Medication adherence is the extent to which a patient follows a stipulatory medication treatment plan . Over 50% of U.S. adults do not take their prescriptions as prescribed, which is estimated to be responsible for 33%–69% of hospital admissions and 125,000 deaths annually . We discovered that there was a suite of digital reminder-based solutions out there trying to tackle adherence at the home medication management level, but realized that onboarding friction stifled their utilization (too many steps to configure, too much data to enter manually, etc.).
That's where we come in! Say hello to MemorAIs.
What it does
MemorAIs is a platform that allows users to scan their on-the-bottle prescription intake directions to generate and download a .ics (iCalendar) file pre-configured with intake frequency, duration, and time information straight to their device.
.ics files are calendar files stored in a universal calendar format used by all major email and calendar programs, including Microsoft Outlook, Google Calendar, and Apple Calendar.
The platform works as follows:
- A user accesses our front-end interface and can choose to either upload or take a photo of their prescription label, including the intake directions from the pharmacy.
- The photo is processed by our system via -
** Optical character recognition (OCR) models, which extract the text from .png and .jpg files and returns substrings, taking both text probability and character coordinates into consideration. ** Regular-expression-based text classification that maps keywords to the specific frequency, duration, and time details needed to generate a calendar reminder.
- The identified details are used to generate and export a .ics file directly to the user on their current device. All they have to do is download it, and their medication reminders are integrated into a calendar of their choosing.
Just scan and forget! MemorAIs will do the remembering for you.
How we built it
We used Flask for the backend, with HTML, CSS, and JS for the front-end. A separate micro-service was created for the OCR identification, recognition, and parsing in Python.
Challenges we ran into
Dependency management: Scripts that ran well on one team member's machine did not on another's due to dependency management and versioning discrepancies. We even (surprisingly) came across this problem while using browser-based notebooks. Troubleshooting this took up a lot of our time and we wished it hadn't.
Time: Working within a 24 hour time constrain was difficult, but we managed to pull off a full implementation!
Team dynamics: We all came to this project with different levels of experience, both in the subject matter and technical expertise. We had to navigate hiccups along the way and learn how to support each other's needs in the moment. At the end of the day, we figured out how to work together in an effective way and came together to create something pretty cool.
Accomplishments that we're proud of
We're proud of everything: the premise & real-world value proposition; the way we handled the classification task (though none of us had really done something like that before), the .ics field mapping and extraction; the camera integration to the UI; etc. But mostly, we're proud of how we worked together and navigated challenges as they came.
What we learned
- Teamwork with people from varied backgrounds
- OCR testing (We got to find out how to identify what pre-trained modules work best on what images)
- .ics creation & field mapping
- UI development & camera integration
What's next for memorAIs
We're excited to hear feedback from our peers and mentors today to see if the idea has potential! The reactions and perspectives of others will largely shape the answer to this question.
In order to be functional, the current memorAIs implementation relies on making a call to a local server that is currently turned off! If the product gains traction, we will search for long term server solution so that anyone, anywhere can use it. Stay tuned!