EndoAware: A Machine Learning-Based Early Pre-Screening Tool for Endometriosis

EndoAware is an open-source, machine learning-driven tool that provides early pre-screening for endometriosis by analyzing user self-reported symptom data. Its core goals are to raise disease awareness, reduce diagnostic delays, and help potential patients seek professional medical advice in a timely manner. This tool does not replace professional diagnosis; it only serves as an auxiliary decision-making reference.

Endometriosis is a common but often neglected gynecological disease affecting approximately 10% of reproductive-aged women worldwide. It is characterized by the growth of endometrial tissue outside the uterine cavity, leading to chronic inflammation, pain, and infertility. Diagnostic challenges include three main issues: diverse symptoms (ranging from severe dysmenorrhea to no symptoms), diagnostic delays (an average of 7-10 years, requiring invasive examinations like laparoscopy), and insufficient awareness among the public and primary care providers (often mistaken for normal menstrual pain).

As an open-source machine learning project, EndoAware's vision is to develop an early pre-screening tool based on self-reported symptoms. Its core goals include: raising women's awareness of endometriosis symptoms; encouraging symptomatic individuals to seek professional evaluation as early as possible; and providing data-driven risk assessments, while clearly stating that it does not replace professional diagnosis.

The project adopts a supervised machine learning workflow: data collection (based on self-reported symptom questionnaires, including structured data such as dysmenorrhea severity and pain types); preprocessing (handling missing values and class imbalance); feature engineering (converting symptoms into numerical features like severity scores); model training (using classification algorithms such as random forests and logistic regression); evaluation (focusing on sensitivity to reduce missed diagnoses while balancing specificity to avoid excessive alerts).

In medical pre-screening, models need to balance sensitivity (priority to avoid missed diagnoses) and specificity (to avoid false positives). For ethics and privacy: compliance with regulations like HIPAA/GDPR to protect data; clear disclaimers (results are for reference only); ensuring algorithm fairness and avoiding bias across different populations.

The open-source model offers multiple values: transparency (public code and models, reproducible research, community review); collaborative potential (attracting clinicians, data scientists, patient advocates, and multilingual contributors to optimize the tool).

Current limitations: reliance on the quality and representativeness of training data, subjectivity of self-reported symptoms, and inability to replace professional diagnosis. Future directions: integrating wearable device data, longitudinally tracking symptom changes, providing personalized recommendations, and integrating with electronic health record systems.

Social impact: empowering patients (validating symptoms, enhancing confidence in seeking medical care, popularizing knowledge); reducing medical burden (triaging patients, reducing repeated visits, optimizing resources). Summary: EndoAware is an important application of AI-assisted healthcare. Open-source ensures trust, and AI's role is auxiliary rather than substitutive. It is expected to play a greater role in women's health in the future.