Through shadowing physicians at the hospital, one of our members felt doctor-patient relations could be bettered if there was a way to make such interactions clearer. Doctors in the hospital often are rushed to see patients and through experience, our member saw that it could take a long time to reach the doctor in charge of their case when clarification or further explanation/questions arise. This was the problem that our members saw in the health industry and decided to try to tackle.
What it does
Med Assist is meant to take in a transcription from rev.ai, a speech-to-text AI, which takes in a .mp3 file and returns a nearly perfect transcript of the recording. From then, the transcription is processed into keywords such as diagnosis and treatment.
How I built it
We used Python and Scikit learn and pandas to implement our ML methods such as creating graphs/figures for data viz, and for creating the neural network which we trained with our transcription dataset. Our neural network is able to extract the diagnosis and treatment given any transcription with ~98% accuracy.
Challenges I ran into
We had a tough time figuring out how to use the machine learning methods, so it took a while to implement these methods into our program. In addition, there is a very limited dataset available currently on doctor-patient interactions and transcriptions of such.
Accomplishments that I'm proud of
We learned how to use ML in our program with minimal experience and had an accuracy of ~98% in our model!
What's next for MedAssist
We would want to develop MedAssist further after the hackathon by implementing into an app and working more on the front-end of our project and the U.I. We would also want to add the extra layer of translation to our project, which would be highly more convenient than the phone translation that is currently provided in most hospitals. We might also want to generalize/develop our translation algorithm to take in other simpler interactions in the hospital, such as general questions on wellbeing and requests from the patient to a nurse.
Next for MedAssist, we would work on incorporating our code into an app so that it can be more easily used in hospital settings.