Endo-Twin[PCOS]

Inspiration

Polycystic ovary syndrome(PCOS) can be challenging to identify due to the diversity of its manifestations. We were inspired by the opportunity to use artificial intelligence and wearable-derived data to investigate longitudinal changes in an individual's physiology, rather than focusing on isolated measurements. This led to our creation of ENDO-TWIN, a prototype that builds a personal baseline and tracks deviations over time. Our goal was to explore the possibility that continual, non-invasive signals could augment the existing approaches to women's health monitoring.

What it does

ENDO-TWIN uses physiological signals such as resting heart rate, heart rate variability(HRV), sleep duration, wrist temperature, and activity to detect changes relative to an individual's personal baseline. The prototype then calculates deviation scores and investigates if abnormal patterns persist over several days. Subsequently, ENDO-TWIN classifies the observed patterns into Stable, Monitoring, and Persistent Deviation.

How we built it

We implemented the ENDO-TWIN prototype in Python using Google Colab, Pandas, NumPy, and data-visualization libraries. We built a synthetic dataset comprising 400 simulated users observed over 180 days each to yield 72,000 observations. This enabled us to evaluate how the prototype would respond to prolonged simulated changes relative to transient disturbances. In addition, we investigated the real-world mcPHASES menstrual-cycle dataset, focusing on associations between the available wearable-derived signals, menstrual-cycle phases, and hormone measurements. Similarly, we used the hormone measurements for exploratory analysis, not as an input to the deviation-monitoring model.

Challenges we ran into

One of the challenges we faced was the need to work with longitudinal datasets that contained missing observations and uneven physiological signal coverage. For example, heart rate and sleep data were generally more widely available than HRV and wrist-temperature measurements, which limited the analyses we could perform with these variables. Furthermore, we had to differentiate between temporary disturbances and persistent deviations. We addressed this challenge in our prototype by combining personal baseline deviation scores with a rolling persistence mechanism. Finally, we had to distinguish between interesting computational findings and clinically relevant discoveries. While the synthetic data helped us to test the logical correctness of our approach, it was deliberately designed to exhibit specific patterns. As such, we cannot conclude that ENDO-TWIN can detect PCOS cases or outperform the existing clinical methods based on these results.

Accomplishments that we're proud of

We are proud of the results of transforming an idea at the intersection of biotechnology, artificial intelligence, and women’s health into a working research prototype. ENDO-TWIN processes 72,000 synthetic observations from 400 simulated users over 180 days, each capturing physiological baselines while detecting deviations. We succeeded in creating a pipeline suitable for longitudinal monitoring of physiological trends with a persistence-based state engine for individual users implemented and a dashboard for visualizing personal trends built. We also analyzed the real mcPHASES dataset to explore the associations between wearable data, menstrual-cycle phases, and hormone levels. Our primary accomplishment is bringing together a mixture of biological, data science, and artificial intelligence expertise into a prototype while also emphasizing the need for scientific validation. We are proud to have created a project that goes beyond a proof-of-concept by investigating how personal monitoring could be used in women’s health research.

What we learned

Our experience with ENDO-TWIN helped us to appreciate how personal baselines, longitudinal analysis, and persistence detection could be combined into an AI-assisted health research prototype. We learned that physiological signals can exhibit substantial inter-individual variation and that personal context should be considered when interpreting changes. In addition, we learned that synthetic data can be used to test the logical correctness of a given approach but cannot be used to establish clinical accuracy. Our exploratory analysis of the real-world data highlighted the importance of carefully addressing missing-data challenges and interpreting statistical associations correctly using clinically characterized data.

What's next for Endo-Twin[PCOS]

Our future scope includes validating this approach on larger longitudinal datasets comprising clinically characterized cases and controls. We also intend to improve missing-data handling, evaluate performance on unseen participants, and benchmark the performance of personal baselines against population-based approaches in a realistic setting. In addition, we plan to create a more accessible dashboard for exploring physiological trends. ENDO-TWIN is envisioned as a future in which women's health monitoring can become more personalized and longitudinal. Instead of attempting to replace clinical assessments, our prototype aims to investigate how persistent physiological deviations could support the identification of patterns that require further research or professional evaluation.

Disclaimer: ENDO-TWIN is a research prototype and is not intended as a diagnostic tool. It does not directly measure hormone concentrations or offer a PCOS screening. Clinical validation is required before this prototype can be used for medical screening or healthcare purposes.

Built With

• ai
• biotechnology
• data-visualisation
• datascience
• googlecolab
• healthcare
• matplotlib
• ml
• numpy
• pandas
• python
• scikit-learn