Home

Studying Bipolar Disorder

Currently, diagnosis and treatment of bipolar disorder rely on careful history taking and mental status examination by an experienced clinician, at times aided by self-report or family-informed questionnaires. These reports, as well as in-person assessments, have to be interpreted by providers in order to extract patterns that could indicate an imminent change in mood. When individuals experience changes in mood, they also may struggle with navigating daily activities—especially ones with high cognitive demand.

With the expanding coverage of wireless Internet access and rapid advancement of mobile smartphone technologies, people are increasingly interacting via typed (rather than oral) communications. BiAffect’s innovative approach aims at understanding and examining the ubiquitous ‘virtual mental-health footprints’ or ‘signatures’ of abnormalities in people suffering from mood disorders, notably abnormalities in cognitive skills.

Using ResearchKit

ResearchKit allows access to iPhone’s proximity sensor, ambient light sensor, camera, accelerometer, gyroscope, compass, barometer, NFC, Touch ID, and pressure sensitive display, which we can use to collect context-sensitive metadata as individuals use the BiAffect keyboard. We will also employ ResearchKit modules like surveys and active tasks to supplement this metadata.

Deep Learning

The DeepMood architecture allows BiAffect to analyze keystroke dynamics data in order to infer the user's mood state using state of the art recurrent neural network (RNN) algorithms.

Participants and Insights

Our target audience is anyone with or without a diagnosis of bipolar disorder who is an English-speaking US resident aged 18 or older and interested in a) contributing to research, b) learning about how their mood interacts with cognition, and c) understanding how keyboard usage patterns and dynamics can be related to neuropsychological or cognitive functioning.

How are you thinking

How are you self-monitoring

Circadian rhythms

Go-No-Go reaction time varies between and within individuals

Example time-series of keypress dynamics

Pilot findings support that keyboard dynamics can identify persons with bipolar disorder and unobtrusively predict depression severity

From 2015 to 2017, a pilot study with 31 participants was completed that validated the hypothesis that keystroke dynamics such as typing speed, frequency of texting and use patterns in social media apps are altered during depressive and manic episodes in people with bipolar disorder. The full scale BiAffect study launched on iOS during HealthSpark at SXSW '18.