MedEase: AI-Assisted Diagnostic Workflow

Inspiration

Doctors have some of the most technically demanding and high-stakes jobs. However, often, their patient time comes at the cost of documenting each and every patient. A study of emergency room doctors revealed that over 40% of a typical 10-hour emergency room shift was devoted to data entry and 4,000 clicks of the computer mouse! That's an incredible amount of time that is taken away from monitoring patients' conditions. We believe that by reducing the time it takes for doctors to document information, we can allow them to use the time to better focus on patients, preventing misdiagnoses and increasing patient safety overall.

What it does

We created a website that allows doctors to effortlessly upload (and access existing) pdf files, which will then be fed to a Retrieval-Augmented Generation Large Language Model (RAG-LLM), producing a differential diagnosis along with important relevant medical information that can be directly input into an Electronic Health Record (EHR) system. To ensure security, we also created a login page, as well as quick access to previously uploaded pdfs. We used MongoDb to store user information, RAG vector database, and PDF storage.

How we built it

We split the tasks into two main tasks: frontend and backend. While one side created the login and document access pages, the other focused on calling APIs and reading information from pdfs. Integration from backend to frontend was done together once the tasks were created.

Challenges we ran into

The usual - syntax errors, bugs, and the like. More notably, those of us that worked on the frontend were rather inexperienced so this was a learning experience in terms of picking up Astro and Typescript. On the backend, querying the vector database from MongoDB was a difficult task.

Accomplishments that we're proud of

We're most proud of successfully integrating front-end back-end interactions (including many of the quality-of-life changes we've made, like easy access to previously imported pdfs), as well as the smooth integration of Large Language Models within the website. Overall, we're just really proud of the way that the project turned out!

What we learned

We learned to create things with an end-goal. As we create, we continually looked for the "so what?" of the project. Doing so really grounded us, allowing us to work without losing sight of the bigger picture, allowing us to better tackle the issue we are facing.

What's next for MedEase: AI-Assisted Diagnostic Workflow

Another way doctors are slowed by the existing systems is the process of ordering medicine (it takes 15 clicks of the mouse to provide a prescription!). Maybe in the future, we can look to integrate MedEase with existing APIs of existing Electronic Health Records in order to allow doctors to order medicine easier - maybe with AI-powered medicinal suggestions that can be adjusted according to the patients' needs.