I-Corps: A machine learning model based on neural networks trained to recognize correlations and patterns that indicates possible medical complications

Suny At Buffalo, Amherst NY

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of a health and wellness software application that supports individuals with chronic illnesses such as prediabetes/diabetes, cholesterol, hypertension, infertility and obesity among others. . Many individuals have questions, preferences, and health issues that are not adequately addressed by routine Primary Care Physician (PCP) care. Based on algorithmic recommendations, the proposed technology is designed to connect users with alternative health care providers who have expertise in treating concerns such as sleep disorders, stress management, lactation, mental health, nutrition, and pelvic floor therapists. Currently, patients rely on recommendations and referrals from their PCP/OBGYN for knowledge about these types of care. In addition, corporations or insurance companies may offer the proposed product as part of their benefits packages similar to preventative programs in tobacco cessation or stress management. The core AI algorithms also may be applied to other healthcare sectors. This I-Corps project is based on the development of a set of algorithms comprising machine learning models. The models are trained to recognize correlations and patterns that could indicate possible medical complications, based on inputs such as electronic health records (e.g. genetic profiles) and IoT sensor data (e.g. oxygen level, heart rate etc.). The research focus is to explore the transformative potential of this sensor-fusion approach, with the requisite sample size of user inputs, to develop risk scores and predict evidence-based medical care pathways, specifically for maternal/post-partum users. While remaining cognizant of potential harmful recommendations generated because of dataset "blind-spots" such as those encountered in other AI applications in healthcare, these findings could then determine optimal recommendations to be delivered to users through user-friendly "nudges" leveraging advancements in behavioral psychology. Prioritizing privacy, the method of data collection would rely on user opt-in consent and a proprietary end-to-end silent authentication mechanism instead of OAuth/RSA tokens that are more vulnerable to hacking and privilege escalation. With a user's opt-in consent, raw patient data as well personalized recommendations could be made available directly to a provider to further inform care. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.