Having talked as a team about some of the real world challenges posed by recognition of Alzheimers-related cognitive decline, we realized that clinicians are forced to make decisions from very limited information about a patient’s baseline and their day to day progression. Frequent clinical check ins may not be pragmatically feasible, especially in the early phases of the disease, when it may be particularly critical to keep a close eye on any changes. Understanding that complex technology can be an even higher barrier than usual for those older in life, and especially those suffering from dementia we wanted to design a tool that would be easy to use, yet still reliable for those receiving the information.
What it does
This tool provides a low friction method of monitoring the mental status of anyone struggling with or concerned about Alzheimer’s related cognitive decline. By sending questions over plain SMS, parsing responses through state of the art Natural Language Processing, and sending only the fields of interest to our backend, this tool enables reliable quick and unobtrusively collection of data that could help a loved one or doctor get a sense of a patient’s cognitive baseline or progression of decline. Given that SMS is an unencrypted communication system this tool would not meet HIPAA standards for regulation of patient doctor interactions so we envision this rather, as a way to empower family members and loved ones to monitor the status of people in their lives susceptible to cognitive decline. Those loved ones would also be able to bring summaries and graphs from the data to their doctor’s attention at a joint appointment or over a secure line, giving the tool some indirect relevance to clinical decision making
How we built it
This tool was built using twilio’s auto pilot service, an automated text messaging bot that uses Natural Language Processing to parse out answers from a User’s response. The service allows you to run your own custom models and tune the results to your specific scenario. For example in this case, if we asked a patient to give a response regarding date and time, we handed off several samples to our twilio autopilot to train with in order to parse out an ISO timestamp from a User’s response who only mentioned date using english words such as “Friday” or “Yesterday” with a high confidence.
Once we had our twilio service up and running, we attached a webhook to our web server. The server ran a Nodejs application that listened for incoming POST requests that contained detailed information about a user’s results when they completed their check-ins. Those results were then parsed and stored into a MySQL database where the records could later be accessed and displayed to a user via a Web Application or Mobile Application.
Challenges we ran into
The challenges we ran into came down to technical and organizational challenges. We had a completely different vision of the project at the start of the hackathon, and realised mid way through that our goal was unrealistic and we were struggling to get anything working with the tech stack we chose. Organizing the team and getting everyone on the same page with the goal and objectives of the project ate up a lot of our time that could have been spent hacking together a proof of concept. We also had the issue of putting our team together blind, and having to learn the workflow that everyone on the team was comfortable with. Technical challenges came in when we introduced a framework that most of us were unfamiliar with, this led us to spend more time in documentation rather than crafting our implementation. A lot of us came from different backgrounds at different skill levels, so understanding where each one of us best fit took some time as well.
Accomplishments that we're proud of
We quickly spun up a basic website with a login and registration page and then pivoted into a previously unfamiliar technology of SMS interface. We rapidly gained familiarity with the tools and some complex topics in web hosting and computer networking.
What we learned
We started with a basic web-stack, with django, html templates, and bootstrap. Many of us were relatively new to this technology and learned a lot in the process. In trying to make the technology as accessible as possible we pivoted to an SMS based interface. To set this up we learned all about http tunnelling, REST-apis, and chatbot technology. Due to the opinionated nature of some of the django defaults we had to switch to node late in the process giving us the opportunity to learn about a different and backend framework that many of us did not have previous experience in.
What's next for MindWatch
We would like to build out a front end for convenient access to the data, expand the database features to include more information to better manage multiple patients and multiple care providers. Our SMS platform also provides the option to extend the tool to phone calls, alexa, and google assistant apps, so we would like to try extending it to those as well.