What it does
69% of individuals with bipolar disorder are initially misdiagnosed and 33% of these individuals continue to remain misdiagnosed for over 10 years. When less than a third of a severely at-risk population is getting the correct treatment that they need - which is dependent on correct diagnosis - there is an urgent need in the healthcare industry that must be addressed. Bipolar diagnoses are most commonly conflated with unipolar depression diagnoses, because patients either are not aware of manic/hypomanic episodes themselves or because clinicians are rarely able to observe manic episodes in diagnostic interviews. Giving clinicians more data on their patients' emotional and mood trajectories on a daily, continuous basis outside of weekly clinical appointment settings will greatly inform their ability to make diagnoses and create treatment plans for patients who may be bipolar.
How we built it
We used the IBM Watson APIs Speech-to-Text and Tone Analyzer in junction with each other. Using Python, we recorded .wav files of recordings for therapists. We first feed the .wav file through the Speech-to-Text API, filter through to search for trigger words found from Affective Word Database from the Stanford Tsai Lab, and then use the Tone Analyzer API to rank the emotional sentiment of the input sentences. Simultaneously we analyze the peak pitch and loudness via Fourier analysis of the sound data. All of this data is analyzed in Python Flask and uploaded into tables for the therapist to view.
Challenges we ran into
Identifying suicide triggers without setting off false alarms (from common idioms/slang) Use of multiple new programming languages