At the International Conference on Acoustics, Speech, & Signal Processing (ICASSP) 2024, Samsung R&D Institute Poland achieved first place in the 2nd e-Prevention Challenge, First Track. The competition is a prominent feature of the world’s largest technical conference on signal processing and its applications. This achievement marks a significant milestone for SRPOL in the realms of wearable technology and health.

The aim of the challenge was to contribute to the field of mental health relapse prediction by analyzing digital phenotypes of patients experiencing non-psychotic relapses. The competition focused on processing data collected from wearable devices such as accelerometers, gyroscopes, and heart rate monitors.

The data regarding non-psychotic relapses, provided by ICASSP 2024 Challenge organizers, contained mainly depression episodes. Considering the strong connection between depression and sleep, where most of the depressed patients exhibit insomnia symptoms, the team primarily explored the utility of the night-time data in the relapse detection solution. Besides that, sleeping patients naturally move less and that property was leveraged by the team to significantly reduce unwanted variations in the signal and device errors, hence gaining an edge over other contestants.

The winning solution employed a Transformer-based Autoencoder that plays a major role in transforming the input data into users’ digital phenotypes. The features extracted at the first stage were then clustered by the Nearest Neighbour Ensembles (iNNE) methodology. The second element is crucial for anomaly detection, which is the key to identifying potential relapses.

Altogether, the team at Samsung R&D Institute Poland – Warsaw – noticeably outperformed competitors at the 2nd e-Prevention Challenge at ICASSP 2024, securing the first place based on average (AVG) of the AUROC and AUPRC metrics:

Detailed explanation of the solution has been published in “Bumblebee Your Way to Recovery: Transforming the Approach to Detection of Mental Health Relapses” paper that has been accepted to IEEE ICASSP 2024 in Seoul, Korea.

The innovative approach in using wearable device data and signal processing techniques has significantly contributed to the prediction of mental health relapses, marking an advancement in health technology and patient care. The solution has the potential to further expand the capabilities of Samsung smartwatches.