The current diagnosis paradigm for mental health conditions involves specialist assessment which is expensive and subjective in nature, resulting limited access of mental health care to people who in need but come from disadvantaged backgrounds or regions. Using digital and mobile technology, we collect multimodal data including video, audio and self-reported mood scores from clinical trials. By analysing different modalities such as facial expressions, speech, language and the time series of self-reports, we have developed tools and models that aim to assist the clinicians for understanding and screening their patients.
For example, we have developed a model that captures the order in which individuals change their mood states, and has demonstrated it’s possible to meaningfully position an individual on the bipolar spectrum. We have also studied the use of signature in learning and summarising signals from non-clinical interviews, and shown its potential in providing useful feedback for clinicians and patients.