symptom check on this website is separate from the app: it runs entirely in your browser and stores nothing.

What to do with a PelviQuest result

Treat it as a reason to act, not a verdict. Take the report to a gynecologist, ideally one experienced in endometriosis, and ask for a clinical evaluation including examination and transvaginal ultrasound. If your clinician is unfamiliar with clinical diagnosis of endometriosis, the NICE and ESHRE guidelines cited below are a useful starting point for that conversation.