A machine learning approach towards endometriosis screening using infrared spectra of urine

Endometriosis diagnosis is challenging due to non-specific symptoms that overlap with other gynaecological conditions. This study proposes a non-invasive Machine Learning (ML) ‒ based urine test using Attenuated Total Reflection Fourier Transform Infrared (ATR-FTIR) spectroscopy for rapid, high-throughput screening. A total of 302 symptomatic patients presenting with pelvic pain and MRI referral indications were recruited. After applying exclusion criteria, 100 patients (50 endometriosis-positive, 50 endometriosis-negative with other gynaecological conditions) were included. Urine samples were self-collected during the first visit and analysed via ATR-FTIR spectroscopy. Two Machine Learning (ML) algorithms, sensitivity-tuned and specificity-tuned, were developed using ∼1,700 spectral variables per patient to prioritize either sensitivity or specificity. There were no statistically significant differences in patient characteristics between groups, as patients with negative results for endometriosis presented with other gynaecological disorders. The sensitivity-tuned algorithm achieved 93 % sensitivity and 57 % specificity, while the specificity-tuned version reached 93 % specificity but only 27 % sensitivity. Given an endometriosis prevalence of 30 % in symptomatic population, the sensitivity-tuned test reduced unnecessary MRI referrals by 42 %, prioritizing patients most likely to have endometriosis. The analysis time was 40 s per replicate, enabling same-day results. This proof-of-concept study demonstrates the clinical potential of a rapid, urine-based ML test to reduce diagnostic delays and imaging costs. Validation in larger, multi-center cohorts is underway to enhance robustness and generalizability.