-
-
Appointment Scheduler (from healthcare side)
-
Symptom Description Form (from patient side)
-
Accuracy of Model - Training & Testing
-
Appoint AI Logo
Inspiration
When we talked to doctors when doing our early research, we found that an appointment triaging system would dramatically improve 3-6 month wait times and reduce administrative burden on healthcare professionals. Thus, we decided to create Appoint AI.
What it does
Appoint AI performs healthcare appointment triaging based on patient urgency. Patients describe their symptoms, then a deep learning model determines how urgent their condition is based on their symptoms, and lastly, an urgency score is assigned to the patient. How early the patient gets an appointment is based on how urgent their condition is. This prioritizes patients with severe or life-threatening conditions, ensuring that they can see a medical professional as early as possible to diagnose/address their condition, instead of having to wait months for an appointment.
How we built it
We created 10,000 rows of synthetic, AI-generated data of patient-described symptoms, and we used a sentiment-analysis model to assign urgency scores to each symptom description. (Note: since no healthcare data is open-source, we had to generate AI data for our project. However, when integrated into healthcare systems, the model will train on their existing data, eliminating the need to use the "placeholder" AI-generated data.) A deep learning/neural network was trained on this data (accuracy: 86%) and tested on a smaller dataset (accuracy: 84%). (Note: we can expect the accuracies to be higher when the model is trained on actual medical data instead of AI-generated data.)
Our project utilized two webpages: (1), a form where patients can input their basic information and their symptom descriptions. Our deep learning model was used to evaluate the urgency score for each patient that submitted the patient form. (2), a site for the healthcare professional where the list of all patients and their urgencies is displayed. The scheduler automatically schedules the earliest appointments for the patients with the highest urgency. The scheduler can then send text messages to each patient asking them to confirm their appointment.
Challenges we ran into
A challenge we ran into was connecting the frontend to the backend. We quite easily wrote Python scripts for the backend and HTML/CSS/JS scripts for the frontend; however, we spent quite some time linking the two, to ensure that the model ran on the real-time patient data and the most recent data was reflected on the healthcare site.
Accomplishments that we're proud of
It gives us massive pride to run our program end-to-end, recognizing the fact that we have created an entire application (generating data, backend, frontend) during a single weekend. We are optimistic that Appoint AI can actually be utilized in healthcare settings to both increase patient safety AND reduce administrative burden on healthcare professionals.
What we learned
This hackathon taught us all a lot about the components involved in creating an end-to-end application. We learned how to generate data and implement a frontend and backend, and we also learned how to link all of those components together so that our overall application works as intended. We also learned the importance of outlining our project before we began coding, since knowing how components would be linked together helped us with the pipeline visualization.
What's next for Appoint AI
The next steps for Appoint AI would be to find a hospital system, such as Emory, Grady, or Children's Healthcare of Atlanta, that has an EHR system (such as Epic) that we could integrate our model into. We strongly believe that hospital systems would greatly benefit from incorporating Appoint AI, improving patient outcomes and bettering the medical realm.
Ultimately, the prioritization system that we developed can also be used in different industries, such as consumer care or inventory management. This indicates that Appoint AI has a lot of potential to improve various industries and make them more efficient.
Log in or sign up for Devpost to join the conversation.