Patient Safety and Quality - The Armstrong Institute
A chronic cough is a difficult symptom to diagnose due to a large number of potential root causes. Moreover, a cough is episodic and sporadic. Symptoms may not present themselves on the day that the patient comes into the clinic, and the provider will send them away without a diagnosis, or will order imaging, the gold standard of diagnosis but which is also costly. My idea is an at-home patch or patches that captures cough sounds and mechanical characteristics and their locations over an extended period of time, and uses machine learning to make differential diagnoses based on the cough data. The goal would be to reduce health care costs for ordering unnecessary imaging, and improve patient diagnosis and outcome.
What it does
It passively monitors the patients acoustic noises to log cough events and related health data. The data is processed with a machine learning algorithm to diagnose any respiratory diseases. In addition, it provides a objective data of the condition of the patient and can be presented to the clinician to assess treatment efficacy and better diagnoses.
How we built it
We used python to develop the machine learning and signal processing algorithm required for extraction of meaningful data from acoustic waveforms. FluidUI was utilized to draw a wireframe of the mobile application that will serve as the front-end user interface.
Challenges we ran into
We originally started to implement an actual working demo on an Android platform. However, setting up of the backend libraries and interfacing with the media API was problematic. Therefore, we had to divert to a mockup for the submission to Medhacks deadline.
Accomplishments that we're proud of
We are proud of the more that 80% diagnostic accuracy achieved by the machine learning algorithm. A solid problem statement and use case derivations including an actual patient interview is another key accomplishment of CoughDX
What we learned
From a conversation with one of the mentors from John Hopkins Medical Hospital we also learned that our approach to patient safety and quality track could also be applied to telemedicine and contribute in increasing access to care. We also witnessed the efficacy of python libraries in implementing machine learning and signal processing algorithms.
What's next for CoughDX
Application of the algorithm to larger datasets, implementation of an actual working mobile application and supporting back-end are the future directions of CoughDX.