EqualCare
Inspiration
Unconscious bias infiltrates nearly every aspect of our lives. Unfortunately, this bias extends into medical research and, consequently, into the datasets that influence and train algorithms. Algorithms are the heart of modern healthcare decisions, influencing diagnoses, treatments, and patient outcomes. However, when these algorithms are trained predominantly on imbalanced medical data, they unintentionally reinforce existing disparities that disproportionately affect marginalized patients.
Historically, women's representation in clinical research has been sparse, notably influenced by policies such as the FDA's 1977 outright ban on women’s participation in clinical trials that was not overturned until 1993. This exclusion has ongoing repercussions to date. Despite women making up half of the population, doctors and algorithms are still trained on male-centric data, leading to a disparate impact where a seemingly neutral practice has an implicitly negative effect on a particular group. Today, according to the British Heart Foundation, women remain 50%*more likely than men to be misdiagnosed during a heart attack. Similarly, Jennifer Kent, MD, notes that 1 in 3 women are misdiagnosed during a stroke. Additionally, according to Neurology.org women are twice as likely to develop Alzheimer's disease after a certain age, yet they represent only a third of the participants in certain medical studies.
Our teammate, Anita, did previous research on this inequality. Her research, analyzing extensive datasets on heart attack, stroke, Alzheimer's disease, depression, and autoimmune diseases from sources including data.gov and the IEEE, found that women constitute a mere 35% of participant data, leaving a substantial gender gap in clinical research. Her research was chosen to be presented at this conference as well as a poster presentation.
What it does
Our project allows users to upload healthcare datasets and instantly receive a visual analysis of gender imbalance. Users also receive an AI explanation of the risks associated from using biased datasets for training AI models as well as possible solutions. Using a sleek UI and robust backend we created a user-friendly experience for analyzing healthcare datasets. Users can upload several datasets and get an analysis of the gender distribution for each one. Furthermore, the user receives a bias score that ranges from significant bias to no bias for each dataset There is also a chart to represent the data in an informative and visually appealing way and a working backend to save the data.
We aim to address inequity in healthcare through intentional and systematic changes. By highlighting the importance of inclusive data collection we can ensure accurate modeling with our project. Giving professionals and industry leaders an easy way to check for equitable data before using it to train algorithms, we help create a more inclusive environment in healthcare.
How we did it
Our frontend tech stack is built using React, making it modular and easy to work with. We use components like GraphView for rendering interactive charts with Chart.js, and DataSummary to display key stats. The UploadForm handles file inputs with a drag-and-drop interface and connects to a Python-based backend via Axios for processing and analysis.
On the backend, we use FastAPI, Openrouter, ChromaDB, Langchain, Superbase, and Pydantic. We utilize these libraries to analyze and parse data as well as store user data and train our AI. We intentionally emphasized a responsive design and asynchronous API calls for scalable and maintainable code.
Challenges we faced
During the coding process, we ran into certain issues that we resolved as a team. Most of our team did not have any experience with React and a few of the backend libraries such as FastAPI. Our teammates Justin and Ethan were so patient and helped us learn how to fetch and process backend data. This teamwork helped speed up our programming exponentially, as we finished the functionality of the project on Wednesday and focused on UI and the video for the rest of the contest time.
Accomplishments we are proud of
As a team we are eternally grateful for the opportunity to face this challenge. Learning from each other and overcoming initial issues brought us closer together and subsequently helped us become better programmers. We are proud of each other for committing to this project despite our fears and inhibitions to create a web application that we are extremely happy with. Personally, a member of our team has faced challenges in healthcare as a woman and we are thrilled to be able to try and tackle this issue.
What we learned
We learned that the success of a program relies heavily on the team that creates it. Without each other we would not be able to bring this idea to life. Anita provided the research and general idea while Ethan and Justin helped the rest of the team learn how to connect frontend to backend data. This was Yulian's first programming experience and he is grateful to have a team that helped him learn the ropes of coding. We learned a lot of invaluable hands on experience is solving a real world problem. The goal of our project is to help pave the way for a healthcare system that is truly inclusive. By making sure algorithms are trained on bias free data, we can ensure the future of AI is responsible.
Log in or sign up for Devpost to join the conversation.