Endometriosis affects 5–10% of reproductive-aged women but faces diagnostic delays of 6–11 years, contributing to prolonged suffering and healthcare inequities. Current diagnostic approaches rely on single data sources and fail to capture the complex, multifaceted nature of this disease. This project develops a multimodal machine learning framework that fuses three complementary data streams: structured longitudinal electronic health records, patient-generated symptom reports from a menstrual-tracking application, and unstructured clinical notes. We apply large language models to extract clinical entities from free-text notes and systematically compare unimodal and multimodal architectures to understand how different data sources contribute to early risk prediction. Critically, our approach incorporates fairness auditing across demographic subgroups to address documented disparities in endometriosis diagnosis and care. By integrating these heterogeneous modalities, we aim to shorten diagnostic delays and enable earlier, more equitable interventions for all patients.

I am participating in SuperUROP because I am drawn to research where methodological innovation meets real-world impact. Women’s health remains critically underserved, with millions of patients facing years-long diagnostic journeys and fragmented data. This project gives me the chance to develop novel approaches for fusing heterogeneous data sources while working toward tools that could improve patient outcomes.