In this project, we aim to develop trajectories of the menstrual cycle by constructing artificial intelligence based digital twins of heavy menstrual bleeding.
Digital twins (DT) potentially offer a model that leverages population level data and combines it with the woman’s unique data, to deliver personalized recommendations for self-management and treatment/care plans to ensure early identification of HMB onset and can be scalable across the global space. The two-way dynamic flow of information between the physical and virtual world (the twinning) can yield more accurate digital representations of the health conditions and enable enhanced in-depth virtual testing of potential behavioral, social, and/or medical interventions.
We will generate a menstrual cycle DT via deep phenotyping, or the integration and automated processing of data from a wide range of factors related to the menstrual cycle. Some of the menstrual factors included (as available) pregnancy test, complete blood count including hemoglobin, serum iron, ferritin or total iron-binding capacity (TIBC), hormonal profile to include thyroid function tests, follicle-stimulating hormone (FSH), luteinizing hormone, estrogen, progesterone, testosterone. Additionally, we would also need information (as available) such as age at menarche, age at menopause, length of menstrual cycle, use of hormone-replacement therapy, use of oral contraceptive pills, etc.
We will approach DT generation using two different AI methods: first, a generative model using variational autoencoder (VAE) and second, using an agentic-AI approach coupling a large language model (LLM) with retrieval-augmented generation, and incorporate causal analysis and uncertainty estimation to determine statistical power. For verification and validation of the digital twins, we will combine the UK Biobank data with data from additional sources such as from the Institute for Health Metrics (IHM, USA), and the AllofUs project (NIH, USA).
