From Digital Phenotype Identification To Detection Of Psychotic Relapses

• Published in: 2023 IEEE 11th International Conference on Healthcare Informatics (ICHI) pp. 276 - 284
• Main Authors: Efthymiou, Niki, Retsinas, George, Filntisis, Panagiotis P., Garoufis, Christos, Zlatintsi, Athanasia, Kalisperakis, Emmanouil, Garyfalli, Vasiliki, Karantinos, Thomas, Lazaridi, Marina, Smyrnis, Nikolaos, Maragos, Petros
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.06.2023
• Subjects: Artificial neural networks; biometrics; Biometrics (access control); Informatics; Medical services; person identification; Probability distribution; psychotic disorder; relapse detection; smartwatch; Statistical analysis
• ISSN: 2575-2634
• DOI: 10.1109/ICHI57859.2023.00045

Timely detection of relapses constitutes an important step towards improving the quality of life in patients with psychotic disorders. In this paper, we design a novel framework for discovering indications of psychotic relapses by modeling the digital phenotype of the patients who wear smartwatches. We start by designing deep neural network architectures that can use biosignals for person identification with high discriminatory performance. Then, we show how these networks can be employed to identify indications of psychotic relapses by looking at the per-person misclassification rate of the network and the corresponding changes in the output classification probability distribution, during different periods of the disorder (normal, pre-relapse, relapse). In order to prove the effectiveness of our approach for detecting relapses, we apply it to one of the largest datasets collected for biometrics in patients with psychotic disorders, with more than 18k days of collected data, and verify the output probability distribution change through extensive statistical analysis.