The International Conference on Acoustics, Speech, & Signal Processing (ICASSP) is the world’s largest and most comprehensive technical conference on signal processing and its applications. The 2nd e-Prevention Challenge: Psychotic and Non-Psychotic Relapse Detection using Wearable-Based Digital Phenotyping is held at ICASSP 2024, which aims to stimulate innovative research on mental health relapse prediction and identification through the analysis and processing of patients’ digital phenotype in the psychotic spectrum.

Samsung R&D Institute China - Beijing (SRC-B) ranks third in the second track called Unsupervised Detection of Psychotic Relapses, which aims to detect the presence of psychotic symptom relapses using continuous measurements from accelerometers, gyroscopes, and heart rate monitors. The competition attracts many followers and participants from the industry and academia. Hence, the excellent achievement of SRC-B in the 2nd e-Prevention Challenge represents another step forward for SRC-B in the fields of wearable devices and health.

In previous research, it has been confirmed that the signals captured by wearable devices can effectively identify different users’ behavior patterns. Considering the similarity of the two tasks in distinguishing user behavior patterns, the user identification task is inferred to be helpful for the relapse detection task. Therefore, user identification is used as an auxiliary task to train the feature extraction network in the system to better explore behavioral patterns from the biosignals. Three different structures of the Transformer-based feature extraction network are adopted in our experiment. Then, the extracted features are used to train an Elliptic Envelope model of each patient for anomaly detection.

Members of SRC-B Language Intelligence Team